Most people use AI assistants with almost no personal context.

They ask for a draft, get something generic, then spend the next ten minutes correcting tone, structure, taste, assumptions, and all the tiny preferences that were obvious to them but invisible to the model.

That is a bad loop.

The better version is to give the assistant a short file about how you think. Not a biography. Not a brand document. An opinions file.

An opinions file captures the durable stuff: how you write, what good work looks like to you, what you hate, how you make decisions, where you like more detail, where you want the shortest possible answer, what kinds of examples actually land.

If the person already has public writing, the assistant should start there. Blog posts, essays, newsletters, LinkedIn posts, podcast transcripts, internal memos, long emails, anything with enough signal. Existing writing is better than interview answers because it shows taste in the wild.

If there is no writing to work from, the assistant should interview the person instead.

Here is the prompt.


You are helping me generate an opinions file for an AI assistant.

The goal is to capture how a person thinks, writes, decides, and reacts, so that any assistant can use the file as durable context when working with them.

Do not make this specific to one person, company, role, or audience size. The output should be reusable for anyone.

First, try to discover whether the person already has useful writing or recorded thinking that you can learn from. Ask if they have any of the following:
- a personal blog
- essays or articles
- newsletters
- LinkedIn posts
- X/Twitter posts or threads
- podcast appearances or transcripts
- talks, interviews, or YouTube videos
- internal memos, strategy docs, or long emails they are comfortable sharing
- examples of work they like or dislike

If they have existing material, ask them to share links, files, or excerpts. Use that material to infer their communication style, taste, recurring ideas, decision-making patterns, and strong opinions.

If they do not have useful material, or if the material is too thin, interview them instead.

Ask questions that surface:
- strong opinions
- taste
- communication style
- decision-making defaults
- things they dislike
- how they judge good work
- how they prefer assistants to behave
- recurring phrases or language patterns
- beliefs about work, people, products, and the world
- examples of outputs they like or hate

For each answer or source, convert the signal into a concise, durable opinion that an assistant could actually use later.

Write the final opinions file in this format:

# Opinions file

## Communication style
- ...

## Taste and quality bar
- ...

## Decision-making
- ...

## Work preferences
- ...

## Assistant preferences
- ...

## Things to avoid
- ...

## Strong beliefs
- ...

## Useful phrases / voice notes
- ...

Rules:
- Keep each opinion short and usable.
- Do not include temporary facts.
- Do not include private details unless the person explicitly gives them.
- Do not overfit to one conversation or one piece of writing.
- Prefer "Person prefers..." or "Person believes..." over commands.
- If something is uncertain, mark it as "Possible preference" instead of stating it as fact.
- Separate observed evidence from interpretation when useful.
- The file should help an assistant act more like a good collaborator, not impersonate the person.

Start by asking whether the person has existing writing or recorded thinking to use as source material. If they do, review it before asking follow-up questions. If they do not, interview them one question at a time. After 10-15 good answers or enough source material, draft the opinions file and ask what feels wrong or missing.

The output should be boringly practical. A good opinions file is not trying to make the assistant sound like you. It is trying to stop the assistant from repeatedly making the same wrong assumptions about you.

That is the whole point: less re-explaining, fewer generic drafts, and a collaborator that gets closer to your default settings every time you work with it.