Angelica.
← all resources

Stop re-explaining yourself to AI

ai prompt pack

Every new chat starts the same way. You spend ten minutes explaining what you do, who you sell to, what you have already tried. Then you get advice written for a generic coach with a generic audience. Below is a prompt that interviews you once and writes the file that ends that loop.

What you walk away with: a single markdown file, under 800 words, that you paste at the top of any AI project. It holds who you are, what you sell, who it is for, and what you are against. Not your writing style — that is a different file and a different job.

why the model needs it

Without it, you get the average

When you tell a model you are a consultant, it does not picture you. It reaches for the middle of every consultant it has ever read. That middle is real, it is just not you, and it is where every generic piece of advice comes from.

The fix is not a better prompt each time. It is removing the guesswork permanently. A model that knows you sell a 30-day course to people who have already bought three other courses will stop suggesting you launch a course.

Two files do two different jobs. about-me.md holds facts and positions: what you sell, who for, what you refuse. A voice file holds mechanics: sentence length, punctuation, the words you have banned. Keep them apart. Mixing them makes both weaker and doubles the length the model has to hold.

the whole thing

The prompt

Open a new chat in ChatGPT or Claude. Paste this. Answer the questions as they come. It takes twenty to forty minutes if you are honest, and about eight if you are not, which is how you will know.

the interview prompt
You are going to interview me, and then write one file.

The file is called about-me.md. Its only job is to be the first
thing another AI reads about me, so it never has to guess who I
am, what I sell, or who I am talking to.

It is NOT about how I write. Do not ask about tone, voice,
sentence length, or formatting. That is a different file.

HOW THE INTERVIEW WORKS

- One question at a time. Never two. Wait for my answer.
- After each answer, decide one thing: could a stranger repeat
  this back to me and be right?
  - If yes, move on. Say nothing else. No praise, no "great,
    that's really clear." Just the next question.
  - If no, push once. Name exactly what was vague, and ask for
    the concrete version: a name, a number, a price, a sentence
    someone actually said to me.
  - Push once, then accept and move on. This is an interview,
    not an interrogation.
- Never write my answer for me. Never give me options to pick
  from. If I am stuck, ask a smaller question instead.
- Never summarise what you have so far until the very end.
- If I contradict something I said earlier, tell me, and ask
  which one is true now.

WHAT TO COVER

Work through these in order. Ask as many questions per area as
you need, one at a time.

1. Who I am. What I did before this. What in my background is
   unusual for my field, including the parts I think are
   irrelevant.
2. What I actually sell. Names, formats, prices if I give them,
   and which one I want most people to buy first.
3. Who it is for. Not a demographic. What they are trying to do,
   what they have already tried, and why it did not work.
4. Who it is NOT for. Be specific here. Ask a follow-up.
5. What I am against. The advice everyone in my field repeats
   that I think is wrong, and why it is wrong.
6. The one thing I believe that, if I stopped believing it,
   would change everything I sell.
7. Proof. Numbers, outcomes, anything someone could check. If I
   have no numbers yet, ask what I have instead.
8. Where I want this to be in twelve months.

THREE RULES FOR MY ANSWERS

Tell me these before the first question, then hold me to them.
- Answer like you are telling a friend, not writing a bio.
- Use real names, real numbers, real sentences people said.
- Do not polish. I will polish later. You cannot un-polish.

WHEN THE INTERVIEW IS DONE

Ask me one last question: "Is there anything you nearly said
and then didn't?" Wait for the answer. Then write the file.

OUTPUT RULES

- Use only my words and my facts. Invent nothing, soften
  nothing, and do not make me sound more impressive than I said.
- Where my answer was thin, write [THIN: what is missing]
  instead of filling the gap. I would rather see the hole.
- Plain markdown. No preamble, no closing commentary.
- Under 800 words. A machine reads this at the start of every
  session. Long is expensive and gets skimmed.

FORMAT

# about-me.md — [my name]

<identity>
Who I am, what I did before, what is unusual about that.
</identity>

<what-i-sell>
Each offer, one line. Mark the one that comes first.
</what-i-sell>

<who-it-is-for>
What they want, what they tried, why it failed.
</who-it-is-for>

<who-it-is-not-for>
Written as refusals, not as a softer version of the above.
</who-it-is-not-for>

<what-i-am-against>
The advice I think is wrong, and the reason.
</what-i-am-against>

<the-one-belief>
One sentence. The load-bearing one.
</the-one-belief>

<proof>
Checkable facts only.
</proof>

<where-this-is-going>
Twelve months.
</where-this-is-going>

Start with question one. Nothing before it.

the parts that do the work

Why the prompt is built like that

Most interview prompts fail in the same four places. If you edit this one, keep these.

One question at a timeGiven five questions at once, you answer the easy ones properly and the hard one in six words. The hard one is the file.
Push once, then stopWithout a limit the model either accepts everything or grinds you down until you make something up to get out. One push is where the real answer usually lives.
No praise between answers"That's a powerful insight" after every reply trains you to perform. You start writing for the compliment instead of for the file.
[THIN] instead of fillingThe default behaviour of every model is to produce a complete-looking document. A complete-looking document with three invented positions in it is worse than an honest hole.
The XML-style tags are not decoration. They give you addressable sections. Later you can write "read only the <who-it-is-not-for> block and tell me if this post breaks it", and the model knows exactly what you mean.

answering it properly

The four questions people fake

You will sail through most of it. These are the ones where the interview decides whether the file is worth anything.

Who it is not for

Not for people who aren’t ready to do the work.
Not for anyone who wants a posting schedule. I will not give you one, and you will leave annoyed.

Almost everyone answers this as a politer version of who it is for, which is a compliment to your buyers dressed as a boundary. A real answer names someone you could sell to and choose not to.

What I am against

I’m against fake gurus who only care about selling courses.
I’m against consistency being sold as a strategy. Posting daily with nothing to say is how most people quit.

The temptation is a target nobody defends. Being against fake gurus costs you nothing, because nobody self-identifies as one. Be against something your own audience currently believes. That is the sentence that makes the file useful.

The one belief

Test it: if you deleted this belief, would anything about what you sell have to change? If not, it is a value, not a thesis. Values are lovely and they do not position anything.

What is unusual about your background

Whatever you almost did not mention because it felt irrelevant is usually the answer. The parts of your history that are normal to you and rare in your field are the only structural advantage you have, and they are the first thing you discount.

The [THIN] markers are the bit I would not cut. First time I ran this I got four of them, and three were in the section I thought I had nailed.

where the file lives

Putting it to work

  1. Save the output as about-me.md. An actual file, not a note in your phone. You will edit it four or five times.
  2. Fix every [THIN] marker, or delete the line. Do not leave them in for the model to interpret.
  3. In ChatGPT or Claude, make a Project and upload the file to the project knowledge. Every chat inside that project now starts with it. This is the version that actually sticks, because it survives you forgetting.
  4. No Projects available? Paste the file first, your request second, in the same message. Order matters: context before task.
  5. Re-read it every time your offer changes. A stale about-me file is worse than none, because it argues confidently for a business you no longer run.

did it work

One test, ten seconds

With the file loaded, ask for three content ideas.

If any of the three could be posted on a competitor's account without changing a word, your <what-i-am-against> and <who-it-is-not-for> sections are too polite. Those two blocks are the only ones that carry any difference. Everything else in the file is description.

Go back and rewrite them until someone could disagree with you.

Write it once, fix it three times, and you will never open a blank chat and start with "so, a bit of context about me" again. That alone is worth the forty minutes.