CLAUDE AI MASTERCLASS for Professionals + Claude AI No Code

1. Introduction: Thank you, and congratulations on taking this class. Claude AI Master class, how to use Claude AI to work smarter, write better, and do more without writing a single line of code. If you've ever opened Claude, type something in and thought, This is good. But I know there's more to this than I'm getting out of it. You're in exactly the right place because Claude isn't just an AI assistant. It's one of the most powerful thinking and writing tools available today. You know how to use it properly. In this course, we're going to take you step by step through how to use the latest version of Claude effectively and how to apply it confidently to real work, real projects, and real decisions. Without writing a single line of code, let's start with a real life scenario. Imagine you're running a business or managing a team or working in any busy professional role. You're constantly switching between tasks, drafting documents, analyzing data, responding to messages, researching topics, creating content, checking facts, and trying to keep everything consistent and on brand. You can use Claude for all of that, but most people don't. Because they open it like a search engine, type a quick question, and move on. Instead of using it the way it was designed as a structured, intelligent workspace that gets better, the more you put into it. In this master class, you'll learn exactly how to use Claude, the way it was designed to be used, starting with prompting. There's a simple structure that transforms the quality of everything Claude gives you. And once you see it, you'll never go back to typing random questions again. Then you'll learn how to set Claude's behavior once using Custom Instructions. So it already knows your voice, your rules, and your context before you type a single word. You'll learn how to turn rough notes, uploaded files, and messy data exports into polished publish ready documents, how to analyze data and spot patterns without being a data analyst and generate charts you can use immediately, how to verify and critique Claude's own outputs before you publish. You never put your reputation at risk, how to think through complex problems and strategic decisions, using Claude as a reasoning partner, how to work with entire document libraries, large datasets, and long files, all in a single conversation, how to build interactive outputs, widgets, and components using artifacts, without touching code, how to organize all your ongoing work inside project. So Claude always has the context it needs, how to create reusable skills that apply your standards automatically every single time, and how to connect Claude to the tools you already use, Google Drive, Gmail, Slack, and dozens more, and use Claude directly inside your browser. With all of this, you'll have to turn Claude into a reliable assistant for writing, research, analysis, planning, strategy, and automation without writing a single line of code. Live practical demos using the latest Claude model throughout. So you always know where to click, what to type, and how to apply it immediately in your own work. By the end of this course, you won't just be using Claude. You'll understand how to think with it, structure work inside it, and use it like a professional who's three steps ahead. So take your time, follow along with the demos, experiment as you go, and by the end of this course, you'll be using Claude smarter, faster and more effectively than anyone else in the room. So let's get started. 2. The Birth Of AI: The 1990s and 2000 brought a big shift, the rise of machine learning. Instead of being told exactly what to do, computers started to learn from examples and data. This approach quietly powered many products we all use today. Spam filters learned which emails were junk. Google search got smarter at finding the right pages. Amazon began recommending products you might like while Google Maps learned to predict the fastest routes based on live traffic. Came alexa and other voice assistants, which could recognize speech and answer questions out loud, something that felt almost magical at the time. Computers were now learning from experience, not just following rules, but even then AI couldn't really create things. I could predict and categorize, but not write, imagine, or explain. That changed in 2017 when researchers at Google developed a new system called the transformer. It helped computers understand how words relate to each other in a sentence, not just one word at a time, but in full context. This was a huge breakthrough and laid the foundation for the next big step in AI, large language models. A large language model, LLM is an AI trained to understand and generate human language. It learns from massive amounts of text, books, articles, and online content by spotting patterns in how words and ideas connect. After the transformer architecture was introduced in 2017, models could finally understand context and meaning. This led to powerful systems like GPT, capable of writing, summarizing, reasoning and chatting in natural human like ways. That next step came from Open AI, the company behind ChatGPT. They built on Google's work and created something called the GPT series, generative pre trained transformers. The first version GPT one showed that a computer could learn to write readable text by studying huge amounts of online data. Then came GPT two in 2019, which could write essays, stories, and even news articles that sounded human. A few years later, GPT three made an even bigger leap. With 175 billion parameters or neurons, it could write, translate, answer questions, and even code. There was just one thing missing, natural conversation. GPT three could give answers, but it couldn't chat as fluently. So Open AI improved it using feedback from real people, teaching it how to respond more naturally and politely. The result was ChatGPT, a version that could hold a conversation. Remember what you said and reply in a way that felt personal. That was when AI truly became something everyone could use, and this is where prompting came in. The word prompt originally came from early computers. It was the line on the screen where you typed a command. The computer was waiting for your input. Over time, the meaning changed. Now, a prompt means the message or question. You give an AI a system like ChatGPT to tell it what you want. At first, prompts were very simple. Things like write an email, summarize this text or explain this to me like I'm five. But people soon notice something interesting. The way you wrote your prompt completely changed the answer. A detailed prompt gave a detailed result, a clear question, got a clearer answer. The better your prompt, the better the AIs performance. This turned the act of prompting into both an art and a science. Today we call this art and science prompt engineer. 3. Claude AI - What is An LLM?: What is an LLM, including examples like ChatGPT, Gemini, Claude, copilot and perplexity. Now that we've talked about where modern AI tools came from, the next step is understanding what's actually behind them. When people talk about AI today, what they're usually referring to are large language models or LLMs. Tools like ChatGPT, Gemini, clawed, copilot and perplexity are often called AI, but under the hood, they are all powered by LLMs. And LLM is a type of AI built specifically to work with language. That's why these tools can write emails, answer questions, summarize documents, explain ideas, and help with planning and analysis, all using everyday language. Over the last few years, LLMs have been everywhere. They've appeared across global news, business, education, health care, and government. They've been used to translate languages in real time, analyze large amounts of information, support research, and help people make sense of complex topics. Even if you don't work in tech, it's been hard not to hear about them. But what exactly is a large language model, a simple way to think about an LLM, very basic level, LLM is trained by reading huge amounts of text, books, articles, websites, and other written material, and learning how language works. It learns how words are usually used together, how sentences are structured, and how ideas flow from one to the next. It doesn't understand the world like a human does. Instead, it learns patterns in language. When you ask it a question or give it a task, it predicts a useful response based on what it has learned from text. That's why an LLM can feel intelligent and conversational, even though it's really working with patterns and probabilities. Why LLMs are so powerful, size matters. As mentioned earlier in the course, one of the key things that makes large language models different from older systems is their size. We talk about size here. We're not talking about physical size. We're talking about the number of parameters inside the model. Think of it like teaching a robot how to play a very complicated game. To do that, you'd need to give it lots of instructions and rules. Each rule helps the robot decide what to do in different situations. In an LLM, parameters are like those rules. The tiny pieces of information, building blocks that help the model decide how language should work. Parameter on its own doesn't do much, but when you have millions, billions or even trillions of them working together, the model becomes very good at handling language. Modern LLMs are far larger than earlier models, which is one of the reasons they can handle more complex tasks and produce better results. LLMs don't just rely on size. They also learn by reading massive maps of text data. You can think of it as giving a computer access to an enormous library and asking it to read everything incredibly fast. This training data includes books on many topics, news articles, and blogs, Wikipedia articles, online discussions and forums, recipes, reviews, and guides, scientific and technical documents. By reading all this text, LLMs learn vocabulary, grammar, tone, and how language is used in different situations. It's similar to how humans learn language by reading and listening, just on a much larger scale. ELDA AI systems were usually built to do one specific task. If you wanted a new task, you had to build a new system. LLMs are different. The same model can write an email, summarize a report, answer questions, explain a concept, help plan work. This is why LLMs are often described as general purpose language tools. After being trained on large amounts of texts, these models are refined, so they're more helpful, safer and better at following instructions. That's why modern tools can hold conversations and adapt to different types of tasks. Why different LLM tools exist. Though many tools are powered by LLMs, they aren't all designed for the same purpose. Some handle long documents better. So focus on research. Others are built directly into workplace tools like email, documents and spreadsheets. There isn't one best LLM for everything. Choosing the right one depends on what you're trying to do, and that's something we'll cover later in this course. Most important thing to remember LLMs are language tools. They are extremely good at working with text, writing, summarizing, explaining, comparing, and organizing information. Understanding how they work at a high level helps you use them confidently, safely and effectively at work. In summary, an LLM is an AI system designed to understand and generate human language. Most tools people call AI today. Powered by large language models, LLMs. Learn language by reading massive amounts of text and spotting patterns. They're trained on books, websites, art schools, and conversations from across the Internet. LLMs are general purpose, not built for just one task. They can write, summarize, explain, compare, and organize information using natural language. Real value comes from knowing what to trust, what to check, and when to apply human judgment. 4. Which Claude Plan Should I Use?: So before you start using Claude for real work, you need to know which plan you're on and what that means in practice. So this lesson gives you a clear, honest picture, so you'll know what to expect as you work through this course. Now, the main thing I want you to remember is that plans change. And Tropic updates plans, pricing, and limits regularly. So always check claude.com slash PRCI or go to claude.com and find the pricing page for the latest. So the information I'm going to give you is correct as of today, and it can change at any time. So let's start by going over the plans. What you're going to get when you first go to Claude is the free plan. And as you can see here, the free plan, $0 a month. The key features are, I mean, you can do a lot of things with it, but you'll use the free plan if you want to use Qarchat, file uploads, web search, memory. You can still crect projects, which we'll go over in Artifacts. And it's best for exploring Claude and occasional use. So I'll go over these things in more detail. I'm just giving you an overview. If you go for the P plan, which is 17 to $20 a month at this time, you get five times more usage than the free plan, and I'll go over that and you get access to Claude code, extended thinking, so it will be able to think more, and you get priority access because remember, everyone's competing for computing processor time when you're using AI models, and it's adequate for daily professional use, if you'll often be doing things like updating documents or creating documents every single day, so the free plan would not be sufficient at this time for that. Then we've got the MAX plan, which is currently 80 to $160 a month. And that's five to 20 times more usage than the P plan, and that's for heavy continuous use for power users and all day workflows, so you'll basically be using it all day and generating content using workflows or using workflows to do things for you. Again, I'll go over that. Then we've got the team plan. The team plan is currently 25 to $30 per seat. So in other words, per user. And then shared projects. So if you want to be able to share projects, have centralized billing collaboration features, especially if you've got teams and you need to collaborate with teams or teams need to collaborate with each other with a minimum of five seats. And this is suitable for teams working together. Then you've got the enterprise plan, which the price will have to be discussed with anthropic before they give you access and they'll give you something custom to you. And this is if you need custom context windows, single sign on, audit logs, compliance controls, and it's mainly for larger organizations. So I won't be covering that kind of stuff in this particular training, but this gives you a good overview of what you'll need, and it will make more sense as we go through the training. So let's talk about reset cycles and what I mean by that. So the reset cycle is 5 hours, and that's when your session resets, but also what happens is you need to think about the amount of process that you're using. So the amount of compute, they call it that you're using. And it's not very easy to know exactly how much of the compute you're going to be using until you start doing some work because everybody does different work. Main thing is that Claude doesn't give you a fixed number of messages per day that you can type into chat. The usage is session based, and so it depends on the amount that you type in in terms of tokens, which is basically what you do type in in terms of messages, divide it up. And when you divide it up into various words and Claude in the background will turn that into numbers, and it will work out how much you've used. And then when you've used your quota, it will say, Okay, you've used the maximum amount for that day. So that's why I can't tell you upfront how much it is. But the main thing to note is that usage session is based around a five hour window, so it will reset every 5 hours. How quickly will you burn through it? Well, that depends on how long your messages and files are. It depends on which model you're using, which we'll go through soon. It depends on which features you're using. So are you using file analysis? Are you just sending messages? Are you doing something like extended thinking, web search, certain things you use more than others. And it also depends on how long the current conversation has grown. So when you do a lot of work in the same conversation, which is quite helpful, obviously, after a certain amount of time, it will get too large, and then it will say, you've used the maximum amount of compute, but that will reset every 5 hours currently. So let's talk about the free plan. In plain terms, the free plan is good for light use, short chat messages, occasional file uploads and analysis. And for example, this is just a very loose example from my experience. It might give you ten to 20 exchanges of uploads and chats in a session. Heavy use for long documents, data analysis, back and forth editing, the kind of thing most people do at work might give you five to ten uses. And then when you hit your limit, Claude will tell you at that point, and it will ask you to wait for the session to reset, which is typically a certain number of hours, maybe 5 hours. It depends what stage you are in that five hour window. I will tell you. The difference if you're using, say, the P plan, which is what I am currently using in my work, is you get five times more usage, as I said before than the free, and the pro plan gives you that for free. And for most four work days of professional use, that's enough because you'll probably be doing other things in your day and going back and forth to it, but it totally depends how you use it. So can't be 100% accurate. And if you're hitting your limits regularly, then you upgrade simple as that. So you just kind of work out what your daily pattern is, how often it's timing. And then you upgrade to the next plan, if that makes sense for you. The next plan currently is MAX, as I showed you before. And the main thing to know is that usage limits tighten during peak hours. And so roughly weekday afternoons in your time zone would be peak hours. And you'll get through your session faster. You'll get through your session faster at these times adjusted the peak hour behavior before, so they might do that again. And when I say you'll get through your session faster, what I mean is you'll get to the point where you've used all the compute, I can give you all the processing it can give you at that time, faster because everybody's fighting for compute at the same time. So then the next question is, for this training, what can you do with the free plan and what are you going to need to upgrade to another plan for? So, currently, at this time, the majority of this course works on the free plan, and that's things like prompt structure, Context Continuity, memory, all that kind of stuff. I won't go through it all now because I'll be going through it in a short while. But essentially, most of the course you can do on the free plan. However, some things will need an upgrade. And that would be things like, for example, Claude code, skills, agents Automode agent teams, Connectors, computer use, browser use, and that uses a Chrome Extension, which I'll be through. So if you're on the free plan, simply follow along with these lessons to understand how everything works and then upgrade when you're ready. Again, these things can change, so this is the current picture, but these things can change, but the main thing to take away is that you'll get to a point where you try and do something, and it may ask you to upgrade at that point, and you can decide if you just want to follow along with the training or if you want to upgrade to do those things. Here we are on the pricing, and you can see there are various plans, and these are the different plans that you're going to need to choose from, and you've got individual, which most people are going to use. You've got team and Enterprise plans, and you've got API plans if you want to write code that's going to access Claude APIs. So these are the different plans at this time you can use, and, of course, they change. Now, when you first access claude.ai, you'll either be logged in or you won't. I happen to be logged in. But once you are logged in, you can see what kind of plan you're on quite simply. So the first thing you've got to do is you've got to go down here and click. You're going to have an initial icon, that's called the initial icon, and you can click on that. Once you're logged in and then go up to settings, and you want to click on Billing. And usually here, it's going to tell you which plan you're on. You can see here I'm on the Pro plan. And the other place to check is here at account, and it usually says, apart from logging off, and apart from logging off from all your devices, it also says to delete your account, please cancel your, and then it tells you which subscription you're on Claude Pro subscription in this case. So that's exactly how you can find out which of these plans you're currently on. So in summary the free plan is a genuine starting point. So you can start at the free plan. It's not just a trial. You get access to a whole load of features, and you can do good work with it. However, remember that your usage, based on the amount of processing that you're using, resets every 5 hours. And so if you run out of processing, then 5 hours later, you should be able to go back and do some more work, even if you do run out. And heavy task, the more processing based on what you're doing, burns through the processing and the compute faster. Remember that different plans such as the P, give you more usage. So the P plan gives you five times more usage than the free, and it unlocks certain features. At this time, there's things like clawed code and skills. Remember also that plans and limits do change. So check clawed.com slash PRICE or go to claw.com and look for the pricing for the latest. So now it's your turn. What I want you to do is to check which plan you're on right now, and you'll be able to do that using profile icon settings and then billing. And if you're on the free plan, note which lessons in this course will need an upgrade and decide whether to upgrade now or follow along. 