What a CLAUDE.md actually does
If you use Claude Code or Claude Projects, you have probably noticed that Claude does not carry memory between conversations. Every session starts cold. You find yourself re-explaining your project structure, your tone preferences, your customer base, your tech stack. A CLAUDE.md is the fix for that.
The file sits at the root of your project (or gets pasted into a Claude Project's instructions). Claude reads it at the start of every session. Done well, it means Claude already knows:
- What the business or project is and who it serves
- What tools, tech, and systems are in play
- How decisions get made and what matters
- What to avoid and why
- The voice, tone, and style for any output
The name comes from Anthropic's own published guidance on structuring context for Claude. It is not a magic filename, but it has become a convention because it works.
Why most CLAUDE.md files fall short
The common version of a CLAUDE.md is thin. A few bullet points about the project, maybe a tech stack list, a note about preferred code style. That covers maybe 20% of what Claude actually needs to be useful.
The rest is harder to capture because it lives in your head: the assumptions you have never written down, the ways your business is different from the generic version of your industry, the things that have gone wrong before, the customers you are actually building for versus the ones you imagine. A shallow CLAUDE.md produces shallow AI output.
There is also a structure problem. A long block of prose does not help Claude nearly as much as organized sections that separate orientation (what is this?) from depth (how does it work?) from process (how do we do specific things?). Most people writing their own file do not think in those layers.
The options: scratch, template, or generator
Writing from scratch works if you are disciplined and know what to include. Most people either write too little (a paragraph that says "I run a marketing agency") or too much (a wall of text with no hierarchy). The main failure mode is omission: you do not know what you left out until Claude gets something wrong.
Templates are better than nothing. There are several good ones floating around GitHub and the Anthropic docs. A template gives you the structure; you supply the content. The problem is the questions. Generic templates ask things like "describe your project" and "list your tech stack." That is fine for a software repo. For a business brain, the questions need to pull out the things you do not automatically think to write down.
A questionnaire-based generator is the strongest option if you want results without spending hours figuring out what belongs in the file. The difference is that a good questionnaire asks targeted questions that extract the right information, then structures the output correctly. You do not need to know what a CLAUDE.md needs. You answer what you know, and the structure comes out right.
What makes a questionnaire better than a blank page
When you sit down to write a CLAUDE.md from scratch, you are solving two problems at once: figuring out what information Claude needs, and figuring out how to express it clearly. That is a lot of cognitive work, and most people drop most of the value on the floor.
A questionnaire separates those two problems. The questions have already been designed to surface the right information. For a small business brain, that means things like: who is the customer and what do they actually want to accomplish, not just their demographic profile. What does the business do differently from competitors, in concrete terms. What does good output look like versus output that looks right but misses the point. What are the three to five workflows that happen most often.
Those questions are not obvious. They come from understanding what Claude actually needs to be useful, not just what feels natural to write about your business.
What AI Brain Docs generates
AI Brain Docs is built specifically around this problem. You answer six questions about your business. The generator produces:
CLAUDE.md -- the orientation layer. This is what Claude reads first. It covers what the business is, who it serves, how decisions get made, the voice and tone, and what to avoid. It is structured the way Anthropic recommends, not in the generic way most templates follow.
A knowledge base -- seven documents that go deeper on the business. Customer profiles, positioning, products and services, competitive landscape, operational context. These are the depth layer that backs up the orientation.
An AI Action Plan -- a prioritized list of the highest-value things AI can actually do for this specific business, based on its workflows and context. Not generic suggestions.
A toolkit of skills, agents, and prompts -- ready-to-use instructions your AI can follow for specific recurring tasks. Drafted for your business context, not generic.
The whole thing takes about ten minutes to complete. The generation runs server-side and you get the results immediately. The CLAUDE.md and knowledge base are available as a ZIP you drop into your Claude Project, paste into Claude Code, or add to ChatGPT custom instructions.
How to use what you get
Dropping the files into Claude Projects (the "instructions" field) is the simplest path. Claude will read the CLAUDE.md at the start of every conversation in that project. The knowledge base documents provide depth when Claude needs specifics.
For Claude Code, the CLAUDE.md goes in the root of your working directory. Claude Code reads it automatically. The knowledge base can live in a knowledge-base/ subdirectory.
For ChatGPT, the CLAUDE.md content goes into the "custom instructions" field of a GPT you configure. The knowledge base documents can be attached as files the GPT references.
Gemini with a system prompt or Gems works the same way.
The files are plain markdown. Nothing is locked into a platform. If you want to edit a section after the fact, you open it and change it.
Who this is for
If you are a Claude Code user building or maintaining a project and you want Claude to stop asking you the same questions every session, a CLAUDE.md is worth the hour it takes to write a good one. If you want that hour back and want the output to be more structured than what most people produce on their own, AI Brain Docs does it for you.
The one-time cost is around $29.99. There is a free tier that includes the CLAUDE.md and a preview of the knowledge base so you can evaluate the output before paying.
If you have been putting off building out your AI context because the blank page is annoying, start the questionnaire at aibraindocs.com/start and have something usable in about ten minutes.