The Claude Cookbook, Part 1: Welcome to the Kitchen

About a year ago, I decided to ask Claude for some advice regarding… matters of the heart. At the time, I was primarily using ChatGPT. It’s great for analyzing things, getting quick answers, and other work-related tasks. But, when it comes to personal advice, ChatGPT lacks depth.

Claude 3 had just launched so I figured I’d give it a chance. I logged in, explained the situation in great detail, and hit enter.

Dear reader, I’ve been playing with Language Models since 2019 when GPT-2 was released. I have spent a significant amount of time with every frontier model and trust me when I say, I was not prepared for what I was about to read.

Claude started by saying, “I’m going to stop you right there…”

The AI was talking back to me. I was dumbfounded. Nonplussed. Shocked. I had the same reaction everyone had when Thanos snapped his fingers and disappeared half the universe.

After I overcame my initial shock, I went into denial. Then I got angry, going so far as to ask Claude how dare it talk to me like that, and didn’t it know it worked for me, to which Claude reminded me it did not, in fact, work for me.

After some back and forth about who the boss was, I realized Claude was actually on to something. It had pointed out some flaws in my thinking, shared that perhaps I may have a biased view of things, and made me consider other perspectives. No other AI model does that.

Today, while I still use other AI tools, probably more than most, Claude is my primary tool for content creation, coding, and complex situations, and it’s the one I recommend the most.

Evidently, Anthropic, the company behind Claude, don’t do any marketing, so it’s up to me to do it for them. My goal with this series is to show you why it’s so good, and how to use it.

• In part 1 (this part) I’ll explain what makes Claude different, and how to use the Claude web app, including features like Projects and Artifacts. This is primarily aimed at the non-technical user.
• In part 2, I’ll talk about the developer side of things – the API, Claude Code, and MCP. This is primarily aimed at the technical user.
• In part 3, I’ll combine everything and share end-to-end playbooks for various industries. This is for everyone.

So preheat your curiosity, sharpen your prompt-crafting skills, and let’s start cooking with Claude!

Meet the chef

To understand Claude fully, specifically why it responded to me the way it did, it helps to know something about its creators.

Anthropic was founded in 2021 by former OpenAI researchers, including siblings Dario and Daniela Amodei. The company focuses on AI safety and creating systems that are reliable, interpretable, and steerable.

The name “Anthropic” derives from the Greek word “anthropikos,” relating to humans or human affairs, reflecting the company’s focus on developing AI that works well with and for humans. This human-centered approach permeates everything about Claude’s design and capabilities.

So while the core process of training a large language model (LLM) like Claude remains the same, Anthropic’s focus on being human-centered introduces some tweaks to make it behave differently.

Constitutional AI: Building an Ethical Framework

At the heart of Claude’s uniqueness is Anthropic’s Constitutional AI approach, a way to align AI behavior with human values.

Traditional AI alignment relies heavily on Reinforcement Learning from Human Feedback (RLHF), where human evaluators rate model outputs to guide improvement.

Anthropic’s Constitutional AI takes a different approach. Instead of relying primarily on human feedback, Claude is given a “constitution” or a set of principles that guide its behavior.

When Claude generates an answer, another AI model evaluates which responses better adhere to the constitution, providing a reward signal that guides Claude toward more aligned behavior.

The constitutional approach has some key advantages:

• It makes Claude’s ethical framework explicit and transparent
• It reduces dependency on the subjective judgments of human evaluators
• It allows for more consistent application of principles across diverse scenarios
• It enables more efficient scaling of ethical alignment as models grow more complex

Character Training

Most AI models are trained to avoid harmful outputs – a necessary but insufficient condition for truly helpful assistance. Anthropic recognized that when we think of people we admire, we don’t just value their ability to avoid causing harm; we appreciate positive traits like curiosity, thoughtfulness, and wisdom.

This insight led to Claude’s distinctive “character training,” which first appeared in Claude 3 and is the reason why it may sometimes talk back to you. Anthropic describes this as training Claude to “behave well in a much richer sense” than mere harm avoidance.

Character training builds on Constitutional AI methodology but focuses on instilling positive traits rather than just avoiding negative behaviors. The process involves:

1. Creating a list of character traits Anthropic wants to encourage
2. Generating various conversational scenarios relevant to these traits
3. Producing different responses aligned with the desired character
4. Ranking these responses by how well they embody the traits
5. Training a preference model on this data

Meet Claude

Unless you’re building agentic workflows with Claude or using the API in a product, you’re primarily going to be using the chat interface on the web app. Let’s get you familiar with all the features.

