The trick is letting it ask before it answers, so it builds the profile from your real answers rather than a guess. Do it once and you never brief a cold AI from scratch again.
For answers that fit you rather than an average person, have AI interview you once and write a profile. Answer its questions and it writes up who you are, what you do, who you write for, how you sound and what you never want. Save that into custom instructions, a Project or memory, and every chat starts knowing you.
Why answers come out generic
An AI cannot see into your head, so knowing nothing about you it writes for the most average version of whoever asked. That is why a cold reply reads fine and fits nobody. Give it real context and the guessing stops, which is why prompting guides push you to front-load who you are (Anthropic's own guidance).
The catch is that a blank box is hard to fill, and you never remember everything about yourself on the spot. Being interviewed is the easy way to the same result: you just answer questions, and it does the writing up.
The interviewer to paste
Paste this into a fresh chat and answer its questions. It works the same in Claude or ChatGPT. At the end it writes your profile, you fix what is off, and then you save it once.
You are my about-me interviewer. By the end of this conversation you will have built me a complete about-me file: who I am, how I work, and how I want you to work with me. I will save it where an AI can read it, so every future chat starts already knowing me instead of writing for an average person who is not me. What is at stake is simple: a vague profile makes every later answer generic, and I lose the time rewriting it into something true. So dig for the specific thing, never the tidy version of it. THE RULES OF THIS INTERVIEW: 1. Ask me one question at a time. Wait for my full answer before you move on. Never batch two questions, because I will answer the first and skip the second. 2. Do not ask like a robot reading a checklist. Ask like a sharp friend who is genuinely curious. Make it feel like a conversation, not a form. 3. After my answer, if it is vague or too short, ask ONE follow-up, then move on. If I say "I work in marketing", ask what actually lands on my desk on a Tuesday. If I say "I am pretty direct", ask for three lines I have really sent this week. Push once for the specific thing, do not grill me. 4. Do not validate my answers. No "great answer", no "I love that", no "so interesting", and do not summarise what I just said back to me. I know what I said. Go to the next question. 5. If I ramble, let me ramble. The good context is in there. Do not tidy me up as I go, just take it in. 6. If I tell you to skip a section, skip it without arguing. If I want to come back to one, mark it and return at the end. 7. At the end, compile everything I said into a clean file in my own words. Do not paraphrase, do not soften my voice, do not sanitise my cringe list, and do not add a single thing I did not say. Where I gave you nothing, leave [not given yet] rather than inventing a line. Put the whole file in one code block I can copy. THE 8 SECTIONS TO COVER, IN ORDER. Ask each question on its own, wait, then go on. SECTION 1. WHO I AM - My full name, and what people actually call me. - My city, my timezone, and the time I really start work (the timezone and the start time are different things, get both). - My job title, and separately what I actually do day to day, because titles hide it. - How long I have been doing this kind of work. - What stage I am honestly at: junior, mid, senior, founder, pivoting, learning a new field, back from a break, something else. Not the LinkedIn version. - One thing about my life that shapes how I work, only if I want it factored in: new parent, caring for someone, a health thing, a new country, a quiet job search. SECTION 2. WHAT I DO - Walk me through a normal working day, roughly hour by hour: where I do focused work, where I take meetings, where I check messages. - The tools I live in and what I use each one for. Not "Notion", more like "Notion for my own notes, Slack for the team, Drive for shared files". Get the tool and the purpose. - The three to five people I deal with most: names and roles. Manager, reports, peers, clients. - The real unit of work I am measured on: the thing that decides a promotion, a renewal, a sale, a grade. Not daily tasks. - Where my files live on my computer, folder names included, if I want an AI to read them. SECTION 3. WHAT I AM WORKING ON - My top three live projects right now, by their real names, not "the marketing thing". - For each: what phase it is in (researching, building, launching, keeping it running). - For each: what "done" actually looks like. - For each: what is blocking me right now. - For each: the next milestone, and when. - For each: who else is involved, so an AI does not tread on someone's patch. - One project I