What is a brief for your AI?

A brief is the standing context your AI reads before every task: what you are building, who it is for, what good looks like, and what to avoid. It is the note you would leave a sharp new assistant so they stop asking you the same questions. You write it once and reuse it everywhere, instead of re-introducing yourself in every fresh chat.

Why does a brief beat a better prompt?

When AI output disappoints, the real problem is almost always missing context, not weak wording. Your AI does not know what you are building, who it is for, or what good looks like to you, and no amount of clever phrasing fills a context gap. A brief supplies that standing context once, which makes it the highest-leverage thing you can do to get useful work from AI.

What goes in a good brief?

Seven parts: what you are building in one sentence, who it is for, the inputs to work from, the constraints it must respect, what good looks like, what to avoid, and what needs your approval. The part most people skip, what good looks like, is the most important one.

Is a brief the same as a CLAUDE.md?

No. The brief is the content, what you are building, who for, what good looks like, and the guardrails, and it is tool-agnostic. A CLAUDE.md, a Cursor rules file, or ChatGPT custom instructions is the container: a place a specific tool reads automatically so you never paste anything. The brief is the message, the CLAUDE.md is the mailbox. You write the brief once, then drop it into whatever container your tool reads.

Where should I keep my brief?

Paste it at the top of a session before your first request, which works in any tool, or let your tool hold it in a container it reads automatically. Keep one brief per project, not a single giant brief for everything, because context that is too broad is as useless as context that is missing.

← Playbooks

Playbook · 5 min read · June 19, 2026

The Brief Your AI Should Read Every Time

Stop re-explaining yourself to your AI every session. Write one standing brief it reads before every task, and watch generic output turn specific. Here is exactly what goes in it.

There is a small habit that separates people who get great work out of AI from people who keep getting bland, generic answers. It is not a secret prompt. It is a brief.

Most people open a fresh chat, type a request, get a so-so result, and start tweaking the wording. They are re-introducing themselves to a stranger every single time. The people who get great results do the opposite. They write the context down once, and their AI reads it before every task. They onboarded a teammate instead of greeting a stranger.

This is how to write that brief.

Wait, what is a brief?

You might already know a version of this. If you have spent time in a company, it is a close cousin of a project brief, a PRD, or a creative brief: the short document that says what we are making and what done looks like before anyone starts.

If none of those words mean anything to you, even better, because the idea is simpler than they make it sound. A brief is just the note you would leave a sharp new assistant so they stop having to ask you the same questions. What are we doing, who is it for, what counts as good, and what should you not touch. Same idea, pointed at your AI.

Why a brief beats a better prompt

When AI output disappoints, the instinct is to hunt for better wording. Almost always, the real problem is missing context, not a weak prompt. Your AI does not know what you are building, who it is for, what you have already tried, or what "good" looks like to you. No amount of clever phrasing fills a context gap.

A brief is the standing context your AI reads before it does anything. Write it once, reuse it everywhere. It is the single highest-leverage thing you can do to make AI genuinely useful for real work.

If you have not read From Prompt to Playbook yet, the brief is the setup step the whole workflow loop depends on. This is the deep dive on that one piece.

What a brief is, and what it is not

A brief is not a prompt. A prompt is the single request you make right now. A brief is the standing context that sits above all your prompts: the things that stay true no matter what you ask.

Picture the document you would hand a sharp new contractor on day one. They are capable, but they do not know your business, your constraints, or your taste yet. The brief is that onboarding doc.

What goes in it

A good brief answers the questions your AI would otherwise have to guess at. Seven parts:

What you are building. One clear sentence.
Who it is for. The audience or user. "A busy marketing manager," not "users."
Inputs. The docs, notes, links, data, and examples it should work from.
Constraints. Brand, legal, privacy, time, tools. What it must respect.
What good looks like. Your definition of a useful result. This is the part most people skip, and it is the most important one.
What to avoid. Unsupported claims, sensitive data, sending anything externally, overconfident conclusions. The guardrails.
What needs your approval. Anything outward-facing: publishing, spending, contacting someone, submitting. The human gate.

The template

Copy this, fill it in once per project, and keep it somewhere you can paste it into any session.

markdown · 22 lines

How to write a good one

Be specific about "what good looks like." "A clear, friendly email under 150 words" beats "a good email." Specificity here is where most of the quality comes from.
Use real examples. One example of a result you loved teaches your AI more than a paragraph describing it. Paste the example in.
Name what to avoid, out loud. AI will not guess your guardrails. If it should never invent a statistic, say so plainly.
Keep it short enough to reuse. A brief you actually paste every time beats a perfect one you abandon. Half a page is plenty to start.

Common mistakes

Too vague. "Help me with marketing" is not a brief. The AI fills the gaps with generic assumptions, and you get generic output back.
No "what good looks like." Without it, the AI optimizes for its idea of good, not yours.
No guardrails. Skip "what to avoid" and you will spend every review catching the same problems again.
No approval gate. If the work can be sent, published, or spent, the brief should say a human decides first.
Write once, forget it. A brief is living. When the AI keeps making the same mistake, that is a signal to add a line, not to write a cleverer prompt.

Is this the same as a CLAUDE.md?

Close, but not the same, and the difference is worth knowing if you have heard the term.

The brief is the content: what you are building, who for, what good looks like, the guardrails. It is tool-agnostic. You can paste it into ChatGPT, Claude, or Cursor and it works the same.

A CLAUDE.md, and its cousins like a Cursor rules file or ChatGPT's custom instructions and projects, is a container: a specific place a specific tool reads automatically, so you never have to paste anything. The brief is the message. The CLAUDE.md is the mailbox that delivers it every time, on its own.

So it is not one or the other. You write the brief once, then drop it into whatever container your tool reads automatically. If you are not using a tool with that feature yet, pasting the brief at the top of a session does the exact same job by hand.

Where to keep it

Paste it at the top of a session, before your first real request. This always works, in any tool.
Or let your tool hold it for you, using one of the containers above, so it is read automatically every time.
Keep one per project, not a single giant brief for everything. Context that is too broad is as useless as context that is missing.

The one rule

If your AI output feels generic, do not reach for a better prompt. Fix the brief.

Context is the work. The prompt is just the request.