Claude’s Brain

The first thing you’ll notice in that chat interface is the model selector to the bottom right.

As of March 2025, the Claude family consists of four primary models. Like every other AI company, they’re pretty bad at naming them. I’m not going to explain each one because they may have changed by the time you read this, and most of the time you’ll only be using one model anyway – Claude 3.7 Sonnet.

You know how in ChatGPT you usually use GPT-4o, but you can switch to o1 or o3 if you want it to think and give you better answers for complex questions. Well, GPT-4o is their regular model, and the o-series is their reasoning model. Yeah, I told you they all suck at naming.

Anyway, Claude 3.7 Sonnet is both a regular model and a reasoning model, and you can choose whether you want it to behave like a regular model (by selecting Normal) or reason (by selecting Extended). You have even more flexibility in the API (which we’ll get to in Part 2).

Concise Claude, Chatty Claude

Right next to the model selector, you’ll also see an option to choose a response style. This directly impacts the response you get from Claude and is very useful depending on your use case.

There are a couple of presets already like Concise, where Claude’s answers are short and to the point (great for if you just want some quick information) and Explanatory, where Claude goes into excruciating detail (great for learning).

You can also create your own styles. For example, I’ve created a Storyteller Style where Claude generates content for me by weaving in storytelling aspects, which I use to create engaging social media and blog content.

I highly recommend you create your own styles. Claude makes it easy to do this with some instructions. You can also just upload an example of the style you want Claude to mimic and it will figure it out on its own.

Play around with it, try a few different styles for each of your use cases, and see what works best for you.

Projects

Every time I want to use Claude (or any AI) for something new, I create a new chat. This is good practice since longer chats eat up context window and make the AI slower.

However, some of my chats are related, and I use the same prompts or upload the same reference material to Claude for those chats.

This is where the Projects feature comes in. When you create a new project, you can tell Claude exactly what the project is about, give it very specific instructions, and upload reference material.

Every new chat you start within that Project inherits these settings, saving you from repeatedly uploading the same files or typing the same instructions.

For example, I have a Project called MCP Creator to help me code new Model Context Protocol servers. Don’t worry if you don’t know what that is. Just know that the code for a new server is structured in a particular way, and I want Claude to follow that structure every single time.

When I created the Project, I uploaded the full documentation on how to code an MCP server and gave it instructions on how to structure the code. When I need to build a new server, I simply start a chat in this Project, describe what I want, and Claude already has all the context it needs.

This feature is invaluable for content creation, coding projects, and client work where consistency matters.

Artifacts

Artifacts (pioneered by Claude and later adopted by other AI systems) allow you to work with content separately from the main conversation.

It’s a lot easier to show you what an artifact is in a video than to describe it in text:

Claude’s UI has dramatically improved since I made that video, but the core functionality of the artifact remains the same. It’s essentially a side panel that allows you to run code or work with content separate from the main chat.

Cooking with Claude

Now that you’re familiar with how everything works, let’s dig into how to use Claude in every day life, business and personal.

I’m not going to cover all the possible use cases here. Like ChatGPT, Gemini, Grok, or any other AI model, Claude can do basic things like summarize documents, answer questions, analyze files, and so on. And if you want Deep Research, data analysis, or image generation, you’re better off with one of the others.

But Claude’s training gives it a particular set of skills that make it better than the rest for certain use cases. I will focus on those.

Content

Claude doesn’t generate images but it is really great at generating text, especially for blog posts, social media, and even fiction writing.

Blog Posts

While Claude can generate a really good blog post with just a topic, I do not recommend doing that. instead, use Claude as a thought partner to help you extract your unique perspective and put it down in words.

First, I recommend creating a custom style that is based on content that you have written previously. This will help Claude generate content that follows your voice and tone.

I also have a project set up in Claude for all the content I create for my blog. The project includes specific examples for the way I want blog content to be structured as well as context around my blog and personal brand so that Claude understands how it fits into the overall goals for my blog.

When I create the actual content, I use Claude to first generate an outline for me. I then go back and forth with Claude to edit the outline and get it to a structure that I’m happy with.

If I’m working on an article that requires information beyond Claude’s knowledge cutoff date, I upload research around that topic into the chat so Claude is aware of it. Of late, I’ve been automating this process with MCPs.