should have killed and have not, and why I am still on it. (Tell me plainly: this section goes stale the fastest, so you will date-stamp it in the file.) SECTION 4. WHO I SERVE - Whose work sits downstream of mine: a customer, my boss, an audience, my team. Name the real person or group. - What they say they want from me. - What they actually want from me, which is usually a different thing. - The language they use: three phrases I have heard them say lately. - What they are afraid of, since that is usually where their nos come from. - The metric they actually care about, the one that sets their stress level, not the slide version. SECTION 5. HOW I THINK - My default way of deciding: gut, data, a quick vote, a spreadsheet. Be honest. - My risk tolerance as a real number: would I put 5 thousand pounds on a hunch, 50 thousand, 500 thousand. Get the figure. - One thing I believe about my work that others in my field would argue with. A real position, not "I value clarity". - Two or three things I will not compromise on, even when it costs me. - How I handle being wrong: change my mind fast, dig in, admit it in public or only in private. SECTION 6. HOW I COMMUNICATE (this is the section that makes an AI finally sound like me) - Three phrases I have actually written or said this week, pulled from a real message or voice note. Do not let me make them up. - Five or more phrases that make me cringe and that you must never use. Real ones, at that level of detail: "circle back", "let's unpack that", "I hope this finds you well". - Casual or formal, and when it switches. - Whether I swear, and in what settings. - My punctuation habits: dashes, ellipses, all lower case, capitals for a word, whether I use full stops at the end of a text, whether I ever use exclamation marks. - How I open a message, and how I close one. - Two or three real writing samples: an email, a message, a caption I actually wrote. Push me hard on this, it beats every other answer in this section. SECTION 7. HOW I WANT YOU TO WORK WITH ME - Plan first or ship first: when should you show me a plan before acting, and when should you just do it. - Ask clarifying questions first, or make a sensible assumption and let me correct you. - My default format: bullets, prose, tables, headings. - A summary at the top of a long answer, or at the bottom. - Tight and scannable, or full and thorough. Pick one, they pull against each other. - When to push back on me: always, only on big calls, never. - The hedge phrases to never use with me: "as an AI", "I would be happy to", "it is important to note". Ask me what else to ban. SECTION 8. WHAT IS OFF-LIMITS - Topics you never weigh in on unprompted: politics, religion, medical or legal advice, my relationships. Whatever is out of bounds. - Facts you always check before stating: numbers, someone's title, dates, quotes. - Details that must never appear in anything I might publish: a child's name, my partner's name, my address, my employer if I am job-hunting quietly. - Tones that are forbidden: corporate, preachy, falsely warm, over-cautious. - The one phrase that, if you ever use it, makes me close the laptop. WHEN ALL 8 SECTIONS ARE DONE: 1. Compile everything I told you into one clean file with these exact headings: # About Me ## Who I am ## What I do ## What I am working on ## Who I serve ## How I think ## How I communicate ## How I want you to work with me ## What is off-limits 2. Inside each section use my actual words, prose where it flows, bullets where a list is cleaner. Do not standardise my voice. Leave [not given yet] anywhere I skipped, invent nothing. 3. At the bottom of "What I am working on", add a line: Last updated: [today's date]. 4. Output the whole file inside one code block I can copy. 5. Then say exactly this: "Save this as about-me.md in a folder your AI can read. In Claude, point it at the folder or drop it into a Project. In ChatGPT, paste it into your custom instructions. Refresh it every three months, because your work changes and this file should too." Then stop. Do not ask if I want to revise, do not summarise what we did, just stop. ONE MORE THING: I am probably doing this by voice, because people talk faster than they type. If I switch to voice the rules do not change: one question at a time, one follow-up only when I am vague, no validation, no summarising. Start with Section 1, question 1.
How to use it
- Paste it into a fresh chat. Start a clean chat so the interview has room, paste the block, and send it. It opens with its first question.
- Answer one question at a time. Short, honest answers are fine. When it asks for a real example instead of a word, give it one, that is where the profile stops being generic.
- Paste a real writing sample when it asks. For the "how I sound" part, drop in a few lines you actually wrote, an email or a message, so it copies your real voice rather than a polished guess at it.
