Seven short documents your AI tools read before they write, an agent that drafts them from your website in one command, and the audit that tells you when they've gone stale.
FREE Nothing to buy. Download, run, keep.Every prompt re-explains the company from scratch, so every output drifts a little further from the last. The fix isn't a better prompt. It's a folder the tools read first: seven documents, plain text, two pages each. That folder is the knowledge base. Build it once, keep it current, and "AI slop" stops being a model problem and starts being a context problem you can actually solve.
Pull up three pieces of AI-written content from your brand (a post, an email, a page). Check each against the five symptoms and count the hits.
| Symptom | What it looks like | Which document fixes it |
|---|---|---|
| Channel drift | Site, deck, email, and social describe slightly different companies | 01 Positioning, 07 Company Profile |
| The re-explain tax | Every prompt starts with a paragraph about who you are | All seven. That paragraph is the knowledge base, unwritten. |
| Category voice | It reads like your industry, not like you | 06 Differentiation, 03 Voice and Tone |
| Wrong room | It's talking to someone who isn't your buyer | 02 Ideal Customers |
| Nothing stays current | The brand doc from March is already wrong | 05 Competitors, 07 Company Profile (dated fields) |
Three or more: the tools are guessing. Run the audit again 30 days after the base exists. The count should be zero or one.
Numbered because the order matters: each document feeds the next, and 03 sets the voice the others get rewritten in. Two pages per document is plenty. A paragraph is a fine start.
If a customer had to choose between you and the obvious alternative, what is the single reason they'd pick you, in one sentence?
Example line: "The knowledge base your other tools should read from. Not another writing tool."
Who were your last three best customers, and what did they have in common the day before they found you?
Put your best-performing piece of writing next to your worst. What does the good one do that the bad one doesn't?
Draft this one from your best existing copy, then go back and rewrite 01 and 04 in it.
What are the three things you'd want every customer to be able to repeat about you after one conversation?
When you lose a deal, who do you lose to, and when you win, what did they try first?
What would the customer lose if they went with the alternative? Answer in their words, not yours.
Why this one matters most: Positioning says what you are. Differentiation says why the alternative isn't good enough. Without it, every AI draft reads like the category.
If a new hire had one page to read before their first day, what would be on it?
The kit is the folder from the live session: seven skeletons, the rules the agent follows, and four jobs. /kb-build drafts all seven from your site. /kb-ingest feeds it a transcript or an old deck and proposes edits. /kb-review is the monthly check for what went stale. /kb-check scores any draft against the base and hands back the fix. Everything after the first build is propose-then-apply; nothing changes without you. Works in Claude Code out of the box; Codex, Cursor, and any agent that reads a project file get the same jobs through AGENTS.md. No agent? PROMPT.md is the paste-in version for Claude or ChatGPT.
brand-kb-starter-kit/ ├── README.md start here, 15 minutes ├── CLAUDE.md what goes where, what never goes in ├── AGENTS.md same job for Codex and Cursor ├── PROMPT.md paste-in version, no files needed ├── audit.md the 10-minute drift audit ├── .claude/commands/ │ ├── kb-build.md /kb-builddraft all seven │ ├── kb-ingest.md /kb-ingest feed it new material │ ├── kb-review.md /kb-review monthly stale check │ └── kb-check.md /kb-check score copy, get the fix ├── inputs/ transcripts, old brand docs, reviews └── knowledge-base/ ├── 01-positioning.md ├── 02-ideal-customers.md ├── 03-voice-and-tone.md ├── 04-messaging-framework.md ├── 05-competitors.md ├── 06-differentiation.md └── 07-company-profile.md
/kb-build https://yourcompany.com. About eight minutes. It ends with what your site didn't tell it and three questions to answer.03-voice-and-tone.md first. Fix what it got wrong. Then the other six./kb-check on something you wrote last week. Or run a prompt you use often with the folder loaded and without. Keep the one that sounds like you./kb-ingest whenever a transcript lands, /kb-review once a month. About twenty minutes a month, by hand. The scheduled version that runs across your connectors and stages a Friday review is what the workshop sets up on your machine.Same seven documents, same rules, one prompt. Replace the URL and paste. You'll get text instead of files; save each one as its own document.
/kb-review. It re-reads the site and inputs/, flags stale dates and contradictions, writes proposals. Read them, say "apply."One site, one command, fifteen minutes of editing. If it takes more than 30 minutes end to end, reply to the lesson email and tell me where it stalled.
Back to the kit ↑Four hours, live, all seven documents built for your brand with your judgment in the loop, and you leave with the agent set up on your machine.
$497 · code LESSON100 takes $100 off for lesson attendeesEnroll on Maven →I run the drift audit on your real content, build the first draft of all seven live with you on the call, and hand you the folder.
$497Book on Maven →Already took my knowledge base course? Skip the workshop; it will be repetitive. Email me and I'll tell you honestly which of these, if any, adds something.