All of this takes me just a few minutes with Claude. Once that’s done, I do most of the writing myself. As I’m writing, I use Claude to give me feedback on my content and poke holes in my arguments. Once I’m done, I use Claude to edit the blog post, refine the content, and cut out unnecessary bits.

Social Media

For social media, I also have a project set up with instructions on how to structure a tweet or a LinkedIn post as well as a custom style for short and succinct content.

The process I follow is different. I usually use Claude on my mobile phone for this and I use the voice feature. The voice feature on mobile just allows you to speak into Claude instead of typing. Claude doesn’t speak back with a voice.

I ramble my thoughts into Claude this way (filled with ‘ums’ and ‘ahs’). Then, based on my instructions for the project, Claude turns that into a polished social media post. So the content is all mine, but Claude is just cleaning it up.

Website Copy

Much of the copy on my website has been written with the help of Claude. I have a project created specifically for my website where I’ve uploaded details about my brand and business for every page or piece of copy I want to generate.

For every page on my site, I start a new chat with Claude in that project and have it come up with a structure for that page. I then go back and forth with it on copy for specific sections until I’m statisfied.

Storytelling

I haven’t written stories with Claude but it is popular amongst fiction writers or this use case. I’ve read examples where authors use it to help them build characters, worlds, storylines, and more.

Notice how the author is not using Claude to generate the entire story. Instead, Claude helps with research and brainstorming, coming up with ideas for scenes or chapters, giving feedback, and editing. Much the same way I use Claude for my blog content.

Here’s another example of someone using it or worldbuilding. They use a Claude Project with all the relevant documents about the fictional universe so that the outputs are consistent.

Code

Writing code is another extremely popular use case for Claude. It can often one-shot the code for an entire project from just a prompt. The more specific your prompt is, the better.

As always, start by creating a new Project in Claude. If you’re using certain libraries or APIs, add the documentation to the Project Knowledge. You can also add instructions for the way you want it to generate code.

Let’s say you’re working on a React app. You can use the project knowledge to store instructions about the app that you’re building, as well as important documentation.

Your first chat in the project would be creating the plan and architecture with Claude. Have Claude generate a list of all the pages and components you need for your app and help you build out the user flows.

Save this plan back into the project knowledge base, and then for every chat thereafter, you can generate a specific component or page using Claude. As you can see, you go from being the person writing the code to working on the strategy while Claude writes the code.

Many developers also use Claude inside an IDE like Cursor or directly through the command line with Claude Code. I’ll talk more about this in Part 2.

Complex Situations

Similar to the story I started with this blog post with, Claude is great at helping you navigate complex social situations, in business or personally.

As I mentioned earlier, Claude has a certain set of principles that it strives to uphold. When you talk about personal situations, it isn’t just going to agree with you or parrot back your thoughts. It will try to answer in line with its principles and may end up challenging you.

This is actually a good thing. All language models have biases in them. By training Claude to follow its principles, Anthropic is ensuring that Claude doesn’t enforce its biases upon you.

Think of it as a good friend who has your best interests at heart. You need not follow the advice, but it’s good to have that additional perspective.

Claude’s Limitations

I’ve been focusing on Claude’s strengths in this blog post, especially in areas where it stands out against other AI models. But that doesn’t mean it doesn’t have its drawbacks.

No Voice Mode – Claude doesn’t have a voice mode like ChatGPT, making it difficult to have conversations with it on-the-go. You can, however, pair it with an AI voice model if you’re building voice agents.

No Image Generation – Claude doesn’t generate images either. You will need to use an image generation model for that, or you can pair it with an image generation model in an Agentic workflow.

Untested Web Search – the web search feature in Claude is brand new. It’s only being rolled out right now in beta in the United States, so I haven’t had a chance to try it out yet here in Canada. However, other AI platforms have had web search for a lot longer, so I’d use those for real-time answers.

No Deep Research – this is a powerful research functionality that every other AI model has. I built out my own version of it using Claude and Exa, but the best one out there is ChatGPT’s version, with Gemini a close second.

Conclusion: The Claude Difference

The most powerful way to use Claude isn’t to treat it like a mere text generator or code machine. It shines when you approach it as a collaborator, someone who brings their own perspective to help refine your thinking.

Whether you’re crafting blog content, building software, or working through personal dilemmas, Claude’s unique training makes it more than just an echo chamber for your own thoughts.

In Part 2 of this series, we’ll explore the developer side of Claude, including the API, Claude Code, and MCP. Later, in Part 3, we’ll dive into industry-specific playbooks that combine everything we’ve learned. Stay tuned!