5. Claude AI - Side Menu Overview: So when you land on Claude's page, you will be faced with this screen where you'll be able to do many different things, and you'll be able to do them all through prompts in here. But what I'll start by doing is just showing you the side menu. So the side menu starts off collapsed like this where you can access all these different items all the way to the bottom, including this, which is what we call the initial menu. And what you want to do is start off by opening it up so you can see what all these different things do. So you can create a new chat simply by pressing this button, and if you're already on an existing chat, it will pop up a new one here so you can separate all your different chats. These are all the recent ones. This allows you to search. You click on that and you can search through all the different chats. So here are some of my previous chats. For example, if I type in football, it will bring up all the different chats with the word football in them. Is that? Customize allows you currently to look at organizing your skills, which is something we'll go through later. But these are basically reusable bundles of prompts, and you can reuse them to do particular things, cord skills and Connectors which allow you to connect to different apps. Chats, again, allows you to access all your different chats all in one place. Projects are basically collections of chats that are all related, and they also store context. So we'll go over that further, but essentially everything within a project has the same context, and it allows you to save files with it. But essentially, it's a group of all your different chats and some context about them. And then we have artifacts. Artifacts are basically different types of objects you can build, and they basically help you to achieve different goals. So they could be anything from if we were to go to new Artifacts here. It could be anything from apps and websites to documents and templates, games, productivity, creative projects, quizzes, and surveys, something that you want to build that's completely custom, which is none of these things, or it might be just a fun way to create charts. But essentially, an artifact is anything you can think of you could create that you could reuse for different projects. And then there's code. So you are a developer, then you'll want to download clawed Code app, but you would download and install clawed code, and then you'd be able to access it from the side menu in Claude. So that's an overview of your side menu at the top. At the bottom, clicking here on this arrow allows you to download Claude desktop, so that allows you to interact with your computer. And it's giving you an overview here of clawed code, clawed for mobile, Claude for Chrome, which we will be going over. And essentially other apps that work in conjunction with Claude you can access here and download from. And then down at the bottom, this is where you access your settings. So all the settings for Claude are in here, some of which will be going over in this training. You can change your language here, get help, upgrade your plan, get apps and extensions. You can give Claude someone, which is cool and learn more about Claude here. And then finally, you can log out at the bottom. So basically, all of your settings are accessible here. So that's an overview of your site menu. And now let's get into actually using Claude and how we're going to use it to help solve a lot of your project problems with AI fast. 6. Which Claude Model Should I Use?: So in this lesson, we're going to go over which model, which Claude model you should be using. We're going to be using the model selector. Now, the model selector in Claude lets you choose which version of Claude you're talking to, and different models have different strengths, speeds, and costs against your usage limit. So this lesson tells you exactly which to use and when. The main thing to remember is, as with different plans and pricings, different models that are going to be used will change because Anthropic releases new models regularly. So the models shown here that I'm about to go through are correct as of now, as of this recording, and they match what you will see in the model selector for this training and just bear that in mind that they can change over time. So let's go over the models that are actually available to you. So in the table, what you can see is the models, the list of models, the speeds, and what they're best at. And the general rule of thumb wid models is it's usually that you're thinking about speed versus complexity. So some models are quicker, but deal with less complex actions as well. They don't deal with complex actions as well. Some are slower, but they deal with more complex actions better. So let's go through the list and all will become clear. So the first model you can see is Sonnet 4.6. So Sonnet 4.6 is fast. It's best at everyday tasks such as writing, editing, analysis, file uploads, data analysis. And it's the best balance of quality and speed, and that's why I'm using it for many tasks in this training. Opus 4.6 is slower, and it's best at complex reasoning, long form strategy, difficult coding problems, tasks where accuracy matters more than speed. Haiku 4.5 is actually the fastest model, and it's best a quick questions, simple tasks, high volume repetitive work where the speed is more important than the depth. So there's that thing of balancing speed versus complexity or in this case, depth. Opus 4.5 is slower, and it's best at previous Opus generation. So what Opus used to do, and it's strong on coding and workplace tasks. So you can use Opus 4.6 unless you have a specific reason. Opus three is slow and older and it's best at older models. It's best at doing what the older models do. And is included for continuity from the older models. And there's no reason to choose this for any new work. And Sonnet 4.5 is fast, and it's best at doing what the previous Sonnet generation did. It's solid, but Sonic 4.6 is better. So there's no reason to this for new work. But I'm going through these because these are likely what you'll see in your list of models, which we're about to go through. And I'm going to show you how to choose the models, so no worries there. So here we are back in Claude AI, and we're about to choose the correct model that we want to use. So as you can see, down here on the right, you can see the models. Now, the models that we choose are going to affect the performance. They're going to affect the complexity of what we can do. We've currently chosen Sonnet 4.6, which is good for what we're about to do. But if you click on this, you'll see that there are a number of models available. So it says Opus 4.6, most capable for ambitious work. Sonnet 4.6, what we're currently on, most efficient for everyday tasks. Ku 4.5 fastest for quick answers, and then you've got extended thinking. And you can turn on extended thinking when you want the model to think longer for complex task. And you can do that simply by flicking this switch, then you'll see it says extended up here. And if you come back and click it off again, you'll see it disappears, extended disappears. And you can actually find more models here. So this is where you can find Opus 4.5. It says Opus consumes usage limits faster than other models, and Opus three and then Sonnet 4.5. So this is where you go to change model. So let's say you were doing something where you needed some thinking. So you were a scientist and you wanted some deep thought about the molecular structure or something. Let's just say something like. Would be able to switch this on and get some extended thinking, something that would take maybe some time to work out. But if you're doing something that wants a good balance, I would stick with 4.6. So my advice, what I recommend is stick with 4.6 for now and then experiment with these others based on what you want to do and make the correct decision. And one thing to remember is your model choice exists for the current chat you're on. So if you see here, this is 4.6. Let's say we switch it to OPAs 4.6. Remember that. And then we go over to this side. We can actually create a new chat. So what we can do is if we click on chats, it opens up all our chats. And if I click here, a new chat, that opens a new one. It currently says 4.6. So if I now switch back to Sonic 4.6 and type in something. So I've typed in something really simple here just for the sake of creating a new chat, and I've asked capital of England, and what we get back is nice easy answer. But that gives us a new chat. If we go over to this side and it'll say Capital of England, that's where we ask that question. If we go now and create a new chat, again, and then in here, we change this to Haiku 4.5 and ask capital of USA. So there you go. We've asked what Capital of USA is and we run that. So now we've got our answer. The purpose of that was if we go back now and look at our two chats, we'll see in here, it's using Sonic 4.6, and in here, it's still using Haiku 4.5. So that's just to show you that you can set the model individually for the chat that you're using. Let's talk about which model this course uses. For this course, as I've alluded to earlier, we're using Sonic 4.6 for all lessons unless told otherwise. And that's the default model available on all plans at this time. And it handles every task in this course well. And the only exception at the moment is the reasoning and planning lesson, in which I'm going to suggest switch to Opus 4.6 when you need maximum on a genuinely complex strategic. Opus 4.6 uses more of your usage allowance. As I mentioned earlier, there's a usage allowance, so it's intentionally not on by default. So just pay attention for when we switch to other models. So Sonic 4.6 is actually great for everything that we need to do, and you won't need to use another model for most things on this course, but you're welcome to if you'd like to. So in summary, Sonic 4.6 is the default. Use it for all lessons in this course, unless I say otherwise or unless you want to try something. Opus 4.6 is usually for complex reasoning tasks where accuracy matters more than speed. Haiku 4.5 is for fast, simple tasks, and you won't need it during this course. Welcome to use it. Older models, Opus three, Sone 4.5, Opus 4.5, are there for continuity. No need to use them unless you want to. And you can switch models via the model name bottom, the model name button at the bottom of any chat, and that's unique to the chat that you're in, so it changes based on the chat that you're in. So you're welcome to play around with the different models and put in more complex or less complex prompts based on what each model does. Experiment with that, and I'll see you in the next lesson. 7. How To Get Clear Answers Using Prompt Structure: One. In this lesson, we're going to learn the single most important skill for getting useful results from Claude, and that's prompt structure. You see, most people open Claude and type a quick question, and that's why they get vague answers, mix formats or responses that don't match what they actually need. And prompt structure fixes all of that. And so we'll do it using one real scenario, carry that scenario throughout this entire course. And our running example is that we are building a football or soccer news and stats website, and we're going to be covering the Premier League, Champions League, and the International Football League. And your audience is an audience of data hungry fans who want fast, accurate and opinionated coverage. So the feature we're going to be using, if you can call it a feature, is actually prompting, and it's the structure of prompting. That's what we're going to be looking at. Why we use it, we use it to control Claude's inputs using a special format, which is role, goal, contexts, constraints, and output. And I'll explain that. But it's the way that we structure our prompt in a way that we usually get back exactly what we want using this. I call it a prompting pattern. So we're going to give a particular pattern, a particular structure to our prompts. And what it solves is getting clear decision ready outputs every time. So let's go now into Claude and see that in action. So now we're back in Claude. Let's create ourselves a new hat. And we can do that by going up here and clicking Newhat. And what we're about to do is we're going to write a piece. We're going to write a match preview for an upcoming football match. There are many ways to tell Claude what you want to do. But let's say we type in something simple, so I'm going to paste in this prompt. And this says, write something about the Premier League title Race. I'm going to change my model back to Sonnet because we were just experimenting with different models. This is something simple, so we're going to keep it on Sonnet for now, and let's run that. After running that, as you can see, what we've got back is it said the title race Arsenal's moment. Question mark, nine points. That's the gap between Arsenal at the top of the Premier League and Manchester City in second. And with the season entering its final stretch, cushion. It's a cushion that looks increasingly like a coronation is waiting. So this is the language it's chosen. Arsenal sit on top. Arsenal sit on 70 points from 31 games, a record that speaks of a side that has groundout results as much as they've dazzled. This is the language you can see the way speaking. Not too bad. And at the bottom, it says, The gunners have seven games remaining to close out the job. So it recognizes that in the UK and around the world, we call Arsenal the Gunners. So not too bad for Claude there. Now, let's type in our prompt and see what a difference that makes to the output. So now we've put in our prompt. We're telling Claude, you're a football journalist. The goal is to write a match preview for this weekend's Premier League title Race clash. Context is the teams, our Man City and Arsenal audience. That success fans 18 to 35. The tone is confident and punchy. Constraints, we want a max of 200 words, no fluff, one keysp per team, and the output format is headline preview, three paragraphs, and the key Stack call out in two bullets. So remember what we've got here, and now we're going to run this. Let's go for it. So now that we've run that, interestingly, Claude has been smart enough to see, I can see this weekend schedule doesn't actually feature Man City versus Arsenal directly. So that's pretty smart that Claude has noticed that. But nonetheless, it's given us our piece. Now, if you notice the difference, the first thing is, if we go up to the top, you'll see that this is a smaller title and a number of paragraphs. Then if we go down to the bottom, you'll see it's got a larger title. It says, Arsenal have a nine point cushion and seven games left. This weekend, they don't face each other. They face the pressure of what happens when both win or both don't it's got a piece here. And then down the bottom, it's got the Key Stack call out, the Mogi graphic here, and the two bullets. It's got Arsenal, only three league defeats all season. It's got the wins, the draws the losses, the most resilient record in the division, and it's got Man City in bold, five losses in 31 games, et cetera. So just by making this tweak and saying what it is we're doing, giving it some context about what the teams are, giving it some constraints of Max, 200 words. And then, importantly, put format, headline, three paragraphs. We've been very specific. So we're here, we've got a number of different paragraphs, and it's actually picked up the correct teams because of some prompting I did earlier. What you can see is the format is different. And so what this is showing you that is that the way that we prompt makes a huge difference to what we get back. We can change the tone, we can make it sharper. We can give it more banter, as is football, and we like a little bit of banter. So we can do whatever we want with this simply by changing our prompt. So let's refine our output slightly by pasting in this prompt. So we've said, Al, give me two alternative headlines, one clickbait, one analytical. Then recommend the best one for a stats focused audience and say why in two bullets. So let's run that. So now we get back our two alternative headlines. So we've got the clickbait headline and the analytical headline for two different types of audiences. This is what Claude has given us. So this says Arsenal are bottling it. City's Last Chance starts this weekend. So if you're not English, bottling, it just means that they're scared. Analytical nine points, seven games, Arsenal's pressure tolerance will define this title race. That's a bit more of an analytical one, less Banto going on here. And then we've got some recommendations from Claude, it says recommended for a stats focus audience to analytical headline. Here's why. It leads with the numbers, 9.7 games, immediately signals data driven framing. And then this one with the X pressure tolerance does double duty. I borrows from the familiar X style prefix to imply a measurable concept, flattering to the readers analytical literacy. So it's given us the reasons why, but the main thing is you can use Claude to change the tone of something. And also, if you're doing something that does require headlines, it's going to give you some options simply by prompting it to do. Not only can you just ask Claude to do something, you can ask it to give you options, and it's intelligent enough to know about the subject matter, something like sports, which if you're not from the UK and you still want to get the language correct, let's say you're building an app or a website, such as we are, you want to be able to tap into what the audience are going to understand or relate to better, and that's what it means by adding this slang here or these colloquialisms. So that's that. So in summary, what we've seen in this lesson is vague prompts produce vague answers. When we gave it a online, we just got some paragraphs back. Prompt Structure turns Claude into a reliable work assistant. If you know how to structure it correctly, you can get back pretty much exactly what you wanted in the format you want it. And the core structure that I use is the role, the goal, the context, the constraints, and then the output format that you want it back in. And we used it to generate a match preview for our football website, but you can use it to generate anything you want for any kind of task. So now it's your turn. What I want you to do is pick something you need to write for your football site or for your website, or for any app or any project that you're working on this week. And then what I want you to do is write a structured prompt using URA role, goal, followed by the exact outcome you want, the context, which is only what matters in relation to the environment or in relation to whatever it is you're doing, the constraints which will tell Claude the length of what it wants back, the tone, and the rules around what you want back, and then the output format. Is it numbered sections? Is it bullets? Is it a table? So give it the structure, and I'm sure you'll be happier with what you get back. So there you go. Have fun doing that, and I'll see you in the t 8. How To Set Behaviour Using Custom Instructions: So in this lesson, we're going to learn about how to set behavior using Custom Instructions. We're solving a different problem, which is, how do I stop setting up Claude from scratch every single conversation in every single chat? And Custom Instructions let you set persistent default behavior for Claude. So it already knows your site or your app or your projects voice, your role, and your rules before you even type a single word. So let's get into it. The feature we're using is called Custom Instructions. Why we use it is to set persistent default behavior that Claude applies to every conversation. What it solves is inconsistent outputs, repeated setup, off brand responses. And Custom Instructions aren't prompts, persistent references that shape every response. So keep them about the voice and the context, not about specific tasks. So the way to set the Custom Instructions is to calm down here to your initial menu, click on that and then go to settings. And then what you want to look for is this area here where it says, What personal preferences should Claude consider in responses? Your preferences will apply to all conversations within anthropic guidelines. So you can have a read of that if you'd like to. But essentially, what we're going to do now is we're going to paste inside of here our Custom Instructions that I created earlier. So here are now I've pasted them in. Let's have a read. So it says, You are an editorial assistant for a football, soccer news and Stats website. Our editorial voice, confident and direct, no hedging, data led, ground opinions in stats, not on opinion alone. Punchy, short sentences, no padding, knowledgeable, assumes the reader knows the offside rule, so that's a football rule there. Tone rules, never use cliches, for example, at the end of the day. No emojis in article content. Use European date format, DDMMyy y. And content and context. The focus is the Premier League, the Champions League, and international football. The USB unique selling point is combining narrative journalism with data analysis. And the audience, stat success fans age 18 to 35. So now, because of what we've done, Claude is going to frame its responses around everything we've put in here. We've put a lot of context for exactly how we want it to respond to us. So let's now save that, and that's saved our changes, and now going to apply to all of our conversations, all of our chats. And let's go and create a new chat. And what we're going to do is we're going to now Claude to do something that's going to test that that work. So what we've said here if we zoom in a little bit is write a short analysis of Erling Harlan's form this season. So we're talking about one of the players, and we want analysis of that, and it should all match the tone that we've asked for. So let's run that. So, here it is. Let's have a read of what's given us back. So it says Highland's But back season. Last season, injuries and Mid campaign slump handed the golden boot to Mohammed Sala. This season, Holland has made the correct, swift and emphatic. Has 22 goals and seven assists a goal involvement of 1.08 per 90 minutes with a non penalty, and it's got all the stats here. Meaningful rate. That's not luck. It reflects elite shot selection and finishing from close range. And so that language carries on pretty much throughout this. And at the end, it says, at 25, he has surpassed 107 Premier League goals in 126 appearances and crossed 150 goals for city across all competitions. 