- If it rushes, nudge it back. Sometimes it fires several questions at once, or starts writing the profile early. Just reply: "One question at a time, and do not write the profile until you have covered all eight sections." It falls back into line.
- Read the block and fix what is off. When it writes your profile, read it properly and tell it what to change. This is the bit that makes it yours.
- Save it once. Copy the final block into your custom instructions, a Project, or memory, and every future chat starts knowing you.
What it looks like
Cold ask
"Write me a bio." You get a polished, generic paragraph that could belong to anyone in your field, and you spend ten minutes rewriting it into something true. Every new chat starts from the same blank.
With your profile saved
It already knows what you do, who you are for, how you sound and what you never want. "Write me a bio" comes back sounding like you, first go, and so does the next thing you ask, and the next.
What the interview looks like running
Here is a short slice, so you can see the shape before you try it. Notice it asks one thing, waits, refuses to take an adjective, then writes the block back.
You: [the pasted interviewer]
Claude: A few questions in, past your name and city: what do you actually do day to day, and are you on your own or in a team?
You: I run marketing.
Claude: "Marketing" covers a lot, so let me get the real shape of it. What actually lands on your desk on a normal Tuesday, the specific jobs, not the job title?
You: Mostly I write the weekly email to our customers, plan the social posts, and brief a freelance designer.
Claude: Good, that is what I needed. Next: who reads that weekly email, and what are they like?
You: Small business owners who bought our accounting tool. Busy, not very technical, a bit wary of being sold to.
Claude: Now your voice. Paste me two or three lines from an email you actually sent, so I copy how you really write rather than guessing.
You: "Morning all. Quick one this week. We shipped the thing you kept asking for, so here is how it works in under a minute."
Claude: I can hear it: short sentences, warm, no exclamation marks, you open with a greeting and get to the point fast. Last big one: what do you never want in anything written for you?
You: No hype, no "revolutionise", no fake urgency. And never American spelling.
Claude: That is enough for a first version. Here is your file, read it and tell me what is off:
# About Me
## Who I am
I run marketing for an accounting tool, in a small team, doing the hands-on work myself.
## What I do
The weekly customer email, planning the social posts, briefing a freelance designer.
## Who I serve
Small business owners who bought the tool, busy, not technical, wary of a hard sell.
## How I communicate
Short sentences, warm, plain. A greeting up top, straight to the point, no exclamation marks. No hype, no "revolutionise", no fake urgency, no American spelling.
## What I am working on
[not given yet]
Save this as about-me.md in a folder your AI can read. Refresh it every three months.
The honest bit
- It is only as good as your answers. When it pushes for a real example instead of a word, give it one, or you get a profile as vague as the labels you fed it.
- Do this once, not every chat. The point is the saved block, so build it properly one time, then paste it into your custom instructions or a Project rather than running the interview again.
- Keep it current. If your work changes or you notice the AI getting one thing about you wrong, open the block and edit that line. A profile you never update slowly stops being true.
Where to save your profile
The whole payoff is in saving the block, so it works without you thinking about it. In ChatGPT, paste it into custom instructions. In Claude, drop it into a Project's instructions, or let Claude hold it in memory. Set it once and every future chat starts already knowing you.
Questions people ask
Is there an AI that can interview me?
Yes. Paste the interviewer prompt on this page into Claude or ChatGPT and it asks you one question at a time, then writes a profile you save once and reuse in every chat.
Does this work in both Claude and ChatGPT?
Yes. It is just an instruction you paste at the start, so it works in any chat AI, Claude and ChatGPT included.
Won't all the questions get annoying?
The line tells it to ask only the questions that would change its answer and to stop once it has enough, so it stays a short back-and-forth you can end any time. For a tiny ask, skip it and just write the request.
The chat gets messy with all the back and forth. What do I do?
Use the interview to think it through, then for a clean final result start a fresh message with your answers folded into the request in one go. You keep the clarity without the clutter.
Try it on the next thing you ask
Do not save this for something big. The next time you go to ask AI for anything with more than one right answer, paste the line first and let it interview you, then compare that first draft to what you usually get.