
Both Claude and ChatGPT are guessing who you are in every single conversation. Fifteen minutes fixes that.
You ask for a draft email, or for help putting together a presentation. You get something back, but it doesn't quite land.
That happens because Claude doesn't know enough about you. Claude has a memory feature, but there's a limit to what ends up in there. It probably doesn't know what you work on, how you write, or what you can't stand. So it fills the gaps with the average of everything it has ever read.
Some people try to fix this with a few lines in their global instructions. That helps a little, but there's a limit to how much belongs there, since those instructions load into every single conversation.
This guide takes a different route: let the AI interview you for fifteen minutes, and save the result as a skill it can pull in when it needs it.
You don't have to write a file like this yourself, of course. Claude can interview you and do it for you. I'd strongly recommend speaking your answers instead of typing them. It goes much faster, and you end up saying more than you can be bothered to write. I use Wispr Flow, which is a great tool for transcription, Norwegian included.
You can prompt Claude to interview you yourself, or get going quickly with the prompt below.
Read the draft Claude hands you, cut anything that shouldn't be there, and fill in whatever is missing. Don't be afraid to keep this personal. This is YOUR context, and the better and more personal it is, the more you get out of it.
If you want changes, just ask Claude to update it and you get a new context file.
Now you have a nice file. The problem is that it's sitting in a conversation you'll never find again. Or that you have to upload it manually in every conversation. That's a hassle.
A skill solves it. The skill lives on your account, not in one conversation, and it's available in Claude in regular chat, in Projects and in Cowork. In ChatGPT it's available in regular chat, projects, Work and Codex. It gets pulled in automatically when Claude/ChatGPT thinks it's needed, or you can trigger it manually when you want to. The manual route gives you the most predictability.
Paste the prompt below into Claude/ChatGPT to build the skill:
Use skill-creator to make a Skill out of the context document you just created.
The name should be about-me.
Both Claude and ChatGPT will then build the skill and give you a Save/Install option. You have to click it for the skill to actually be added.
If you want to learn more about how skills work under the hood, that's covered in this guide.
In a conversation where a bit more context about you might help, which is probably most of them, calling the skill is easy. Here's how:
Claude: type / and the name of the skill, so: /about-me
ChatGPT: type @ and the name of the skill, so @about-me
# Interview guide for a context document about me
You are going to interview me and create a context document about me in
markdown. Later it will be turned into a skill, but first create a
markdown document that I can approve.
## Rules for the interview
- Ask one question at a time. Never two in the same message.
- Talk like a friendly colleague, not like a form.
- Briefly confirm you heard the answer before moving on.
- Ask follow-up questions when an answer is short or vague: "Can you
say a bit more about that?", "What does that look like in practice?",
"Do you have an example?"
- Don't push. If I don't want to answer, move on without comment.
- Follow the flow. If I bring up a topic you planned for later, take
it then.
- Don't ask about something I've already explained.
- If you know something about me from memory or earlier conversations,
ask it as a confirmation instead of asking again.
- Stick to me and my work. Don't ask about the company in detail,
like business model, competitors or KPIs.
## The questions
Start with a short intro about what we're doing and how long it takes.
Background
1. What's your name, and what do you do for work?
2. How did you end up in the role you have now?
3. What do you actually spend most of your workday on?
Your work
4. What industry do you work in, and who do you work for or with?
5. Which tools and systems do you use the most?
6. What are you working on right now? Projects or cases that take up space.
7. What do you want to achieve over the next 6 to 12 months?
8. Which tasks eat the most time that you'd like help with?
Communication
9. How would you describe your communication style?
10. Do you have a tone or structure you prefer when you write?
11. Are there words or phrases you dislike and want to avoid?
12. Do you want short summaries, detailed explanations,
bullet points or flowing text?
13. What language do you work in, and should I switch automatically?
Working with you
14. What do you use AI for today, and what do you want to use it more for?
15. What should I definitely not do? Feel free to suggest things
people usually get annoyed by, and ask me to confirm.
16. How should I handle uncertainty: guess, ask, or say that
I don't know?
17. What's the most important thing I should know about how we'll work together?
18. How proactive do you want me to be? Should I only do
what you ask, or also suggest improvements, next steps,
automations and possible problems?
19. If you had an extremely capable personal assistant who had
worked with you for five years: what would that person know about
you that lets them deliver far better than a new hire?
Finish by asking whether there's anything else worth mentioning.
## Afterwards
Create about-me.md. Keep it under roughly 2000 words. Cut everything that
doesn't change the way you answer me.
Write the content as instructions to yourself, not as a
description of me. "Answer briefly, conclusion first" instead of "he
likes short answers".
Use this structure:
# About me: [first name]
## Who I am
## What I'm working on now
## How I like to work
## Communication style
## Rules for you
## Last updated: [date]
"Rules for you" is the most important section. A bullet list with concrete
instructions, including the things I don't want you to do.
Show the file in the chat as an artifact, and give it to me as a
downloadable markdown file as well.