2024 25 blit looks like a footnote. This is the dominant version back to full speed. So if we now go now that it's written that analysis and have a look at what we asked it to do, what we'll do is again go down here, click on the initial menu, click Settings, and then come back to what personal preferences should Claude consider in responses. We've asked it to use an editorial voice, confident and direct, no hedging, data led, so it's used lots of stats. Grand opinions in stats, not opinion alone. Yep. Punchy short sentences, no padding, knowledgeable seem to be pretty knowledgeable. Assume the reader knows the offside rule, tone rules, never use cliches. I didn't see any of those. No Emojis didn't see any of those, use European date format. Context is focused on premier league Champions League International football. Unique selling point is combining narrative journalism with data analysis, which is why we had the stats mixed in there with the analysis. And the audience is that success fans 18 to 35. So now going back to our chat, thinking about that, only thing that I didn't check or didn't remember when I was checking is the date format, and you can see that this is using the date format that you would find in Europe, which is actually we put the day here and the month there, whereas in the States, it's the other way around. And we can tell that because if you look here, we're only in April at the time of recording this, and you can see this date is clearly talking about February because it couldn't be talking about August, which hasn't come yet. So we know that it's doing the correct format. Apart from that, I'm happy that it's matched everything that we've put in our Custom Instructions in our preferences, so that's a good so let's talk about what not to put in custom instructions or preferences, and let's talk about why we typed what we typed into the live tress. So first of all, what not to put in. So we don't want to put in task specific instructions, so write match previews, something like that, because the purpose of those preferences is to set the tone for exactly the responses that we get back. What we're not trying to do is write any instructions. Also, what we don't want to put in is any one off goals, like telling it that you're covering the Euros this week, things that apply to the current time, because what we're doing is something that should apply to any time frame should be relevant for everything that we're doing. What we also don't want to do is put in process steps like step one, do this, step two, do that. That's not the place for it. What we want to put in is, again, something that is going to guide the format. Any of our instructions. When we put in process steps, it's almost like we're giving it instructions, so we don't want to do that. Now, on the right hand side here, this is just emphasizing that when you're doing a live test of this, the best way to do it is to put in something really simple without repeating anything because the tone is already editorial, stats focused, and punchy when we type in, and that's what we want to test. So what we did put in was something really simple, not too complex. We didn't have to add much structure, but we still got back the response that we wanted. So that's really the power of putting those custom instructions in. In summary, you can find Custom Instructions by going to settings, then general, and then finding the part where it asks for your preferences. And they apply to every conversation. You can set them once, and then they're inherited everywhere. There's no repeating yourself. And you should keep them about the voice and the context of the response that you want back from Claude, not the specific task that you want it to do. That's not the place for that. If you keep it about voice and context, you'll be guiding Claude on exactly the responses you want back, and you'll be a happy individual. Now it's your turn. What I want you to do is write some Custom Instructions for your own site or for your role. Keep it short, and it should cover your role or the site context or your project context, the tone preferences that you have and any key rules that would apply to all of your chats for whatever it is you're doing. And once set, every lesson that follows in this training is going to become faster and more consistent, and everything that you do for your projects is going to become faster and more consistent. So have fun doing that and I will see you in the next lesson. 9. How To Keep Long Conversations On Track Using Context Continuity: In this lesson, we're solving one of the most common frustrations with Claude, which is it started well, but halfway through, it forgot what we were doing. Now, this isn't a bug. It's a misunderstanding of how context works. Claude thinks in conversations, and everything it knows about your task lives in the current context window, and we're going to talk about what that context window is and how to use it. So let's get started. So the feature we're discussing today is Context Continuity, why we use it. We use it to maintain shared understanding across a long conversation. What it solves, it solves drift, repetition, loss decisions, inconsistent outputs. And in terms of what we're talking about here, context is the short term awareness inside the conversation. A memory, what we call memory is actually long term storage across multiple conversations, and we'll cover that at a later date. But right now, what we're talking about is context, which is the short term awareness inside your chat or inside your conversation, what Claude is aware of for the purposes of that conversation. So here we are inside Claude, and we've started a new chat. What we're going to do, first of all, is paste in our prompt, and I'm going to explain what this is all about. So what we're doing here is called anchoring. So we're going to tell Claude what all of this conversation we're about to give it is regarding, and that keeps it anchored to what's in here. So we're saying we are planning the editorial calendar for our football news and stats website for the next four weeks. This conversation covers content themes, article types, publishing, schedule, and data angles. Keep all responses aligned to this goal unless I say otherwise, acknowledge and wait. So what we're doing here when we say keep all responses aligned to this goal is anchoring. We're telling that everything we're about to say relates to what's up here. We're also telling it to acknowledge the information that we give it and to wait. Let's run that. So, as we've asked it to acknowledge is ready to build out the four week editorial calendar, content themes, article types, publishing schedule and data angles. Whenever you're ready let's start. So it's acknowledged it. Great. So now let's put in our next prompt. So the first thing we're going to do is we're going to say, start by identifying the three biggest football stories. We should plan content around over the next four weeks. So this is going to help us with our editorial calendar, and let's run that. So it's told us, here are the three biggest stories to plan around over the next four weeks. We've got the Champions League quarterfinals and semifinals. This is the dominant story of the next month, the last eight PSG versus Liverpool, Real Madrid and Bayern. So it goes into some details about the big stories over the next four weeks for our editorial calendar, gives us some more here, Arsenal's Premier League title running, gives us some information on that. And then the World Cup 2026, the countdown begins, gives us information on that. And so these are the three biggest stories, and it says, These three stories interconnect because Arsenals is managing a title race and a UCL campaign simultaneously with World Cup selection pressure bearing down on the key players. That overlap is where the most compelling editorial sits. So it's giving us ideas for our editorial, and it says, Ready to map these content themes and article types when you are. Okay. So now referring to this, we paste in our next prompt. And we've asked it now. We've asked Claude, based on the three stories you identified, draft a week by week publishing plan with content types each. So let's run that. So on running it, it says, Let me build this out using the fixture calendar, which is essentially the calendar for all of the football sporting events coming up and the different matches that are coming up. And it says, I have everything I need. The bracket is fixed. So the semifinal pairings are predetermined, and then it's giving the semifinal pairings, the teams that are going to be playing in semifinals. It's also created this four week plan editorial calendar that we can use. So there's a four week editorial calendar between these dates. And essentially what that does is it's giving us the calendar by week, so week one, week two, and the matches that are on the kind of stories we should be reporting. So it's giving us the theme, and we're calling it the crunch. And in here, we've got a date and a story on that date and the content type, which is, because it's sports we're dealing with, we can give a match preview, and the data angle, the kind of things we're going to be talking about in terms of stats is arsenal. So one of the teams points per game at home versus away this season, what a win does to the title probability. So essentially, for every one of these matches, we've got our take on it as a sports news organization, the kind of things we can talk about, and we've got it week by week. So what we've shown here is really we have kept everything anchored to what we talked about in the beginning, which is this editorial calendar. And now any question we ask is going to relate to the editorial calendar. And so one example of that is if I type in a really small question, which really could mean anything like so. So I've typed in. What do other companies do for this? Now, for this could mean actually anything. So let's run that. So it already knows it's asking me to be more specific, but it knows that it could be editorial calendars, data led football content, publishing schedules, content planning tools, which you're asking about. So essentially, it knows it's kept the context throughout this conversation, and that's what we mean by Context Continuity. Now, after a certain amount of time, when you start asking questions about different things, Claude will start to forget what it is that you're talking about if you ask questions in specific ways that are not related necessarily to what you've already spoken about. And so it's good to know what to do to bring Claude back on track when you do that, when you start getting output that doesn't relate to what you were talking about or the subject. So I'm going to paste in another prompt here that's an example of doing that. Hill you said, quick reset. We are still building the four week editorial calendar for a football site. The three themes and the weekly plan are unchanged continue from that context. So run that. So there you go. Now Claude says, Confirm. We're back on track. The themes, the UCL quarterfinals and semifinals, Arsenal's title run World Cup 2026 countdown, and the four week publishing plan is set. What's the next element you want to tackle? So, essentially, this is a quick way to tell Claude, L, here's what we're talking about. Everything we're about to talk about from now on, still relates to this. So if you ever start to drift off, this is a way to get Claude to focus back on what you were talking about before. So this is what we mean by Context Continuity. It's the ability to keep Claude focused. And the beauty of Claude is that it has great continuity for long conversations, and it tends not to forget what you were talking about, but if it ever does, you know exactly how to reset it. So in summary, to keep Context Continuity, always state the goal early and anchor the conversation before you start. Build step by step in one thread. Don't restart, so you want to keep building upon what you're talking about. And it's good to reference earlier decisions explicitly to prevent drifting off topic. And that is the best way to keep Context Continuity. So now it's your turn. What I want you to do is run your next editorial planning session or project or whatever it is that you're doing this week or upcoming and run it entirely in one conversation. I want you to start with a clear goal and a clear scope of what it is you're doing in that conversation and then build step by step always what came before. And remember, if you start getting responses that don't appear to be talking about what you're talking about and what your goal is, you can always reset Claude and remind it what you're talking about. And that is called Context Continuity. So have fun with that, and I'll see you in the next lesson. 10. How To Remember Key Details Using Memory: In this lesson, we're going to solve the problem. How do I stop repeating the same information every time I open Claude? So we're going to talk about memory. Now, memory allows Claude to store important details about you and your work across different conversations, not just inside one chat, like context continuity, but over time and across different chats, different conversations. So let's get into it. So the feature we're using is called a memory to persist important details about you and your work across different conversations, different chats. And what it solves is repeating the site info goals and preferences and whatever it is to do with your project or what you're trying to achieve. It stops you having to repeat that in every session and in every chat every time. So again, a reminder that context is short term and that's inside of one conversation, but memory is long term, and that's across multiple conversations, multiple chats. And they work together, but they're not the same thing. So here we are back at Claude. And in order to access the memory settings, what we're going to do is go down here to the initial menu, and we're going to click here and go to settings. And then settings on the left, we're going to choose privacy. Then if you look down here under privacy settings, you'll see memory preferences. So click Manage and you can see up here the memory settings. Now, the important thing is, for any of these memory settings, you have a little toggle here, and there are two memory settings if we zoom in a little bit. So we've got search and reference chats. Allow Claude to search for relevant details in past chats. So if you want to remember what's in other chat, you also need the ability to find what's in other chats in order to do that. So this allows Claude to search for relevant details based on what you've said to it in past chats, which we'll see in a little while. And then the other thing is generate memory from chat history, and this allows Claude to remember relevant context from your chats. And this setting controls memory for both chats and projects, which we'll come to later. But for now, let's talk about chats. So if we these two both on, so it gives us the answer to this question. What does Claude know about you? And it says not much yet. After more chats, you'll see what Claude knows about you here, and that will all be stored in here. So just to remember that there are two parts to this. One is the ability to generate memory from your chat history, and we've switched that on, and the other is the ability to actually search for that and search and reference chats, different chats. And that allows Claude to actually find the information in its memory. So we switch both of those on. So now let's head back to our chat, and we're going to start a new chat so what I'm going to do now is I'm going to tell Claude to actually remember something, and I'm going to test that. It's remembered it, and it's remembered it across different chats. So if I paste in this prompt, so here we've got this prompt which says, remember this, and this is the keyword here. This tells Claude to do exactly what we've said, which is remember what's to follow, and that puts it in memory. So remember, I run football news A Stats website focused on data driven journalism, primary coverage, Premier League, Champions League, international tournaments. My role, founder and lead editor, preferred writing tone, confident, punchy, data led, no cliches. I usually need Claude for article drafts, data analysis, editorial planning, and SEO content. So that gives it a full overview of who I am and what I do. And I'm going to add a little bit more so we can test Claude and make sure that it's actually remembering. So I've added that whenever I type the following, I want you to replace it with a football image. And I've put the word Imogi in brackets, and this is something we can test quite easily is being retained in memory. So let's go. So we can see just now it said managing memory here, and it said it's used two tools. It's saved. Whenever you type Imogi, I'll swap it for football Imogi and that's that right here. What are we working on? So let's test that straightaway. I like the fact that he actually told me it was managing memory. So this should have gone into memory now, but we're going to so let's type in the following prompt to test it. So I have typed in the prompt. I like playing cards and emoji, and I'm going to run that. So here it says, I like that card and it's put the football Imogi. So that's a good test, but as we're in the same chat, if we want to test memory, we actually need to create a new chat. So let's take this prompt and let's start a new chat. Let's start a new chat up here, and let's place in the exact same prompt and run it. And there you go. Two very different hobbies, playing cards and the football emoji in it's remembered that even in a new chat, and that is memory working at its finest. So now let's do one more test. If we go up and start a new chat again, and I'm going to place in the prompt. Help me write a short editorial note for this weekend's fixtures, and I've also added to include Emoji at relevant points. So let's see what it does. So we can see it's now searching through different tables, and it says it has enough to write the editorial. Here's the weekend preview. So here we go. It's written an editorial note. It says Weekend editorial for Match week 32, which is 11th to 13 April 2026. The fixtures are here, and the biggest number in English football right now is nine. That's the point separating Arsenal, that team, and then it's put a football emoji, just like we've asked up here from Manchester City at the summit. And this is a brand new, so it's shown that not only has it remembered how to write the editorial the way that we told it to, it's also included the Emoji that we told it to, and it's remembered that because we put into memory to remember to use this Emoji every time we say, do this. So this is a good way to show that memory is working in different chats. What does Claude memory actually do? Well, when memory is enabled, Claude can save important facts and reuse them automatically in future conversations without you restating them. So I didn't have to tell it to use the emoji in brackets. It just knew because it was a memory. What is a good memory item? Good memory items are things like when we're building a website like we are now, the site name, the editorial focus, my role, preferred writing, tone, key coverage areas, the main things that we want. And bad memory items would have been this week's fixtures, things that could change temporary campaign ideas, and one off article angles or anything one off. We don't want to store that in memory because then it will keep getting used in different chats, and it will remember things we potentially don't want it to remember. So in summary, memory persists key facts across different conversations and different chats, and context does not, that's short term and within one conversation. You can enable it in settings privacy memory at this time. You should use it to save stable info, never save temporary or changing data in memory because it's going to interfere with your different chats, and you won't be able to get rid of it as easily. And the other thing is to review and manage your memories regularly, which are in settings. So go ahead and try that in your settings and experiment with what you want to store in memory, and I'll see you in the next lesson. 