AI Guides › Claude Mastery

Writing An "About Me" File Claude Actually Uses

By Nigel Guy · 3 min read

People who write an "About Me" for Claude tend to write it like a diary entry — a paragraph of background, a bit of personality, how they got into their line of work. It reads well and does almost nothing, because none of it is structured for a model to pull facts out of under time pressure. An About Me file isn't there to be read for charm; it's there to be scanned for specifics.

The mechanism: the four-block structure

1. Role and context, one line each. Your job title, who you work with day to day, and the kind of output you're usually after (emails, code, analysis, drafts). Not a career history — just enough to calibrate a default register.
2. Standing preferences, as a list. Formatting habits (tables over prose, short paragraphs, no headers under a certain length), tone (direct, no hedging, British spelling), and anything you correct often enough that it's clearly a pattern rather than a one-off. If you've told Claude the same thing three times across different chats, it belongs here.
3. Recurring facts, not one-off ones. Names of tools, projects, or people you'll reference repeatedly — the things that would otherwise need re-explaining at the top of every new chat. Leave out anything true only for this week.
4. Explicit exclusions. A short line on what not to assume — for instance, that you don't want unsolicited praise on your drafts, or that a particular topic in your work is confidential and shouldn't be extrapolated on. Silence reads as "no opinion here"; a stated exclusion reads as a boundary.

Write each block as short labelled lines or a small table, not paragraphs. The point is that any one fact can be found and corrected in isolation, without editing a wall of text around it.

What to skip

Skip narrative entirely — the "how I got here" story that feels natural to write adds nothing a model can act on. Skip listing preferences you've never actually needed enforced; a long file of hypothetical rules is harder to scan than a short one of real ones, and dilutes the parts that matter. And don't duplicate what the built-in memory feature already picks up on its own — if it's reliably inferring your role from ordinary conversation, you don't need to also spell it out here.

Guardrails

• Where this file lives — Settings, a Project's custom instructions, or a pinned first message — and exactly how much of it gets used on any given reply is a product detail that changes; check the current setup rather than assuming a particular field behaves as it did a version ago.
• Anything you write here should be something you'd be comfortable being visible to anyone who has access to the same account, Project, or shared workspace — treat it as a working reference document, not a private diary, regardless of where it's technically stored.
• A good About Me file improves defaults; it doesn't guarantee a correct answer on any single request. Keep stating anything genuinely important directly in the conversation where it matters, rather than assuming the file alone will carry it.

All 751 AI guides · JulieMango plans from £17/mo