The Exact README Template That Makes My AI Actually Useful

If I could only teach you one file, it would be this one. The README is the piece most people are missing from their AI setup, and it's the difference between an AI that guesses and an AI that knows how you work.

A README is just a plain-text file that sits at the top of a folder and explains what's inside and how to use it. (The name is a programming convention. It literally means "read me first.") Mine sit at the top of every folder in my business, and they do two jobs: they teach the AI how we work with whatever lives in that folder, and they point it to the right place for whatever I'm asking. It's an onboarding doc for the tool.

Here are the exact templates, three versions, plus the test that tells you whether yours works.

What a README has to do

Every good one answers four questions, in order:

1. What is this? One paragraph of orientation.
2. How do we work here? The preferences, the never-dos, the vocabulary.
3. Where do things live? A map of the subfolders, so nothing gets hunted for.
4. What are the rules for working in this folder? The behaviors you want every single time.

Everything else is optional. A README that answers those four in one screen beats a five-page one nobody maintains.

Template 1: the client (or project) folder

# [Client Name] — README

## What this folder is
One paragraph: who this client is, what we do for them, and the
current state of the work (e.g., "monthly content retainer, currently
mid-website-rebuild, launch planned for fall").

## How we work with them
- Preferences: [formats they like, how they want to receive work,
  meeting cadence]
- Never: [the things they've said no to, topics to avoid, formats
  they hate]
- Their vocabulary: [what THEY call their offers, audience, and
  programs — use their words, not generic ones]
- Voice: their copy follows [link/path to their voice guide]

## Where things live
- /meeting-notes — every call, filed by date, newest first
- /strategy — positioning, brand guidelines, plans
- /content — drafts and shipped work, organized by month
- /data — analytics and performance

## Rules for working in this folder
- Read the two most recent meeting notes before drafting anything
- All copy follows the shared rules in [path to your rules doc],
  plus this client's voice guide above
- If something I ask for contradicts their strategy doc, flag the
  conflict instead of guessing
- Finished work gets filed in /content under the current month

Template 2: the department (or role) folder

For folders that mirror your org chart, where each role is an instruction file:

# Marketing — README

## What this folder is
The marketing department: one instruction file per role. Each file
defines that role's job, standards, and examples.

## The roles
- director-of-marketing.md — strategy, channel decisions, review gate
- content-marketer.md — long-form drafts, newsletters
- social-media-manager.md — platform posts, scheduling notes
- copywriter.md — sales pages, emails, launch copy
- graphic-designer.md — visual direction, brand standards

## Rules for working in this folder
- When I ask for work, adopt the matching role file FIRST, then the
  shared rules in [path], then do the task
- If a task spans roles, say which roles you're combining
- Role files get updated when I give feedback — save corrections to
  the relevant role, not just the chat

Template 3: the personal folder

Same skeleton, smaller stakes. Health, home, money, travel:

# Health — README

## What this folder is
Everything health: workout logs, food notes, wearable exports,
practitioner notes.

## How I work
- Goals right now: [targets]
- My experts: [nutritionist, trainer — their feedback outranks
  AI suggestions, always]
- Never: [the advice patterns you don't want, e.g., "don't suggest
  cutting foods my nutritionist approved"]

## Where things live
- /logs — daily food and workout notes
- /data — wearable exports by month
- /expert-notes — what my practitioners actually said

## Rules
- Trends and observations are welcome; diagnoses are not
- When my expert's notes conflict with a pattern you spot, flag it
  for my next appointment instead of overriding

The test: ask a stranger

A README works when the AI can answer this correctly with nothing else:

Read the README in this folder, then tell me: what would you do
first if I asked you to [draft this client's newsletter / log
today's workout / write a sales page]? Where would you look, and
what rules would you follow?

If the answer maps to how you actually work, the README is doing its job. If the AI improvises, your README has a gap, and the gap it exposed is exactly what to add. Run this test once per folder; it takes two minutes and catches everything.

Keeping them alive

• Update the "current state" line when things change. A README that says "mid-website-rebuild" a year after launch teaches the AI stale context. I touch mine whenever the work shifts phase.
• Feed feedback into them. When you correct the AI twice for the same thing, that correction belongs in the README (or the role file), not in chat purgatory.
• Keep them one screen long. The moment a README needs scrolling to skim, split the detail into its own file and link it.

FAQ

What's the difference between a README and my AI rules doc? The rules doc is global: how you write, what you never do, your standards everywhere. A README is local: how to work in THIS folder, with THIS client. READMEs point back to the rules doc so you never duplicate rules.

Do I really need one in every folder? Every folder your AI works in, yes. Folders it never touches can wait. Start with your two busiest clients and your rules folder; that covers most asks.

How long should a README be? One screen. Orientation, preferences, map, rules. Detail lives in linked files.

Does this work outside GitHub? Anywhere plain files live: Obsidian, Drive, Notion. The convention (a file named README at the top) is what matters, not the platform.