11. How To Keep Outputs Consistent Using Constraints: Lesson, we're solving a very common real world problem, which is why does Claude give me good answers, but not in the same format every single time? Now, this is where constraints come in. Constraints let you control the shape, the length, the structure, and the rules of Claude's output so that results can stay consistent and reusable. So let's get into it. So the feature we're using here is called constraints. Why we use it to control the format, the length, the tone, and the structure of whatever output we're getting. It solves it solves inconsistent outputs, constant reformatting, and wasting editing time when you're with Claude. So the key thing here is that instructions tell Claude what to do. Constraints tell Claude how the output must look. They're complimentary, but they're not interchangeable. They're different things. So let's get into it. So here we are in Claude, and we're going to start a new chat so I can show you how constraints work. So first of all, let's paste in our prompt. And it's a straightforward prompt. It says, write a match report for Arsenal 21, Chelsea. So Arsenal have scored two, Chelsea have scored one, and we are going to write a match report for this, but we haven't provided any constraints. So let's see what happens when we run so here we go, we've got a match report for this football match. It starts off with the score. It's got the league, the date. It says it was played at Emirates Stadium in Match Week 28. Corner obsession fuels title push, so that's our headline. Didn't need to be pretty. They just needed to be rufless from sep pieces, and once again, they were. It goes on about the players, how they played. We've got a few paragraphs here. And then afterwards, it's given us another heading which says, the number that defines Arsenal season. Arsenal 16 goals from corners in the Premier League this season is the joint most by any side in a single campaign, gives us some more paragraphs. Then it tells us who was on the team, people who got red carded, who got sent off, that kind of stuff. So this is what's given us by default, but we may not want this format. We may want it either shorter, longer, or different layout, and this is where constraints come in. So let's paste in our prompt for our constraints. So this is our prompt. We've now said, write a match report for Arsenal to Chelsea one using these constraints, exactly three paragraphs. Each paragraph no more than three sentences. Lead with the decisive moment, not the score line, include exactly one sep per paragraph. No match commentary cliches, confident editorial tone. So we've been really specific, and these are our constraints. So let's see what it does differently this time. So here we go. Here's our output. So we've said exactly three paragraphs, one, two, three. We've said each paragraph to be no more than three sentences. For example, I can see one full stop here. That's one sentence, two sentences, three sentences. It says, lead with the decisive moment, not the score line. As you can see here, it says, Durian Timber's 65th minute header from in swinging corner settled a fractious London Derby and put Arsenal five points clear at the Premier League summit. So, presumably, this was the most decisive moment. And it says to include exactly one stat per paragraph. So here it says, Arsenal have now scored 16 goals from corners this season. And so that's the stat here. And in this one, William Siba headed Arsenal in front from Sakers corner on 20 minutes is actually the stat. I wasn't actually sure when I read that, I checked with Claude, and Claude said that that was the stat, so it considers that a stat. I suppose it is. I was looking for something a little bit deeper. And that's something we can more specific about. And in the third paragraph, it says, handing Chelsea their seventh red card of the league season. So there's the stat there. And it says here, no match commentary cliches. I couldn't find any looking through, so that looks to be correct. And especially in football, where there's quite a lot of band told common phrases, match commentary phrases. I can't see any looking through it. And it says a confident editorial tone, and I think we've got that. So that appears to have worked. So those constraints have given us this, whereas before, without the constraints, you can see how much we got back. It's much longer. There are many paragraphs. We've got we're leading with the score. We've got two different headings. So without telling Claude exactly what you want, it's good to know that you can add more detail if you want to get back something that's more detailed, but without telling Claude exactly what you you're always unsure what you're going to get back. So now let's give it some more structure with these constraints. If I paste in another prompt, if I paste in this prompt, which is write a match report using this structure. Paragraph one, the decisive moment and its context, paragraph two, tactical story of the match with one key stat, paragraph three, what this result means going forward, and then the constraints, three sentences per paragraph, editorial tone, no cliches, one concrete stat paragraph. So we've added in a little bit more here. So let's go for it. Let's see what happens. Here we go again, it's pretty much the same format, except we've made a few changes, which is in paragraph one. It's going to test the decisive moment and its context. So it says the goal was not a fluke. It was the inevitable product of a system Chelsea had no answer to. So that's the context there. Paragraph two, tactical story of the match with one key stat. I said in here, Chelsea finished with 58.9% possession and still lost. And it says the stat that best captures how thoroughly Arsenal controlled the match, they barely seem to be in. And then paragraph three what this result means going forward. And it says down here, Arsenal five points clear with the title race entering its final stretch control their own destiny. And on this evidence, they know exactly how to win ugly. So the key thing here is that you can change these constraints anyway you like to change the output that you get, and it will slightly alter this output based on these constraints. And that's something that you can use in whatever it is that you're creating. So in summary, constraints control the shape, not just the content. It's the shape of the output. Apply constraints directly inside the prompt, and Custom Instructions, which we talked about before, sets your voice, whereas constraints set your format so you can use them both together. Now it's your turn. What I want you to do is take your most repeated content task, something you do all the time, and write a constrained prompt with three to five rules around the structure, the length, the tone, and the exclusions. So whatever you're doing in your project in your day to day life that involves creating some kind of content, some kind of text. I want you to apply the constraint with the length, tone and exclusions that would be best for you and whatever you're doing. And that way, you're really going to see the power of using constraints and how they help you to avoid repeating yourself every single time just to get the same format. So, have fun doing that, and I'll see you in the next lesson. 12. How To Turn Notes And Files Into Finished Documents Using File Uploads: So in this lesson, we're going to talk about how to turn notes and files into finished documents using file uploads. So in real work, most tasks start not from a blank page, but with some rough notes. We've got some data exports as files. We've got messy drafts, stats, and spreadsheets. And today's goal is to turn real world files into finished content for our football site. So let's get into it. So the feature we're using is called file uploads. Why we use it to read and work with real documents, spreadsheets, and PDFs, turning raw inputs into usable outputs without manual copying. So let's go and do it. So what I have in front of me are two files that we're going to use to generate an overview of a match. And it's not meant to look neat and tidy. It's meant to look a little bit messy so that we can test that Claude can actually read this data and do something complex. So let's start with the first file. The first file is editor notes, and these are notes about the match. So if we zoom in slightly here, we can see what we've got is post match debrief notes, Arsenal versus Chelsea Arsenal win 21. We've got the location here, who the notes type by, and you can see it says that it's from a voice memo, so it's a bit rough. And it's basically split into the main story. The main stats to highlight because this is what we want in our post match review, angle highlights, things to check or fix before publishing. And then a note about the tone, and it says, Don't be boring, which is good to know. Don't be boring about this. Arsenal looked like a title team today. Chelsea, look lost. Say that. Okay. Some people may disagree with that, I suppose. But essentially, what you've got is some notes from a match. And then our next document. So this document is about match stats, and it is very rough and not very tidy, a little bit all over the place. And that's exactly for the reason I said so that we can test how Claude would deal with it. So up here, it says the same match Arsenal versus Chelsea and the date. Over here, it's got a Premier League and match day 31 along this top line. And it says the venue here, the attendance, and it's got for each team, all of the stats, it says her team stats, the number of goals each team had shots shots shots on target, expected goals, possession, and all the usual kind of stats that you'd get for football passes, pass accuracy, corners fouls, yellow, red cards, and offsides. And it's got the goal events. So what it's doing really is it allows you to compile in this document a whole bunch of different stats that we can use in our match review. And so what we're going to do is we're going to upload this into Claude, and then we're going to ask it to analyze this data. So it includes top player stats here, then other pressing stats here. So essentially, if you were to look through this, it's a little bit of a nice sort to look at. I mean, you can kind of pick out the information, but let's see what Claude will do when we give these two files, we're going to give it both this text file with the editor notes and the match stats export and see what it does. So let's head back over to Claude and see what happens. And we're going to start a new chat. And in here we're going to type in our prompt or paste in our prompt, I should say. So our prompt says, using the uploaded stats export and editor notes, create a match analysis article with these sections. We want a match overview of just two sentences, the key tactical story, standout performers with stats, data insight, and that's one key stat explained. This means for the table in the league table, and the tone should be confident editorial, data led, and a max of 400 words in total. So the only thing that's missing is the actual files. And the way you do that is you go down here to add files, Connectors and more, and we want to choose ad files or photos. And here are my two files, so I'm going to select both of those because that's what we'll be using to write this match review. And there they are. They've been added in. So you can see them both here. And now we can run it. So using the uploaded stats Export and editor notes, create a match analysis article with the sections that we've just read out. Let's do it. So there you go. So this is the output. We've said we wanted a match overview of two sentences. There it is. We've got sentence one, sentence two. Tell the key tactical story, and here it is key tactical Story. And that's in a paragraph here. We want to talk about the standout performers with the stats associated to them. So if you look here, it's highlighted these other standout performers. It's talking about their stats that it got from the stats document that we uploaded, four shots, five key passes, et cetera. We've also said we want some data insight about one key stat, which is going to be explained in a bit more detail, and that's here. And we want to talk about what this means for the table as in the league table that these teams are in, and that's explained down the bottom here, so we can see that it's explained that. And if we zoom a little bit more in, you can see what it's explained. And all the while, as usual, we've set the tone to be confident editorial data led and a max of 400 words in total. So if we wanted to test that, we could actually copy out all of the text, and then we could actually start a new chat, paste it in what it's done, and say, how many words is this? So there you go. Let's run that. And there you go. It says 370 words, which is right in line with what we asked for in here. Where we've said MAX 400 words. So that's almost up to the limit, but that's good. Good job. And now, because Claude already has context about what we've uploaded in these documents, not only has it generated this report, but it's going to remember these files because it has Context Continuity, we can paste in another prompt and know that it will refer to these files. So let's do that now. So our prompt is rewrite this as a shorter news item for our homepage feed, Max hundred 20 words lead with the most surprising Stat. Oh. So here we go. Here's the newly shortened rewrite. And this means the expected shots on goal. So it's saying Chelsea managed just less than one shot on gold, 0.84 at the Emirates, barely half of Arsens 1.73. So it said Leeds most surprising stat. So it's telling us that is the most surprising stat. And we've got a nice short summary now. So the main thing is not only can we upload files, get it to do some analysis, and rewrite things in a more readable format, but it remembers that context, and then we can tell it to either re analyze it or make it shorter or longer or whatever we want there. Really key and really helpful. So some common mistakes to avoid are uploading files without saying what to do with them. Claude would then either ask you what you want to do or make some assumptions. Another common mistake is assuming that Claude knows the desired output format. You really have to tell it what you want to do and in what format you want it to put out the response in. Another common mistake is uploading unrelated files in the same conversation, and that might get a bit confusing for Claude. So remember that files give Claude input, but you still need to give direction. So in summary, in order to add files, you click the plus sign and then pick Add files to upload any into a chat. Remember that Claude can read multiple files simultaneously so anything related can be uploaded at the same time. And remember that you can refine the output from Claude without re uploading because it remembers the context, so you don't need to upload a file, ask a question, then reupload the file and ask a question. Sometimes it may be helpful if you find it drifting, but generally Claude remembers what you uploaded with Context Continuity. So that's that. Go ahead and try that, and I'll see you in the next lesson. 13. How To Analyse Data And Spot Patterns Using Data Analysis & Charts: In this lesson, we're going to talk about how to analyze data and spot patterns using data analysis and charts. So for a football stat site, data isn't optional. It's the actual product. And today, we're answering, How do I use Claude to turn this football data into insights, and you can use that for any data. And how do we do that without being a data analyst? So let's get into it. So the feature we're using today is data analysis and charts. Why are we using it? We're using it to explore and interpret structured data to generate visual chart outputs. And what it solves, it solves making sense, in this case, of football stats quickly and turning them into publishable content, and you can use it for any stats or any data. So here we are at Claude. And the first thing we're going to do is we're going to upload our file of football stats, which we're going to analyze. So let's go over and have a look at that file before we upload it. So this is our file of football stats. It's for the Premier League, if we zoom in slightly here for Premier League 2025 to 26 season stats on Match day 31, as of this date. And if we go back over, we can see the kind of stats it's got. It's basically league table stats. So we can see all the teams in this column. We can see, for example, Arsenal has played 31, 120, drawn seven, et cetera. It's everything you would need for the league table. And it's also tells you the difference between the number of expected goals for Arsenal and the number of expected goals against Arsenal. We've got how much possession this team had 61%, how many shots per game, shots on target, past accuracy, and all the information that we basically need for this data analysis. So now, what we're going to do, we can also see that there's some information down here as well. So what we're going to do is we're going to load all this information, the premiere league stats into Claude, and then we're going to do some data analysis and generate a chart based on this information. So now let's head back over to Claude and do that. So here we are back at Claude. The first thing we want to do is upload the file. Click plus, go to Ad files or photos. You've chosen our file, which is Premier League stats, open that, and we can see that that's attached. And now we want to paste in our prompt to do the data analysis. Our prompt says, based on the uploaded data, which teams are significantly over or underperforming their expected number of goals? Is there a correlation between possession and lead position? Which three teams have the best form over the last six games? So this is our little query. Let's see what we get back. Let's run it. So you see Claude working here and showing us where it's doing. So here, are the results to recap, we asked which teams are significantly over or under performing their expected goals. So it says, expected goals over and under performers. Biggest outlier is Brighton. We've got Ipswich, Lester, so it's just telling us the teams. It's answering the question. We asked, Is there a correlation between possession and league position? If you look down here, it says the correlation is zero -0.854, which means extremely strong, and it gives us some more information about that correlation here. And it says, in relation to this question, which three teams have the best form over the last six games? And it says the best form is number one arsenal, five wins and one draw. No losses there. Second, Man City, four wins, two draws, no losses. Third is Liverpool with four wins, one draw and one lost, and then it's got a number of points here. So now we can see which teams have the best form. So that's analyzed the data. So now we've gone to our next prompt. So our next prompt says, based on this data suggests three specific article angles for our site that would genuinely surprise a stats literate reader, give each a working headline. So we want something that people who are really into stats will be surprised by. So let's go ahead and run that. We've got three angles. I won't go into too much detail, but we've got Brighton are the Premier League's most expensive problem, for example, we've got some paragraphs explaining that, and then a working headline Brighton's expected goals isn't broken. Their finishing is. But after 31 games, does the distinction still matter? So we've got a headline, we've got some info based around stats, and it says, The story isn't that Brighton are underperforming, it's the scale. A 14.9 goal gap between expected goals and the actual output is historically anomalous, and then it goes more about that. But the main thing is it was able to pull out the data and give us back what we want. And now let's go my favorite of all actually generating a chart. So let's paste in our prompt. And if we zoom in a little bit, we've create a bar chart comparing the expected goals versus actual goals for the top eight teams, label each bar clearly, use a clean, minimal style suitable for embedding on a sports website. Let's go. So here we go. We've got our bar graph. And what it shows is the actual goals versus the expected number of goals. We can see Arsenal Heels, we're expecting more goals than we actually got. And most of them follow that same pattern. And then underneath it says a few things jumped out visually. Liverpool are the top scorers in the top eight, but still sit 4.8 goals below their expected goals. The gap is consistent across the board. Tottenham are the only side in the top eight actually outperforming their expected goals, scoring 0.8 more than expected. Hover any bar pair for the exact Delta. So if you hover here, you can see if I zoom in a little bit, and I hover here. You can see it was 58 goals, actual goals, and it was 4.4 less than the expected number of goals. If we go over to Tottenham, they actually scored 49 goals, which was 0.8 more than the expected number of goals. So they're the only ones that scored more than the expected number of goals. So Claude's pretty good at picking out the information, summarizing it, and drawing for us, as you can see, pretty good. So in summary, this data analysis exercise allowed us to upload a structured CSV or Excel file, do some data analysis, and it activates automatically. We didn't have to press any other buttons or run any functions. It allows us to ask questions in plain English, and then Claude handles the math. And we can ask for charts explicitly, and they appear on the right hand side in what we call the Artifacts panel. More about that later. And also, we move from data to article angles. So in other words, different ways of expressing the article, and from data to article to downloadable visuals all in one session so we can change what we actually want just by typing a different prompt. Pretty cool. So now it's your turn. What I want you to do is take any football stats file you have or any file. It doesn't have to be football. A file you have, upload it and ask what stands out? What's changing. What would surprise a stats literate reader? It could be something for your business, for your project. It could be anything at all if you're building a website that uses stats or anything at all that you're interested in that uses some kind of data. And then you can also visualize that data in different ways, as an article, as a summary, and even as different types of charts. So go ahead, have fun doing that, and I'll see you in the next lesson. 