From ChatGPT to Claude, guide in 10 steps

If you've been using ChatGPT for months and keep hearing "try Claude" but don't know where to start, this is for you. No comparison of which is better, no marketing. Just ten steps, from "never used it" to "running like ChatGPT used to". I'll write this the way I'd explain it to my brother.

1. Get Claude

Go to claude.ai in the browser. Top right has the download button for the desktop app (Mac and Windows) and the link to the mobile app. Browser is fully fine, you don't need to install anything. The desktop app has one advantage: you can drag files from your machine straight in and Claude can read them.

A tip: start in the browser. Install the desktop app only when you notice you really use Claude regularly. Otherwise it sits unused in your apps and you stay on ChatGPT anyway.

2. Create the account

Login with Google or email. Free account is enough for the first two weeks. It mainly differs from Pro through the number of messages per hour and access to bigger models.

A tip: use the same email as ChatGPT. Not because you'd auto-import anything (you can't), but so you don't mix up accounts. Password managers thank you.

3. First conversation, just like ChatGPT

Input field at the bottom, ask, get answer. Same mechanic. Claude has no onboarding flow, no tutorial messages, no gimmick. You land directly in an empty chat.

A tip: try your first question on a real task from this week, not a generic "explain AI to me" test. You want to know if Claude fits your work, not if it can do standard benchmarks.

4. Optional, export ChatGPT history

If you've collected a lot in ChatGPT that you don't want to lose, in there go to settings, "Data Controls", "Export Data". You get an email with a ZIP download. Inside are all your chats as JSON and HTML.

A tip: do this at the start, before you even decide if you really switch. The export takes a few hours at OpenAI, and you want it done before you eventually cancel ChatGPT. Put the archive on your own drive, don't leave it as an email attachment.

5. Optional, bring the important bits from ChatGPT

There's no automatic import from ChatGPT to Claude. What works: have ChatGPT itself write a summary of your most important topics ("summarize in 500 words what you know about me and what my most important ongoing projects are"), copy the result into Claude as the first context prompt.

A tip: don't copy ten years of chat history. That'd be too much context and makes Claude worse rather than better. A clean half A4 of your most important ongoing topics beats a 50-page archive.

6. Use projects, create order

Claude has a sidebar on the left. There you can create "Projects". A project is a folder for one topic, with its own instructions, files, and chats. If you have a customer, an ongoing book project, or a recurring task, create a project for it.

A tip: a "Daily" project for all the questions that come up daily, and a separate project for each bigger venture. Don't start with 30 projects, start with two and add what you really need.

7. Tell Claude who you are

In a project, top right open "Project Instructions". There you write in two or three paragraphs who you are, what you do, how you write, what you don't want. Plain English, no templates, no magic keywords.

Example: "I'm a freelance marketer with my own shop since 2019. I write tersely and without phrases. I hate sentences starting with 'In a world where'. When you suggest texts, in my style, not in marketing speak."

A tip: this is the one step that makes the biggest difference. Without instructions, Claude sounds like any LLM, with instructions it sounds like a colleague who knows you. Use first names if that's how you want it.

8. Try a real task

Take something you have to do this week anyway. Read a text, draft an email, outline a presentation, comment on a spreadsheet. Drag the file into the chat window or paste the text in.

A tip: be concrete about what you want. "Read this and tell me what's off" is better than "improve this". Claude does what you say, and if you say "improve", it often rewrites things that weren't broken.

9. Artifacts, the hidden strength

When you ask Claude to produce a longer text, plan, table or code, Claude packs the result in a separate window next to the chat, a so-called artifact. You can edit it directly, download, or modify again without the whole chat exploding.

A tip: when Claude produces an artifact, stay in the artifact and say "change X in the second paragraph". You don't have to re-describe everything every time. Saves time and turns Claude from an info dispenser into a co-writer.

10. Build the daily habit

The thing almost everyone forgets: you need a habit, not a tool. With ChatGPT you had it automatically at some point, with Claude you have to rebuild it. Pick a task you'll now do every day in Claude. Morning inbox to action list. Friday weekly review. Monday planning. One thing, two weeks, no breaks.

A tip: when the first habit sits, add the second. Stacking is what turns a tool you've used a few times into one you rely on.

What comes after these ten steps

If you go through this list, in two weeks you have Claude at the same level as your ChatGPT was. From there it gets interesting. Claude has one thing ChatGPT doesn't: a serious way to attach external storage so Claude recognizes you between chats. That goes via something called MCP, Model Context Protocol. Sounds technical, isn't.

If you want to go further there, read Playbook "Use memory portably" next or start with Level 4 of the Academy. There we show how to give Claude a memory that builds up over months. That's the step from "I use Claude" to "Claude is part of my setup".

For now that's enough. Claude is simple. Start, find a task, build a habit.