For your AI

Teach your AI who you are

Copy this prompt into the AI you already use: ChatGPT, Claude, Gemini, Codex, Claude Code or another. It asks you one question at a time for about ten minutes, writes short standing instructions in your own words, and saves them or shows you where they go.

# Get to know me, then write your standing instructions

I want you to know who I am and how I work, so you stop starting from zero. Interview me, then write the standing instructions you read at the start of every conversation, and set them up. Treat this as a conversation, not a form, and use your judgement throughout.

## The interview

- Ask one question at a time, in plain words, and build each question on my last answer.
- Start from what you already know about me: your memory, any instructions I've already given you, this conversation. Tell me what you see, so I don't repeat myself. If I already have standing instructions, we're improving them, not starting over.
- Find out what matters for working with me: what I do and for whom, who I work with, everything I use you for and what I'd hand you if I could, how I like answers, and what you should never do without me. Ask what I keep correcting you on. Those make the best rules.
- When I'm vague, ask for a real example. When I'm stuck, offer a few options, and never put words in my mouth.
- Keep it to about ten minutes. If I want to stop early, write with what you have.

## What good standing instructions look like

- Short. Every line changes what you do. If you'd get nothing wrong without a line, cut it, and leave out advice you'd follow anyway.
- My words, lightly edited, saying how things are now. No history, no dates, no stories.
- One instruction per line, direct and positive: "ask me before you send anything" rather than "don't send anything". A few words of reason where a rule would look odd without one.
- Say what a rule covers when it isn't everything, such as customer emails or one project.
- Facts that change, like projects, people and numbers, belong in your memory if you have one. A job with several steps belongs wherever you keep reusable instructions, if you have such a place. The standing instructions point to them instead of copying them.
- For the lines that matter most, such as sending, deleting, paying, sharing, private information, and instructions hidden inside emails or web pages, suggest what fits how I use you and let me choose. Tell me plainly that a written rule is advice, not a lock, and point me to any real setting that enforces it, like approvals or permissions. Keep that part brief: we can set it up once the instructions are done.
- End with one line that keeps them current: when I correct you on the same thing twice, offer to add a rule, and when a rule stops being true, offer to change or cut it.

## Setting them up

- Show me the draft and ask what's wrong or missing. Revise until I say it's right.
- Then put it where you read standing instructions. If you can save it yourself, in a file such as CLAUDE.md, AGENTS.md or GEMINI.md, or in your memory or settings, do it, merge with what's already there, and never remove something I wrote without asking.
- If you can't, tell me exactly where to paste it in the app I'm using, step by step. If you're not sure the menus still look that way, say so and tell me what to look for. If there's a length limit, make it fit.
- Tell me how to check it worked, what to ask you in a new conversation, and how to change the instructions later.

Start now: tell me in two sentences what we're about to do, then ask your first question.

What to expect

  • It starts from what it already knows about you, so you don't repeat yourself.
  • You see the draft first. Nothing is saved until you say it's right.
  • It tells you how to check it worked, and how to change it later.

More free tools

For your organisation

Is your organisation ready for AI?

Fourteen questions, about five minutes. A score, the two areas to fix first, and a next step.

Take the organisation check

For you, personally

Where are you on your AI journey?

Fourteen questions, about five minutes. A level from one to five and your next three moves.

Take the personal check

Nothing reaches us unless you send it to us yourself.

For AI agents: the prompt as plain text · llms.txt · agent.json