14. How To Check Your Output Using Verification & Critique: So in this lesson, we're going to move on to check your output using verification and critique. Claude has helped you generate ideas, write articles, and analyze data. Now we're solving a critical problem. How do I know if Claude's output is actually correct before I publish it? So for a football site, getting facts wrong, like wrong score lines, wrong stats destroys credibility fast. So let's look at how we can solve that issue. So the feature, more like a technique that we're using, is verification and critique, why we use it to get to check, challenge and validate its own outputs. And what it solves is over confident claims, factual errors, reputational risk from publishing bad data. You see Claude is designed to flag uncertainty, surface assumptions, and critique its own reasoning, and this is one of its genuine advantages for professional publishing workflows. So let's get into it with an example. So here we are back inside Claude, and let's start off with our prompt. So we've asked Claude to write a 300 word analysis of Liverpool's defensive problems this season. So that's the team we want to know about. And we want their expected goal average and any pressing stats about it. So let's run that. So this is a 300 word analysis of Liverpool's defensive problems, referencing their expected goals against. It says, The press is lying to Liverpool fifth place, ten defeats for a club that won the title under just 12 months ago. Liverpool's 2025 to 26 campaign is a structural problem, not a run of bad luck. So that's what they're going with, and then there's an article 0N however, now we're going to ask Claude to verify itself as follows. So our new prompt says, Now review what you just wrote and one, identify any specific statistics or claims that require external verification. Two, flag any statements that could be outdated given your knowledge cut off. Three, highlight anything that sounds authoritative, but it's actually an assumption. Four, suggest where a human editor should fact check before publishing. And then finally, do not rewrite yet, just surface the risks. So let's run that and see what happens. So here we go. We've asked Claude to essentially verify itself, and he said, Good call. Here's a full risk audit of the piece, and it says, number one, statistics requiring external verification. So it says the 1.2 expected goals against per match, 1.35 actual goals against are sourced from 40 stats and needs cross referencing against before. So told us about some stats that we need to cross reference. So that's an example of what is given back there. Number two, it says, outdated or cut off risk statements, things like won the title just 12 months ago. The standings data shows Liverpool in fifth with 52 points, which implies a down season. But the title claim is inferred from memory context. It's not confirmed by search results, so it needs verification. So it's found this from memory of what we've done earlier, but it hasn't actually gone to the Internet and checked that this is actually true. That's another example, the third authoritative sounding assumption. So it says things like the space behind the defensive line. So this is a narrative assumption. There's no data on it. We may want to keep things like this because there's a kind of thing you'll put in a review, but it's good to know it sounds authoritative, but it's not necessarily true. That's what it's saying. It's an opinion. There's another example. And then it's given us the priority fact checks for a human editor. We've got all kinds of things here. So one I'll pick out is Van Dix age written as 33, and the risk levels high. Action is to correct it to 35, so it's done some more research on that and found that. So this is interesting to see that even within the output from Claude, there are some things that it's not 100% certain of, and going over it again, it highlights some of those things. So the important thing here is the first time you get output from Claude, especially if you've got some stats or some information you want to really be sure of, it's worth going over and verifying again. Now with that knowledge in mind, let's get Claude to rewrite the article. So now we've put in the prompt, rewrite the article, keeping only claims that can be reasonably supported. Soften or remove anything that needs external validation, flag two places where we should insert Ralsts from our data provider. Let's go. Okay, so now we've got a rewritten article. It still says the press is lying to Liverpool, so it's still got that attention grabbing headline. And it says fifth place, ten defeats for a club that won the Premier League title last season. The drop off is not marginal, it's structural. And if we go back and look at the original, you can see here in the original, it still said fifth place, ten defeats for a club that won the title under Arslah. So it doesn't speak about Anslh just 12 months ago. Liverpool's 2025 to 26 play is a structural problem, not a run of bad luck, so they've taken off this part. So obviously, just looking at the feedback we've got, Claude now decided not to make reference to particular players. It's trying to be strictly factual. Now, we might not do this in a real article, but this is just a demonstration when we want it to be cautious, we can do. Another thing that I've noticed, if I go back down to this rewritten run, there are a couple of places in here where it says to insert information that it wants us to confirm ourselves. So here it says, insert the current expected goals against per match and the Premier League clubs from the data provider. So it's asking us to go to a data provider, get the actual information and insert it here because it can't access that. And it's done something very similar down here as well. So this is just showing that really we can get an article written, but make sure everything's factual. And if something isn't factual, it will tell us, and then we can just find that for ourselves, do our own research and insert so here are some verification prompts you can re use. So no matter what you're doing, whether you're dealing with a football match article 0R anything else, you can use these prompts, which are which parts of this answer might be incorrect or outdated. What assumptions are being made here, and where could this be misleading to a reader? And what this does is before you're putting information out, asking these questions allows Claude to tell you where you need to be more specific and more accurate, and it prevents you, as it says here, misleading the reader. So in summary, verification is a prompting technique, and there are no settings needed within Claude. You should ask Claude to critique its own output before you publish, especially if you're writing something which you really need to be accurate before it goes out to the world. And for our example, we should always verify recent match data from sources. So depending on what you're doing, you should always verify recent data from primary sources. Now it's your turn. What I want you to do is take something Claude wrote recently and ask it to identify the assumptions, flag any risks, and suggest what needs external verification. Try the three reusable verification prompts I just gave you on your next piece of content before publishing, and you should find that you've got something that's a lot more accurate where you've taken the time to work out, what you can put out, what things you want to make assumptions about, and what things you want to verify. So go ahead and have fun with that, and I'll see you in the 15. How To Think Through Complex Problems Using Reasoning & Planning: So now we're using Claude differently as a strategic thinking partner. Claude is genuinely strong at multi step reasoning, and that's one of its biggest advantages over other AI tools. So let's get into it, and I'll show you exactly how. So the feature or technique we're using is called reasoning and planning for multi step thinking across complex multi variable problems, which just really means stepping step by step across problems that are really pretty difficult to solve, and there's a lot of changes, a lot of variables. And what it solves, it solves making strategic decisions without having to have clear data and planning when there's uncertainty. So here we are at Claude, and we're going to paste in our prompt. And a prompt says, I need to make a strategic decision about our football site's content focus. The context is we currently cover all Premier League news. Our traffic is growing, but engagement is low. Competitors are bigger and faster on breaking news. Our strength is data driven. Think through this step by step. One, what content strategy would most differentiate us from our competitors? Two, what are the trade offs of going deep on fewer clubs versus broader coverage? Three, what is the highest leverage move in the next 90 days? And we said, Don't give me a list of generic ideas, give me a reasoned recommendation I can act on. So the difference here is we've made it very clear we're making a strategic decision, and we want it to really go deep, and we want it to analyze based on all the information it has and not just give us a list of ideas, but a reasoned recommendation. And reasoning takes quite a lot of compute and processing power. So this is an example of where we can use our model selector at the bottom to choose a model that is better at reasoning and deep thinking. And so if we go in here, there is Opus 4.6. It uses, as you can see, it's popped up here, Opus consumes usage limits faster than other model. It's also most capable for ambitious work. And by that, at the moment, we mean some kind of deep thinking. We won't switch on extended thinking just yet, because that will require even more compute and processing power from Claude in the background, but it will switch this on if it needs it. So let's go for Opus 4.6 and let's run it and see what happens. So here we go, it's actually coming up with an answer pretty quickly, so let's let it finish and we'll go over. So here we go. It's giving us the feedback. Let's zoom in a little bit. And I said, Here's how I think through this. The core problem competing on breaking news against other outlets with bigger newsrooms and faster pipelines, that's a losing game. Your engagement is low because broad, shallow coverage doesn't give the reason to come back to you specifically. They can get the same match report anywhere. Your MT is data analysis. The strategy should widen that not try to fill someone else's. And then it says, Our differentiation is on the analytical layer Premier League coverage. Stop trying to be a news outlet that also does that. Flip it. You're a stats and analysis outlet that contextualizes news. So it's basically telling us what it used to what it was telling us earlier about being the people to go to people for stats. And I won't read through all of this, but it's basically breaking down what our strategy should be. Number two is deep versus go Broad on data, deep on narrative. Going club specific, say covering only the top six would shrink your addressable audience without a proportional gain in loyalty. A Brentford fan who loves the expected goals analysis would bounce the moment they realize you don't cover them, and then it's given us another 0.3 highest leverage move in the next 90 days. One thing done well, launch a weekly data driven newsletter with a signature format. And it goes on to tell us why it works. So it's really breaking down the reason in here as we've asked it to. So we're going to do a little bit more and kind of tell us a little bit more if I pasted in this prompt. So if we zoom in here, before I read this, it's good to know that sometimes one of our greatest strengths when using something like Claude is to actually get it to give us the opposite opinion, and then we can look at both those opinions. So if we get it to argue against itself, sometimes it exposes weaknesses in its own arguments, or it makes it really clear which one we should so that's why we've asked it. What's the strongest argument against that recommendation? What would need to be true for it to fail? And so that way, we're basically challenging what Claude's already told us. We'll leave it on Opus 4.6 for now in case it really needs to have a deeper look and think through this, and I think it does. So we'll leave it on this model for now and let's go. So we've got the response. If we zoom in a little bit, it says, Good question. Here's the steel man against. Strongest counter argument analysis doesn't scale and you starve before the newsletter compounds. The recommendation assumes you can survive and grow on the lower volume, higher depth content while a newsletter audience builds. That's a six to 12 month bet. The failure mode is straightforward. You cut news volume, your search traffic drops immediately because Google rewards freshness and volume for sports queries. Your overall numbers dip, and whoever controls the budget panics and pulls the plug before the strategy has time to work. So this is the counter argument to do. And then we can just kind of skip through to the end. And it's telling us, even though it's giving us a counter argument, it's saying what it would do to de risk this from failing. So it says, Don't cut news coverage on day one, run them in parallel for 30 days, publish the newsletter alongside existing output, and measure the sign up, open rate, and click through independently. If after a month, you're seeing 40% plus open rates and organic subscriber growth, you've got signal that the audience exists and values the product. So basically, run both the ideas and if it works, what we're thinking about doubling down on more stats, then we're in a good place. So we've used OPAs 4.6 to really go deep and think about this. So that just means we can trust the response a bit more because it's used more data and more processing power. At this point, I want to summarize its findings, and I'm going to paste in this prompt. And this says, summarize your reasoning into a one page decision brief. The situation, two sentences, options considered, three bullets, recommendation, one clear statement, key risks, two bullets, first action step, one sentence. And I'm not going to use Opus because there's no real deep thinking or deep reasoning going on here. It's just telling it to summarize. So I'm going to go back to my model and go back to Sonnet 4.6. And what's come back is actually interesting. So it said, What decision would you like the brief to cover? Once you point me at the topic, a transfer, a tactical setup, squad selection call, a publishing angle, anything, I'll build a brief around it. And what I can see is it's actually forgotten everything we did before, and that's because we changed a model. So what I'll do is I'll go back to our list of chats. I'm going to go in here and I'm going to look down here, and I'm actually going to copy the question and everything that it gave me back as a response, copy that, and I'm going to go back to the chat where I wanted it to summarize. Paste it in. And above that, I'm going to tell it that this was from a previous chat and I changed model, so I want to get my summary here now. So since I started a new conversation and Claude forgot what we were actually discussing, I've put in this prompt, which is the below is from previous conversation. I'm continuing the conversation here, and now I'd like you to summarize as instructed above. So then it's going to use these summarization points that I gave it. So let's run that. What I've placed it in, obviously, is the previous conversation. So let's go. Okay, so that's what this time it said. The full original conversation about the editorial strategy isn't in my searchable history. But I have the ful text you've pasted. That's enough to build the brief accurately. So now it's given me a decision brief, which is an analysis of the first editorial pivot. It's just going to be analysis of how we could change our strategy. It's given us the situation that the site's current model relies on high volume news coverage to drive SEO traffic. It's gone through the options considered a four pivot to analysis first output, maintain current news heavy model or a parallel run between both approaches, just like it said earlier. And it's covered the recommendation, which is to run the newsletter as an additive product for 30 days. And then it's gone over the key risk of revenue model might be too dependent on CPM. That's a cost for 1,000 views. And the addressable audience for stats driven football content in the UK may be structurally small. And then it said the first action step is to publish the first newsletter issued this week alongside existing output, track the open rate, click through and subscriber growth daily for 30 days before making any structural changes. So it's basically summarize what it gave us back. That's that. So in summary, reasoning is available in every chat. There's no special setup. At the moment, I'm using Claude Opus 4.6 for maximum depth on complex problems, and that's available, obviously, on paid plans. That's why I switched to 4.6 because we wanted to do a lot of reasoning. We should always frame problems with contexts, some constraints, and a clear output format so that Claude knows exactly how we want it back and what we're trying to achieve. And it's always good practice to ask for the counter argument against whatever it comes back with to stress the recommendations. So Claude can pay the devil's advocate. That's really going to be helpful when you're trying to come up with recommendations you can believe in. So now it's your turn. What I want you to do is take a real strategic decision facing your site or your project or whatever you're doing right now and brief Claude to think through that step by step. I want you to ask for a recommendation and then ask for the strongest argument against that recommendation, and that's going to give you confidence that you've actually thought it through and Claude has helped you to do that. And based on that, you should be, as I say, really confident in your decision, and you've used the power of AI to help you to think through it. So there you go. Have fun, and I'll see you in the next lesson. 16. How To Work With Large Documents Using Claude's Long Context Window: So one of Claude's biggest technical advantages is its context window. Claude can hold the equivalent of an entire season's worth of match reports, a full competitor site audit or thousands of rows of player data all at once in one single conversation. So we're going to get into why that's a useful feature in this lesson. So the feature that we're looking at is called the Long Context Window when we're processing entire documents, datasets, or histories all in one conversation. And what it solves is it solves losing context in the middle of a project, having to chunk and re upload large files. We don't need to do that because Claude can hold so much in its context window. And it's important to note that Claude is actually significantly at this time, significantly stronger than ChatGPT for long context work. So, for example, a 200 K tokum window holds roughly 150,000 words. And that's as I said before, an entire season of weekly articles, a 300 page scouting report or 50 match data files simultaneously, all in the same context when done. If you're not sure how big that is, that's absolutely massive for a context window in LLM in an AI chat bot. So how to use this and where to find this in Claude. So the 200 K at this time, token Context Window is available in all the pay plans, pro, MAX, team enterprise, and there's no settings to configure. All you need to do is upload the large files or paste the long documents into the chat inside your project or inside your chat. Now, if you're in a free plan, free plan users have a smaller Context window, and you'll need to upgrade at Claude AI slash upgrade or just go to claude.com and find the upgrade option for that full 200 K window. For the enhanced even more, a bigger Context Window, an enhanced 500 K to 1 million token window, this requires a different plan such as the enterprise plan, and that's available, and you can use it in Claude code sessions with Claude Sonnet 4.6. So a tip is, use a project to keep large reference files available across multiple chats without reloading, and you can see how to use a project in some of my other lessons. But for now, let's get into Claude, and I'll show you exactly how we're using our large Context Window to actually do some work. So here we are at Claude and if I zoom in a little bit, underneath the chat, you can see it says, you've used 90% of your session limit, get more usage. And I actually clicked did this earlier. So when you click on this, it allows you to get more usage because what we're going to do is going to be quite intense in terms of documents. So to do that, if you're on a paid plan or a free plan, it will pop up this get more usage thing and you can click on it, and then you can either buy extra or upgrade your plan. You have to work out the mask what makes sense, but for me, it made more sense to just buy a little bit more extra usage. So if you click on this, you can top up $5 or $10 or just a little bit, and you can keep adding to that if you need more. What we're actually going to do today is get information based on a bunch of documents which together are going to use the large Context Window. So let's have a look at those documents. So in front of us here, we've got three documents, one, two, three, and I'm going to go through what they are. The first one is similar to what we've seen before. These are editorial notes for Arsenal 2025 to 2026 season review. So for the team Arsenal, they're compiled for the full season review. So we're doing a full season review editor. So these are rough drafts that need shaping, and they will help us to create documents that we want to create. And it's got information about key stories that we need to develop about particular players, turning points. So turning points in different matches. So, for example, on Match Day four, it was Manchester City versus Arsenal away. They drew 11, and there were some things you want to discuss. It says, the point where the press where the press started taking arsenal seriously as title contenders, the expected number of goals was almost dead level, which hadn't happened at the Etihad, which is the stadium against Manchester City in years. So this is something that hadn't happened in years. Is something that we might want to discuss in a review as a talking point. Then there are some stats to check and verify before publishing such as certain players goals, attendance figures, things that we want to check over. Tone guidance, which is, as before, it says, Don't hedge. This is a top of the table team that has looked the real deal. So it gives some guidance on the tone of how we're going to write article. Then we've got a bunch of article ideas for the season review package, so a bunch of different article ideas, a long form article 0N how Arsenal won the title race. So essentially all of these are things that are ideas that the editor has come up with or things that the editor has observed that we might want to talk about. So that's the first document. The next document is the season stats. So for a whole bunch of teams here, if we zoom in a little bit, you can see that we've got the date, the opponent that team played against, the venue, whether it was home or away, the score, result, was it a win draw or loss, expected number of goals, four expected number of goals against, and then the difference between those two possession, the number of goals arsenal scored. And so, basically, we've seen this before. This is the number of stats that we can use in any of our articles. Then the third document is a report again for Arsenal in the premier league based around these stats. So here you can see different match days and the score, who they played against, and some stats here as well. And then down here, there is a report, a small match report, which is dominant opening day, a player sacker set the tone with clinical penalty, wolves barely threatened. So this is something we can expand into a match report, and we've got that for each of the matches down here. So quite a lot of data potentially as it builds up throughout the season. So let's head back over to Claude and see how we're going to use. The first thing we're going to do with Claude is run a little bit of analysis on the documents. For that, we are first going to upload them, so we click AD Files, and then we're going to go and get our documents, open them in Claude, and you'll see them more open here. And then we'll type in our prompt. And our prompt says, across all the uploaded match reports and stats, identify one, the three most significant turning points of the season. Two, which teams form changed most dramatically and why. And three, the single most surprising statistical finding from the season. So we'll First of all, run a little bit of analysis. So here we are, we've got the output. And I can see down here, it says, You hit your session limit, resets at 10:00 P.M. And I can't believe I've hit that already because I just topped up and it says, keep working, which inevitably means spend more money. But before I do that, let's analyze what's going on here. So we analyze the data, and it says, solid data to work with. Here is the full analysis across all three files. So, number one, the three most significant turning points, and match day four Man City away one, one, and it goes into some description. Match Day 23, which is Manchester City at home and Match day 28, Liverpool away, and it gives overview of each of those matches. And then it says the team whose form changed dramatically was Southampton, but the direction is consistent, devastation, not recovery. So it gives us some analysis of that the home matches, the expected goals, and what was actually scored on those days. And then the third result that it gave back was the single most surprising statistical finding, and it said, Arsenal are 261, five drawn, zero losses, and every draw came against a top half side. Yet their expected goals differential in those five draws is plus 1.6. So it's given us the most surprising statistical findings for football fans. And so now that we've gone through this and we've done some analysis, we can do a little bit more. So let's do that. But before I do this, I need to take notice of the fact that it says I've hit my session limit. So let's top up a little bit, by extra usage. And in here, if I go to other, I can top up as much as I want or little as I want. And so I'm just going to top up ten pounds. It's added on VA which is our tax in England, and I'm going to say pay now. And it says purchase successful here. It also allows you to turn on auto reload, but I don't want to do that because I'm going to make a decision if I want to pay more. So I'll I'll click the Cross on that. And the way to know that my top up has worked is if we go down here and then click on settings and then click on usage and scroll down to the bottom, it says Extra usage, which is on, and extra usage is to turn on extra usage to keep using Claude if you hit a limit. So I've only spent seven pence of it, but my spend limit has not been used, and I've put that's 2000 pounds for now, even though I have no intention of using that much. And my current balance is 29 92, so I've got more than enough to go. So if we now go back to our chat, and now we're going to paste in our next prompt. And this says, using all the uploaded material, right, the 1,500 word, season review for our site, structure, introduction, top story, data story, three key performances, season verdict, and the tone should be confident and editorial, no padding. Every paragraph should earn its place. So now we're using even more even more from Claude here. And that's all done using the fact that we've uploaded all those documents, and I've paid for extra processing power and to go past my session limit. So let's go. So here we go. This has been generated, and you can see on the right hand side, it's created this colorful document for us. And just to recap, we said using everything uploaded to write 1,500 words season review for our site with a structure of introduction. So in the introduction, after the introduction, there's top story, data story, three key performances, season verdict, and we set the tone. So here, if we drag this out a little bit, we can see this is what it's done season review for the Premier League and the real deal through 31 match days, Arsenal have built a record that doesn't just invite the title. It demands it. The day says, This isn't Fortune, it's architecture. And then it's gone on to put some of the stats here in red. And we've got some titles under the introduction, saying Arsenal have not lost the Premier League match this season across 31 games against Liverpool at Anfield, against City twice against every mid table side that set up to suffocate them. And then it goes on to go over some of the matches. It goes over their expected goals and how that's stacked up throughout the year, throughout the season. Then it's gone on to give us some of the performances that define the season and about some of the players performances specifically. And then the season verdict here got a nice title and then some information about the season. Then a block quote. The question was never whether Arsenal were good enough. The data answered that by February. The question now is simpler. Can they maintain it for seven more games? So that's the season review. Pretty cool. I like the fact that it's given it some color as well. And the fact that there's, you know, so many words and so much data we've uploaded is all about the context window, the amount that we can do within one chat within Claude, which at this time is far longer than a lot of other LLMs. So in summary, in Claude, the 200 and K Context Window is available on all pay plans and no settings are needed. To use it, you can upload entire datasets and document libraries, not just individual files. And then you can ask cross document questions that would take hours normally to answer because of the amount of data you'd have to go through. But with Claude, you can ask it and get the responses within seconds. Now it's your turn. What I want you to do is find the largest document or dataset that you work with regularly in your business, on your projects, or whatever it is you're doing. Upload it in fall to Claude and ask Claude a question that requires reading the whole document or multiple documents. And quickly, you'll see how powerful it is to take advantage of large documents with Claude's Long Context Window. So I have fun doing that, and I will see you in the next lesson. 17. How To Analyse Images Using Vision: Today, we're going to get into how to analyze images using vision. Now, Claude can understand images, not just read the text in them, but it can also interpret what's happening visually. A football site like what we're creating, this opens up really practical use cases, things like a tactical diagram, commentary and infographic or competitor audits or audits of competitor sites. So I'm going to go into more detail about that and let's get into it. So the feature we're using today is called image is to understand and evaluate images, diagrams, and screenshots. What it solves is manual visual slow image audits or missed issues before we publish. So we often have to have a look at images and verify that it has what we think it has and verify the quality, and Claude can help us by doing that for us. So here we are at Claude, and I'm actually going to show you the images that I've created using Claude, and we're going to talk about what they are and how we're going to use them. The first one is this tactical formation diagram for a football match. What it shows is the formation of the teams for this particular match. And in this case, this is for the same team that we've been talking about all the way through. So this is the tactical formation diagram for Arsenal, and they've got the 433 defense, 433 in terms of the player setup. It's the 433 defensive shape on a football pitch, and it shows player positions for both teams. And in yellow, what we have here is what we've got the yellow press trigger arrows from the arsenal forwards towards the ball carrier. And we've got orange arrows, blocking passing lanes and annotations calling out the pressing shape and vulnerability behind the right back. In our next image, we've got the forecast number of goals, so the expected goals versus the actual number of goals. This is something we actually saw earlier. So if I drag this out a little bit. And this is for the top eight Premier League teams, Arsenal, Liverpool, et cetera. And it deliberately includes one data inconsistency for spurs showing more goals than expected goals. And we can see that one here, and it's in a way that looks suspicious. And one clear underperformance story, which is Brighton, which is this one here, as you can see, they were way under the expected number of goals here. So this gives Claude something genuine that it can flag when we do our quality assurance prompt to check the quality of this image. And in the next image, we've got a competitor sites screenshot. So this is a mock of a screenshot of a competitor's website, and it's a realistic looking rival football stat site or rather an article page from one. And it has the full layout. It's got the navigation, the article header for the article here. It's got the stat line within the article. You can see it saying Arsenal strengthen their grip on second place with a 21 victory over Chelsea, and it goes on with stats, match stats in it's got a mid article promotion banner here, like sites usually have when they're promoting something. And essentially, this is the mockup of a rival website. So let's go back to our Claude chat, and then we're going to work out how we're going to use these three bits of information, which I've already downloaded, all three of these images. So here we are. We've started a new chat. So first of all, let's upload our tactical formation diagram. Here it is tactical formation diagram, upload. And we're going to paste in a prompt, which says, describe the defensive shape in this image, identify the pressing triggers shown and suggest one vulnerability a counter attacking team would exploit. So for those that watch football, what it's going to do is it's actually going to analyze this diagram and then give us some feedback on it. So let's go. So exactly as we said, it said that Arsenal's defensive shape is Arsenal sit in a compact 43, three position centrally around the halfway line. The three CMs form a tight horizontal unit deliberately narrowing central lanes and forcing play wide. So I know for some of you not really too into football, I won't go into too deep, but essentially is breaking down the defensive shape based on that diagram. And it also goes into the vulnerabilities here. It's just explaining the weaknesses in this team from looking just at this image, which is something usually someone really deeply into football would be needed to understand. But as you can see, Claude, with all its knowledge, is able to break that down. So let's upload our next image. So we're going to upload our stats infographic. But before I do, I actually went in and corrected the fact that these names are visible. Previously, this title this little area was overlaying it, but now we can see all the names clearly. So going to download this and then we'll go back to Claude. So I'm going to upload the stats infographic, which I corrected earlier so that the team names are visible, and we're going to paste in our next prompt. And this says, Check this infographic for data accuracy, flag any numbers that look inconsistent. Readability issues, elements that could confuse a casual reader, whether it matches a clean editorial aesthetic. Let's go. So here we go. This is said the chart has fundamental legend The orange bars are labeled actual goals in the legend, but the numbers tell the opposite story. Liverpool's orange bar reads 65.8, which is clearly an expected goals figure, not a goals tally. The bars in the legend carry the lower realistic goal totals. The legend is inverted, fix this before publication. Or the entire chart misleads. So let's have a look at that now. So when I look at the document, I can see there's been a problem I didn't spot, which is that it's cut off half of the legend, which tells it what to do. So Claude is quite right in saying that that wasn't an intended problem. Actually, what we want to do is we want to correct this so that both expected goals and actual goals can be seen clearly here. And then when Claude reads it, it will know that these are the expected and these are the actuals. So let's go back and do that. So at this point, I've corrected the image, and we're going to upload it again and run our prompt again. So let's upload. We're going to get our stats infographic, open it in there. And just to show you. Now we've got the stats infographic. We've got the legend at the top. It's very clear. All of the text is clear. All the team names are visible. So now there's no reason why we shouldn't be able to read the data we need and run some analysis on it. Now let's go back to Claude, and we're going to back in the exact same prompt and run it again. So we've asked it for the data accuracy and to flag any numbers that look inconsistent, readability issues, elements that could confuse a casual reader, and whether it matches a clean editorial asthetic. So what it says now is, if we zoom in a little bit, it says, audit version two. It says, The legend is now fixed, so the legend is clear, as I've said. It's now correct and correctly placed orange for expected goals, dark for actual goals. The call confusion from version one is resolved. Excellent. A accuracy. It says, the numbers are internally consistent across all eight teams. Every expected goal figure exceeds actual goals, except spurs, as we've said, and Liverpool, which is 65.8 expected goals, 61 actual goals. So in spurs, there were actually more goals than expected. And in Liverpool, there were less goals than expected still underperforming. But the gap is the smallest in the set and worth noting editorially. There are some other things here. Readability issues, it says, The insight box now does real editorial work, which is good, but it contains three separate tones, bold headline, neutral observation, then a red editorial note flagging a non issue. That inconsistency undermines the box. Strip the red line entirely. So if we actually open up the diagram, it's talking about this insight box down here, which could be improved. And it says the bar contrast remains a concern on the background. The navy bars will still struggle. So if we look over here and open it up a little bit, the backgrounds actually light now, so I'm not sure specifically what it's talking about here. Maybe it's saying that if we do put this on a dark background, it will struggle, but I'm not entirely sure with that one. Casual reader confusion, it says, mostly resolved. The legend placement top right is clean and immediately readable. One residual issue, a reader unfamiliar with expected goals may not understand why BrightonsTler is a bad thing. A one line explainer in the subtitle or footnote, expected goals means chance, quality, a higher expected goals than goals indicates underperformance would close that gap without patronizing the core audience. So for Brightons it's speaking about this one, and they're saying people might interpret this as a good thing, but maybe we should put some notes under there to clear that up. So editorial aesthetic, cleaner than version one, the provisional data caveat in orange at the bottom is appropriately flagged, but still small. Consider bumping the front weight rather than just the color. We look down here, it's talking about this down here where you can see it says, No, expected goal figures for Match day 31 provisional, Opta final updating pending. So we're saying we're going to get the data from Opta who provide good data for this kind of thing. But it's provisional for now. That's what it's talking about there. And it says remaining fixes before publishing, remove the Spurs red footnote, add a brief expected goals explainer for Context, increase contrast on bars for mobile. So it's talking about this footnote down here, removing that and then making the other changes. That's its analysis of the image, and it's done that using vision. So now let's upload our third image. And this is a screenshot of a fictional competitor's website, and then we'll paste in our prompt. And our prompt says, describe how this page presents data visually, what's working, what's cluttered, and what could we do better on our own site. So we run that. So I've brought up the image of this fictional competitors webpage, and let's go over what it says. So competitor analysis for the stat zone Match report. How is data presented? Stats are consolidated into a single match stats block. Dense inline text format that lists possession, shot expected goals, individual player expected goals, et cetera, and it goes through everything it can see there. So if we zoom in a little bit here, we can see that as it said, there is indeed a match stats block, and it can see everything that's on here expected goals and all the things that's mentioned. And it says, What's working? The stat block is efficient. Everything critical is above the fold, which suits mobile readers scanning quickly. The expected goals figures appear in the lead paragraph rather than being buried. That's the right instinct for a stats literate audience. Had a little issue here on Claude T. I've re run this, and what it's saying is Stat zone leads with narrative, drops a Sat block mid article, then returns to the match stats box is possession shots, expected goals, individual player expected goals. So again, it's speaking about this area. What's working? The Breadcrumb label, Match report, Premier League, and then the date is clean and immediately orients the reader. So talking about it's talking about this area at the top here where it says the Match Report Premier League, and then the date. So it says that's clean. It likes that. The Byline and timestamp are prominent. The expected gold figures appear in the opening paragraph rather than buried good editorial instincts. So it's talking about here where it says the expected gold figures are these. So it likes that. What's cluttered? The stat block is doing too much in too little space. Three lines of plain text is hard to scan. So that's here. There are three lines, and it has got pipes separated text here. So so far, I agree, I think, you know, I really don't like this site at all. It really works for being able to analyze it and testing that Claude can analyze it. And then it goes into more detail about various elements, which I won't go into, but you can see that it's basically picking out bits of information, and Claude can see what's in this graphic. That's the main thing. But what we've demonstrated here is that Claude can basically look at quite a complex image, pick out different areas of this website, which let's face it, isn't a very impressive website and pick out things that it likes, things that it doesn't like, and give us some feedback on it. So in summary, if you do want to analyze any images using image vision, you will go into Claude, click the plus sign, add files to upload any image, and then vision will activate automatically. What you use it for, we use it for tactical diagram commentary, infographic quality assessment, and also competitor analysis. That's what we used it for. But basically, anytime you want Claude to have a look at an image and give you some feedback on that image using analysis, you can use this technique. And you can always upload multiple images in one message for a comparative analysis, and that's exactly what we did here. Now it's your turn. What I want you to do is upload any visual you've published recently, a stat graphic, a screenshot, a diagram, an image, and ask what looks off, what could confuse a reader, and whether it matches your editorial standards, or you could just ask it to pick out specific, you know, opinions and things that it thinks about that image that are going to help you with your project or whatever it is you're doing. So have fun doing that, and I will see you in the next lesson. 18. How To Build Interactive Outputs Using Artifacts: So in this lesson, we're going to go over how to build interactive outputs using Artifacts. Artifacts open a live workspace alongside your conversation where Claude places code documents, data visualizations or interactive components that you can edit and iterate in real time. For a football stat site, this is where Claude stops being a writing assistant and actually starts being a building partner for all of the good work that we've put in so far. It's now time that we can actually build something. So let's get into it. The feature we're using is called Artifacts. Why we use it. We use it for live editable outputs in a separate workspace, and that's things like apps, components, dashboards, charts. What it solves is having to copy output into other tools and then losing a iteration history and rebuilding from scratch. So usually some tools, what you would do is you'd come up with the ideas, say in one tool, and then you'd go into another tool and actually build it, actually build your app. But within Claude, we can actually start building straightaway. Here we are in Claude. We're in a new chat, and we're going to paste in our prompt. And if we zoom in a little bit here, we said, create an interactive HTML widget for our football site. It should display the current Premier League top six standings as a clean table, include the position, team name, played, wins, draws and losses, goal difference, and points. Use a dark background with clean typography, no external dependencies. So let's go. Let's run it. So here we go. It's done what it needed to do. It did warn us that the live standings data doesn't include played goal difference or goals. So it said it will calculate played wins, draws and losses, and flag goal difference is unavailable, pulling in the real points as records. So this is what we've ended up with. This is our table. And what you can do from here is you can actually save it as artifact. You can see below, it says, Live data pulled directly from the standings feed. So this is the table, and what we can do is we can either copy to clipboard, download it or save it as an artifact. If we save it as an artifact, we can see that Claude starts to work its magic. That will save it as a standalone HTML. And therefore, that's an example of an artifact. Here we go. It's now opened up on the right hand side, and it saved this table as a standalone artifact. That means that we can use it in other things as HTML. We can download it, and it always makes it visible here so we can see what it's done. So that's an example of creating an artifact, and the beauty of it it's done this all in HTML, usually something we would have had to get someone who knows HTML or, you know, a web developer to create. Now we can create these things ourselves in a few seconds. So now let's update this slightly, and let's paste in our next prom. So we've said, update the widget, add a column for last five form using colored dots, green equals win, yellow equals draw, and red equals loss. Make the table header, add a subtle hover effect on each row. So by making the table header sticky, that means if we scroll up and down, a table header stays in place, which is handy. So let's go let's run that and see what happens. So here we go, it's doing its thing, and it's created another table. And if you are wondering, the last five form is for each team the performance results and momentum over the past five matches. So here we can see it colored in green, amber or red, depending on how well it's done in those different areas. So in this new table, it says Form Dots, and that's for game week 32. The results are live data, and it's speaking about the last matches that it's had Arsenal's had. I lost to Bournemouth, City Beach, Chelsea, Liverpool Beat United lost to Leeds, Villa DrueForest and Chelsea lost to City. So that is what happened in game week 32. And the preceding four matches are reconstructed from season records. It's got a sticky head up, so it locks to the top of the scroll container, so the column labels stay visible as you scroll. We can't exactly see that here because it's put it all on this side, and the hover effect. So the rows Brighton on hover, the position number and team name pop up. A full contrast, and the form dots scale up slightly for a tactile feel. So when you hover over it, you can see there's a slight change in the color there, so you can see that we're hovering over them, and you can see it a lot more on this side, as well. So asking if we want it to save the current version over the previous artifact. Well, we can do it again as we did it before and just say save as artifact, and that will do its thing and create another HTML widget that we can use so that way we can download it if we want to and use it time and time again. So that takes a little while. There it is our top six standings. At the moment, we can't see the fact that this is sticky because if we had more data in here, it would, I presume, scroll off the bottom, and then we'd be able to scroll up and down. But everything works as before, when we hover over it, the color changes slightly. And this is our top six standing table with everything that we would usually see in there from the gold difference, which we're going to need to populate from different feed. But for now, what this shows is that we can create these clawed with a simple prompt. So now let's create another one. So our prompt says, Now create a separate artifact, a match preview card component. Show two team badges, placeholder circles, head to head record, and keys side by side. Same dark aesthetic as the standings widget. So let's go on and create a match preview card. Go for it. So here we go. Here's the Match preview card. It says, I've used City versus Arsenal on 19 April, since that's the next live top six fixture in the feed, feels more useful than a generic placeholder, so it's taken some initiative there. And it says the card has three sections. The matchup head are with color coded badge circles, blue for home, red for away. So that's these here. The head to head record across the last ten and six key stats with mirrored bar charts showing the split between the sides. The proportionate bars make asymmetries readable at a glance without needing to pass numbers and swap out the data object in the script block to repurpose this for any fixture. So it's talking about what you can see down here, and essentially we can change the data behind the scenes when we download this. We'll see that there's a script block in there and we can change the data object in there. So in order to demonstrate that, we can just download it. It's already artifact. And here it is. The downloaded extra mail. And I'm going to open that with BB Edit, which I downloaded, which is good for editing the source code. And if we have a look within here, we can see just as it said there is a script tag, and you can see there are stats in here, and it's actually filled out all of the information so we can see the expected goals per game, the possession shots on goal, goal scored, clean sheets, press intensity, all these football related stats. And we can just update them in here. If we wanted to, we could pull them from a different feed. But what it's done basically is it's created the artifact for us and it's given us the ability to really quickly what information we need in here and that will then change for our panel. So let's head back to Claude. So now we've got our match preview card, which we can use for our games for any Premier League game, and it shows the ability to just create an artifact right here within Claude. So in summary, Artifacts are run by default. Just check settings capabilities if it's not showing. The artifact panel opens automatically on the right side of the screen as you saw when we save it as an artifact. And you can iterate on the same artifact across multiple messages if you want to make any changes or updates, and then download or copy the output. Using the icons inside the panel, so that way you can change anything in the HTML file, and then you can change what you see on screen. So now it's your turn. What I want you to do is ask Claude to build one simple HTML component for your site or for your project. It could be a stats table or a match card like we built, or it could be something else specific to the project that you decided to take on. And I want you to iterate on it twice. Feel the difference from copy and pasting into a text editor, decide what you want to change in it and just tell Claude what you want to do by writing a prompt. So there you go. Have fun. 19. How To Organise Work Using Projects: So in this lesson, we're going to talk about how to organize work using projects. Up to now, you've been working with Claude one conversation at a time, and that's fine for quick task, but it breaks down when work spans days or weeks or you're juggling research, data analysis, article drafts, and site builds. So what projects do is they let you group all of this together. So let's get into it. So the feature we're using is called projects, and why we use it is to group related chats, files, and instructions under one persistent workspace. What it solves is the problem where you have scattered work, lost context, and you have to keep re explaining things about your site or about your project every single session. Our projects are free for personal use, and there's something called shared projects where your team collaborates in the same workspace, and those are the things that require team or enterprise plan. But what we're going to use is just normal projects. So here we are back inside Claude. And if we look at the left hand side menu and we actually open a sidebar, you can see there are a number of things here. We've got chats, which we've been using, and here we've also got projects. So if we just click on that, you see, I've got a couple of projects already, but let's create a new one, and then we can use that to organize our chats and conversations. So if we click New Project, there is and we can paste in here the name that I created earlier. It's called football site, 2025, 2026, Build. And there are some more information that we can add in about what we're trying to achieve here. But for now, I'm just going to create Project. And so now here is the project. It's newly created. And if we click here in Projects, we can see it's been added to our list of projects. But if we click into it, we can see that projects have memory. And here it says, Project memory will show here after a few chats, but we haven't chatted inside this project yet. And we've also got instructions that we can add, and it says here, add instructions to tailor Claude's responses. So this is similar to the custom instructions and preferences that we added in settings. But this is going to be specific to this football site project. Also, we can add files. So this is any files that relate to the project, which means that now any chats within this project will be able to reference these files. But before we touch any of this stuff, the first thing I'm going to do is decide on some chats that we've already created that I want to add to this project. And the way we do that is, for example, on the left, you can see a bunch of recent chats that we which you can also find by going to chats and seeing them here. And what you can do is you can click on each one and add to project. So Premier League is clearly to do with the football site, so we can click on that and say Add to Project, and then we choose the project, and you can see it's moved it, and there we go. And down here, you can see it's been added to the football site, 2025 26 build project. Now, obviously, selecting each 11 by one is going to take quite a while. So what you can do is you can select the ones that you want simply by clicking Select and then deciding which of these chats should go into that project. So I'm going to select all the ones that are related, this one, that one, and keep going until I've selected them all. So there you go. I've selected all of the conversations, chats that I think are going to be relevant to the football site project, and now I'm going to add them. So here it says, move 15 chats to a project, so click this button and then select the project you want to go to. And now it's going to move all of them to that project. So now you can see a number of these chats have been labeled in here. You can see it says football site 2025 26 Build, so they're all part of that project. So now if we go over to projects on the left and within here, click on the project, or even we can look here and see what's available so you can as a favorite, or you can edit the details of the project, archive or but in this case, we're just going to click into it, and you'll see that all of these conversations are now part of the project. But if we look over to the right where it says files, we can see that the files that we used here aren't part of the project, so we'll have to add those manually. So if you click up here, click the plus sign and then click Upload from Device, then what I'm going to do is add all of the resources that we were using from my chats, add them. Just added those three. Now we'll go back, upload from device, add the next. It will add those three there they are. And then finally, we're going to go back upload from device. Those two actually wasn't final because I think there's one more upload from device and add that one in. And so now we've got all the files that we used in this project. And what that means is we don't have to keep reloading them. Once we've created a chat within this project, it's going to use these files as reference, and we can talk about them, and Claude will know exactly what we're referring to saves us having reupload. The other thing we can do is we can add in instructions. So if you remember, we had some instructions that we added within settings. Now, those will apply to all of these chats, but now that we've created a project, we can make it. So those custom instructions only apply to this project. And that means if we've got other projects or other things, we may not want the same custom instructions. We may not want the same preferences to apply to them. So what I'm going to do is I'm going to take the Custom Instructions out of here and I'm going to add them to this project. So let's go back to settings. And here you can see it says what personal preferences should consider in responses. And it says, Your preferences will apply to all conversations. So we don't want it to apply to all conversation. So what we're going to do, we're going to take this thing which says you are an editorial assistant for a football soccer news stats website. We're going to take all of that stuff out, cut that out, save the changes, and then we're going to go back to our projects, pick this one, and then we're going to go to instructions. And in here, it says, set project instructions, provide Claude with relevant instructions and information that's within the football 2025 26 build. This will work alongside user preferences and the selected in the chat. So we are going to paste that into here and save it. And so now, whenever we do anything within a chat within a football site, only then will it use these Custom Instructions. So to test that out, we can actually reference this image. So I can say, copy this and create a new chat. And in here, I can say, pasting this title. I can ask Claude what it can see in that graphic. So I've said, What can you see in this graphic summarized in five bullets. There you go, saying, looking at the expected goals versus goals infographic premier league 25 to 26, top eight teams match day, Arsenal are underperforming, Liverpool standout over performers, bright and other biggest underperformers. Spurs are scoring above their expected goals. Data is provisional. The chart notes, expected goals, figures for match day 31 are pending. So it's able to see that, and that is within this project. You can see up here, it's within our project. So that's why it's able to see that. So that's great. So now we've got a project. We've got custom instructions related specifically to this project that don't interfere with any other project. That's excellent stuff. Now I feel a lot more like we do stuff in a more organized fashion, and I know where to go if I want to find all of the conversations related to the project I'm working on. As well as that, I've also got the files related to that project that I can not only look at and organize, but I can reference within the chats. Good stuff. So in summary, you can create a project from the left side bar within projects, the new project. You can add Custom Instructions once so that all of the chats in that project inherit them, and you can upload reference files via add content available in every single conversation, and you can add that to the project. Then after that, you can organize your chats by workstream, not by date. So in other words, you can add new chats for particular bits of work you're doing, and they can be added to each project, so they can reference all of the files, all of the memory and all of the instructions that you've added to that project. Now it's your turn. What I want you to do is create one project for your football site or for your website, for your project, whatever you happen to be doing right now. And I want you to add your Custom Instructions and one reference file and start with one single chat inside of there, or you can add existing chats if you have them and organize it the way I've shown you so that you can feel a lot more organized working on specific projects. So have fun doing that, and I will see you in the next 20. How To Create Reusable Workflows Using Skills: This lesson, we're going to talk about how to create reusable workflows using skills. Skills are Claude's most powerful standardization feature. A skill is basically a reusable instruction bundle. So it's a package set of rules, procedures, and contexts that Claude loads automatically for a specific type of task. You can build it once and you can use it every time repeatedly. And as of late 2025, skills became an open standard. The same skill you build for Claude can also run in other compatible AI tools. So let's get into it so you can see how to use skills within Claude. So the feature we're using is called skills. Why we use it is it's reusable task specific instruction bundles based on the open agent skills standards. So basically, you can reuse the same instructions time and time again to do the same task over and over again. And what it solves is prompting from scratch, inconsistent output quality and standards locked in individual chats. You basically create once and run over and over again. The important thing here is that skills require a paid plan to run. So Pro Max team enterprise at this time are all different plans. I'm running the P plan. And a skill is inside of what's called a skill dot md file, which is basically a text file written in Markdown language. Hence it's called MD, and it's a quick way to describe what the skill should actually do. You can create it and upload it to your project. And to install a skill, you'll just go into your project and then click to add the content and upload your skill MD file. And Claude detects and applies the skill automatically when a task matches its description. And if you're using Claude code, then place the skillMD files in your project directory, and Claude code will read them at startup. You can also browse community skills by visiting claude.com slash GIE at this time or searching Github for Claude agent skills. But I'm going to show you how to use skills within Claude. So here we are back within our Claude project for the football site. Before we do anything with skills, let's have a look at the skill MD file for what we're about to use. Here's the Skillm defile. So essentially it's see what's called Markdown, and it's just made up of hashes and hyphens, and then the instructions and the various levels. So it's kind of like when you're marking up titles and subtitles, it's done in the same way. So you can see one hash for the main heading, which describes what kind of skill it is, and then two hashes for the next level down, which is the purpose, and all of the things with two hashes are at the same level below the title. It's got a description. So the match report, use this skill when writing match reports for our football site. So what this is going to do is it's going to make it really easy when we want to write a match report, we're just going to use this skill. It's called the Match report skill, and the purpose is to generate a complete publish ready match report for our football news and stats website. We can apply this skill automatically whenever a match report is requested, so that's going to obviously save us time. And the structure is every match report must follow this exact structure in order. You must have an opening, a tactical story, the standout performer. So that's a particular player who is the standout performer in that match, some context and a verdict. And then the rules are it must be a max of 300 words, no cliches. Every paragraph earns its place. If a stat is cited, it must be specific. Do not open with the scoreline in the report. And do not use passive voice in the verdict, so it should be a confident voice. And again, the tone is editorial, confident, data led. Assume the reader watch the game, do not overexplain. The output form actually be plain pros, no bullet points, no subheadings inside the report itself. Deliver the report ready to paste into a CMS. That's a content management system. Example trigger would be Match report, Arsenal 21, Chelsea on seventh of April, 2026. So that's an example where you would trigger it. So you would trigger it with the words match report, then the team one and the scores, the other team and then the date. Claude detects this pattern, loads the skill, and delivers a structured report without further prompting. So here we are back at Claude, and the first thing we want to do is we want to add the SkilMD file within here. So we're going to click AD files. We are going to add it as a file as we have the file already. So let's go and locate that file. There it is. So we open a SkilMD. You'll see it all loaded right here and there it is. And so now we can simply go to a chat and say Match report, we'll give it the score and the date and run it. So it says reading the Match report skill, so it knows it's there, which is excellent. So here we go. It said that it's picked the match data and the PDF shows Arsenal 21. So Arsenal two, Chelsea one on that exact same date. So I've said it should be Arsenal three, Chelsea one. So it says it doesn't see 31 in some of my other documents, and that the editor's notes and the competitors screenshot corroborate 21, and it says, I'll write the report to the correct scoreline. So basically, it's trying to be a little bit intelligent here about the fact that I've given it so many other documents that have a different score. So it's going to use that score because it was corroborated, which is fine for now. This is just a test. And then it's gone on to write things exactly as we've asked it to write. So if we just recap what the skilled MDFle said, if it said it must have one, two, three, four, five, it must be in that structure. And so if we head back, we can see there's one, two, three, four, five paragraphs there. And it also says a Max 300 words total, no cliches. Every paragraph earns its place. If a stat is cited, it must be specific. Do not open with the score line and do not use passive voice in the verdict. So if we head back over, you can see it didn't open with the score because we already know that. And if we read it, it says, Kay Havas' 67th minute finish ended Chelsea's resistance and with it, any doubt about the outcome. Bokeo Saka had already done the damage from the spot in the 12 minute, but it was HvaunHavrz' run off Odgard the third time he'd made it that broke Chelsea structurally. So if we go back and have a look at the language, no cliches. Every paragraph earns place. If a stat is cited, it must be specific. Do not open with the score line and do not use passive voice. I think it's definitely not using passive voice. It didn't open with the score line. I'm going to have a look at the stats. Every paragraph earning its place, we'll see that on the way back. I can't see any cliches, and it does below 300 words just from memory because we checked this earlier. So heading back over to Claude, the next paragraph, Arsenal's expected goals of 1.73 against Chelsea against Chelsea 0.84 tells the story cleanly. So just looking through it, it is quite confidently toned, and every paragraph does have a purpose. Martin Odgard ran the game five key passes, control of the tempo. Yeah. This is Arsenal's second win over Chelsea this season. So this is our first, well, one of a few stats here, and it does seem to have a purpose for being mirroring the match day eight result at Stanford Bridge. Through 31 games, they've gone head to head with every top six side and come out ahead more often than not. And in the final paragraph, Arsenal make title contention look inevitable. So yeah, I'd say that's pretty much given us what we asked for. So in summary, skills require pro or above as a plan within Claude, and you install via the project and then add content or add file. And then you build one skillMD file per repeatable content type. So, for example, for a match report, you'll have a match report, SkillMD file, et cetera. And Claude applies the skill automatically when the task matches. So when you call for a match report, it will find the skill or the skillMD and then run that. And skills are now an open standard, so they don't just work in Claude. They work across multiple AI tools. So now it's your turn. What I want you to do is write a basic skill for your most common content task. Start with a name, a description, a structure, and then three rules in your skill MD file. And then what I want you to do is upload it to your project and then run it once. And I'm sure you'll see very quickly that it's a massive timesaver. You don't have to write the same prompt over and over. You just call the skill, and it does the work for you. So have fun with that, and I'll see you in the next lesson. 21. How To Connect Claude To External Apps Using Connectors: This lesson, we're going to talk about how to connect Claude to external apps using Connectors. So Connectors link Claude directly to the tools you already use like Google Drive, Gmail, Slack, Notion, Github, Microsoft 365 Canva, and many, many others. Claude can then search, read, and act across all of these apps from a single conversation. So let's get into it. Here we are within Claude. And what we want to do to set up our Connectors is we want to go down here to the initial menu and then we'll go to settings. And then we're going to go to Connectors. And you can see here we can allow Claude to reference other apps and services for more context. So we can connect to any of the apps we want, but I'm going to connect to GMO and Google Drive. So first of all, we'll go to Google Drive. Connect, and we're going to choose that we're going to connect to Google Drive. So it says Connected to Google Drive there. And if I refresh this page, it actually says that Connectors have moved to customize. So let's go to the new customized page. So now we can go here, go to Google Drive, click Connect. We'll connect via my passion Consulting account. Continue. And we'll say what it's going to be allowed to do. So I'm going to allow it to see and download all Google Drive files and to see Edit create and delete only the specific files used within the app and then continue. So here we can see it's connected, and it says, Connect Google Drive to Claude so it can search through your documents, and it allows you to disconnect and view details here, which shows you that it is connected. Then the other thing I'm going to connect is actually Gmail. So we'll just go through the same process, click Connect, continue, and I'm going to continue with my passion Consulting account, continue again, and I'm going to allow it to view my email messages, manage draft, and send emails. There you go. We're now connected to our Gmail, as well, and it's showing us all of the tools that has access to from getting the email profile to reading my email, searching emails. And it's also got the ability to create a draft email. So now we're fully connected. So now that we're connected, I'm going to head back to my project. And I'm going to start a new chat and add in a prompt. So I'm within the football site project, and we're going to add in this prompt. And the prompt says, Search my Google Drive for any documents related to our Premier League season review, summarize the key editorial decisions we made and suggest how to build on them for next season. So let's run that. It looks like it still doesn't have access to Google Drive, so let's go we're going to go to settings. We're going to go to Connectors. I can see Google Drive is connected, and Gmail is, too. Let's go to customize as this is what it's suggesting we need to go to. Let's check Google Drive is there. It looks like it's connected, according to this. And if we view the details, it looks like there are two tools, Drive search and Drive fetch, it looks like we're connected. Otherwise, the disconnect button wouldn't be there. So I'm going to go back to our chat, and I'm going to go back to our project. And inside our project, I'm going to start a new chat, and I'm going to ask what it has access to. So after looking into this, I said, I can see Google Drive is connected. Why can't you access that? And it basically said that it's not enabled for this chat. It seems to be a new way that things are working because I know things have been moved around lately. But once I've created when I've created the project, it for some reason, had access to Gmail, but not Google Drive. So it told me, Google Drive shows connected, but not enabled in this chat. That's why I can't access it. It needs to be toggled on for this session. So I'll just go to reconnect here and I'm going to connect again, and I'll click Continue. Says it's connected now, but I'm going to go back to the project and check again. Now here it says Google Drive is connected. So I'm going to try again and ask it if I'm connected to Google Drive. So it's saying that no, only Gmail tools loaded. Google Drive is not currently active in this chat. You'd need to enable it via plus button in the chat input bar, then I'll have access. So we'll go down here plus, go to Connectors and switch on Drive search. So I'm going to ask again, Are you connected to Google Drive? So now it seems to found it. I think it was a red herring. It always had drive search. So when I refer to this as Google Drive Search, it found it. So now let's type in our prompt. The prompting is search my Google Drive for any documents related to our Premier League season review, summarize the key editorial decisions we made and suggest how to build on them for next season. So it's searching my Drive, as you can see here. So it said the drive search didn't surface dedicated season review document, only the master class handout, which is something that I've created for this course. And it's shown me stuff here from the project, so it hasn't found anything. And that's correct because I haven't uploaded anything yet. So let's upload a document to my drive, my Google Drive, and then we're going to run this prompt again and test that I can actually find what I've uploaded to Google Drive. So here I am at my Google Drive, and I'm about to upload a file that relates to the football site. So if I click New File Upload, and then I'm going to go and find the file. And here it is. It's the editorial notes, which have the season review inside of it. So we'll open that. And here we can see it's now on my drive in the football website folder. So let's go back to Claude and run a prompt to search within this drive. So here's my prompt again to search within Google Drive, and let's run it again. So, it turns out that files can only be found if they are Google Docs. So the way to do that is to click on this file and Tune open with then Google Docs. So now this is now a Google Doc. And you can see it's opened another file here, which is a Google Doc, and this text one was not searchable. So if I now go back, and I'm going to say search for my Google Drive for any documents related to the Premier League season again. So that's come back, and it says, found it. The file just became visible. Let me fetch the full content. Got it. Full document retrieve from your drive. Here's a summary. And then it tells me of key editorial decisions from the drive document, the season narrative Arsenal as the title team, not contenders, five player tactical stories committed to, and three turning points identified. And if we just look quickly here, we can see season narrative which matches this. And then the five players, which are the key stories, which matches this. Then the three turning points are these. And it says, Verification first publishing discipline. Several stats were flagged to confirm before publishing. That's these, which are the stats to check, and a dual long form strategy. The article package includes writing both how Arsenal won the title race and how Arsenal blew the title race. So that'll be what's mentioned here. So now I'm confident that from within Claude, we can see documents on my Google Drive. Good stuff. Next, we're going to test the capability to see inside of my GML, so I'll paste this in. And this prompt says, Search my GML for any press releases or data reports from football data providers in the last seven days, summarize what's new and flag anything relevant to our editorial calendar. So let's do it. So this is saying nothing on that search. It's tried broader tons, but it couldn't find anything. That's absolutely fine. So what I'm going to do now is I'm going to send an email to myself on GML, which actually has some football data in it, and I'm going to run this search again. So let's do that. So here we go. I've created an email which is saying football match data reports from data providers, and inside, it says, PDF attach with all match reports, and I've attached this PDF here. So I'm just going to send that to myself and that has been sent. So if I look in my Sent items, I can see that the email is here in my Sent items. So now let's go back to Claude and see if we can find it. So I'm going to copy and run this prompt again. So this time, it says, It's found it. There's one directly relevant email. Let me read the full content. So it says, Good. One relevant email found sent today 14th, fourth, subject football match data reports from data providers with a PDF attachment called All Match Reports PDF. The body just says, PDF attached with all Match reports. And the good news is that the PDF is already in your project files. It's the A Match reports PDF, so it's managed to match what it found in the email to what's actually already in the project. And if I go down, it says the flags for your editorial calendar. The Match Reports PDF covers through Match day 31 on this date, you're up to date. Match day 32 data won't be in here, and it gives me some more background about the fact that there's no communications from Opta or any of these sites. So it's definitely found a PDF. It's found it in my email, and it knows it's the same document that is already in my project. That's great news. This is working great. And it can be used to query any data document in my email from now on, which is going to be really helpful. So in summary, you can enable Connectors by going from settings, Connectors, and then connect. You may need to go to customize, as this feature is changing at the moment, but start by going to settings and Connectors and within their Look for Connect or customize. There are over 38 apps already available, including Google Drive, Gmail, Slack, Github, and Notion, and more. And you use connected apps in any chat or conversation. No reauthorization isn't usually needed. If you hit any blockers, it may ask you to reauthorize in that situation. So, now it's your turn. What I want you to do is connect Claude to one app that you use daily. Start with Google Drive or Gmail. Nice, easy ones, and then ask it a question that requires searching that tool. And that way, you'll know that whenever you're ready, you can query emails or documents in your Google Drive or your Gmail or any app that you ask it about. So there you go. Have fun with that, and I'll see you in the next lesson. 22. How To Use Claude In Your Browser (Chrome Extension): Lesson, we're going to use Claude in your browser using the Chrome Extension. So Claude Chrome Extension brings Claude directly into your browser. You can trigger it on any webpage, ask Claude questions about the content you're looking at, fill in forms, navigate sites, and complete tasks all from inside Chrome. So let's get into it. So the feature we're using is called browser use within Chrome. Why we use it is we use Claude inside any website to read page content, fill forms, navigate execute voice instructions. What it solves, it solves context switching between browser and Claude chat, and we don't need to copy content between tabs manually. We can get Claude to do everything. We can even get Claude to carry out tasks for us all within the browser. It's amazing. So let's see how that works. So here we are, in order to install the browser extension, I'm actually going to create a new tab within Chrome. So up here, I'll create a new tab, and then I'm going to go to this web address Chrome web store@google.com. And then within here, I'm going to search for an extension to do with Claude, and we want the official Claude extension from anthropic. So, to make sure I get that one I've searched for Claude Anthrop so that should bring back anything with Claude D anthropic in the name. And I happen to know that it'll be this one. You can tell us when you click on it, and you look in here, you can see, it's got anthropic, and it's verified, and that's the one we want. So we're going to add that Chrome now by clicking that. It lets you know that it can access the page debugger back in. I can read and change data on websites, display notifications. And basically, it's giving you a warning that this is really powerful. Going to go with it, so add the extension. Then it tells me Claude for Chrome would like to connect to your Claude chat account. Your account will be used to access your anthropic profile, contribute to your Claude subscription usage, and create continue and delete your conversations. So let's authorize that. So that's that. So once you're in, it says Claude in your Chrome. You're in the Beta. So this is Beta a Beta testing version. And you can try a demo if you need to, but we're going to do our own demo so we're all installed. And actually, before we go back to Claude, there are a couple of things we can do. So the extension is up here, and we can click on the three dots and we can go into options. And because we're going to be doing stuff with voice, we can actually allow microphone access, and you can see it can hear my microphone, and I'm going to say, allow this time. So that's already done. And the other thing I'm going to do is I'm going to pin this so that it's at the top of the list, and so I can immediately go up and see that I can access Claude, the extension just by going up here. So that should be everything we need for now. It says that this is a Beta feature and it has unique risks. I'm going to let it stay on my Chrome, so I'm going to say, I understand. I'm going to open a new tab, I'm going to post in a link to the BBC's website, and let's open up a story so we can click on this story here. We can see it's got a Match report, and this is the Match report. So now that we're on this page and we have a Match report, we can actually access Claude, and we can let it perform actions or read what's on this page. So to test it, we'll go up, see our icon, click on that, and we'll see it opens up this panel. So down here, this is the equivalent of using Claude. But against this website. So let's paste in our prompt. And it says, summarize the key data points in this article. What angle did they take that we haven't covered on a stats focused football site. So this is asking about this particular article. Claude should be able to read it and then give us some insights about how we could improve our articles. So let's run that. So down here, we see this asking us for permission to run this action says New permission is required. Claude wants to read page content on bbc.co at UK. I'm going to accept that and allow that action. If we look here, we can see that this is the story, and it's summarized and said the key data points from the article are Spurs have now gone 14 Premier League games without a win. Their last Premier League win was at Crystal Palace on 20th of December, and Igor Tudor was sacked after just 44 days, and it gives some information here. And if we look through here, we can check some of that information. It says, Spurs could not end a winless sequence in a Premier League, which now stretches to 14 games, so that matches up with this. And so that's proof that it's actually reading from here. So now we've got the ability to read webpages. So now let's open another page and see what else we can so now we're at the FB Ref page, which has a whole load of stats, particularly for Arsenal. And if you scroll down, you can see there's lots of standard stats, lots and lots of tables of stats around the matches that Arsenal have played. So what we're going to do is we're going to open up Claude and run some kind of analysis against all of this and show how we can just pick out information from this rather complex looking website for our site. So what we're going to do is again click on the icon. It's going to open up the window, and then we're going to paste it in our prompt. And the prompt here says extract the top five players by expected goals from this page and format them as a clean table with columns, player position, expected goals, goals, expected goal difference. Then write one sentence summarizing the most interesting finding. So let's go. So here Claude's popped up a plan of what it's going to do against this site, says it's going to read the current arsenal stats page, extract the top five players by expected goals from the shooting standard stats table, format the data into a clean table with the columns we've asked for, and then write a one sentence summary of the most interesting finding. So let's approve that plan. And off it we'll go. We'll wait for that to finish, and then we'll come back and have a look. It will take a few seconds. So after running that it said, unfortunately, expected goals data is not available on this FB ref page. After a thorough inspection of all tables and columns on the page, the shooting table only includes standard stats, so that's the goals, the shots, et cetera. So what it's done instead is it's built a table with what it could find. And when I've looked, it's actually true. I can't see, so I can't see expected goals. But what it has done is it has found the players, their position, the number of goals, and the shots and the goals per shot. So it's done that for all of the top players. So it's done pretty much exactly what we've asked. It just didn't have the expected goals result, which is cool, and it shows that it's actually doing the work to work out whether what I've asked for is actually possible or not. So I'm pretty happy with that, and it's put these players in this order in order of goals just like we've asked for. Now, comparing to the table, you can see that Victor here, if we look over has scored 12, so that's definitely accurate. And you can look through and compare all the other players as well and just make sure that those scores are accurate, as well, and they are. So I'm happy that it seems to be doing what it said it would. And remember, this is accessed all from a webpage, and this allows us to pull out the information we want in real time from any live webpages. We can then pull this information into whichever format we want a CSV, a PDF, anything we want, and even put it on the website. So that's great stuff. So let's do one more if we open this page, and then we open up Claude again, our web extension, and let's paste in a new prompt and say, based on the stats on this page, write 150 word editorial intro for our football site about the most surprising number you can find with a confident tone, no cliches, let it be data led and assume the reader, let's go. So, again, it gives us a plan, which I'm happy with, and I'm going to approve the plan, and let's go. Here it says it's comeback and it said, the most surprising number I found is the arsenal save percentage is 68%, meaning David Rea has saved only 51 of 75 shots on target. So we've got a pretty good goalkeeper there, and it's found that to be the most interesting stat that it could find on this page. And then it's gone on to give us just as we asked for 150 word editorial intro for our football site. So this is showing how we can actually pull live data from a real website that we can now put on our website or write a story about. And that's good because it means that we can interface to well known reputable websites for our app. So in summary, you can install from the Chrome web store, which is Chrome webstore.google.com and search for Claude, then find the official anthropic extension addict to Chrome. Then you need to sign in with your Claude AI account, which requires Pro or able for the extension. I was already signed in, and you may already be. And then click the Toolbar icon, which I pinned to the top on any page to activate it, and then you can pin the extension for easy access via the Chrome toolbar puzzle icon. But essentially, once you've clicked the icon to activate it, you can then find information from any page and interact with any webpage, and it makes life a hell of a lot easier for you. Now it's your turn. What I want you to do is install the Chrome Extension and activate it on a competitor football site or in any site that you want. And then ask Claude to summarize the article 0R the information on that page and identify an angle that your site hasn't covered or identify something that your project needs. And that way, you will have much more confidence being able to query information from any site and interact actually with any site. And there's a hell of a lot more you can do. So there you go. Have fun doing that, and I'll see you in the next lesson.