Free · The Four Ways People Ruin Their Brain

Four ways to wreck the document everything runs on.

If you've built a Business Brain — the one plain document that holds what you sell, how you talk, and what an AI must never say in your name — this is the maintenance page. Four failures. All common, all quiet, all of them produce confidently wrong drafts while looking fine.

Haven't built one yet? Start here — it's free and it takes about twenty minutes.

1 — Leaving the instructions in it

The blank template has bracketed prompts: [2–3 sentences: what the business is]. If those are still in your document, the AI will treat them as facts about your business. It doesn't know they're instructions. It reads them as content.

The tell: drafts that are strangely vague in exactly one area, or that repeat back template language.

Fix: open your Brain, search for [, and replace every bracketed instruction with something true. The only brackets that survive are [FILL LATER: ...] — deliberate gaps, which are honest. A gap the AI knows about is safe. An instruction it mistakes for a fact is not.

2 — Writing it like marketing

The Brain is for the AI, not for customers. It's the single most common mistake and it feels like doing a good job.

"We craft artisanal experiences with a passion for excellence."

That teaches the AI nothing. There is no fact in that sentence. Drafts built on it come out as brochure copy, because brochure copy is all you gave it.

"Cakes from ฿1,800. 7 days notice. 50% deposit. We don't deliver outside the city."

That teaches it everything. Every draft from that Brain can now answer a real question correctly.

Fix: read your Brain and delete any sentence that doesn't contain a fact, a number, a rule, or an example of how you actually talk. If a sentence would look at home on a homepage, it's in the wrong document.

3 — Letting it go stale

A Brain with December's availability in it, read in March, produces confidently wrong drafts. Not vague ones — specific, fluent, wrong ones. It'll tell a customer you're fully booked when you're empty, in your own voice, with total conviction.

This is the failure that costs real money, because stale drafts don't look like errors. They look like answers.

Fix — the one-line rule: something changed? Update the line the same day. Fifteen seconds. Update the date at the top while you're there.

That date is your honesty meter. If it's three months old, you don't have a memory document — you have a historical record, and the AI can't tell the difference.

4 — Skipping the boundaries

Section 6 — the never-rules. It's last, it's the least fun to write, and it's the section that makes wrong drafts harmless instead of expensive.

Without it, the worst-case draft is: an AI confirming a date you can't make, quoting a price you don't offer, or apologising for something you didn't do — in your name, to a real customer, fluently.

With it, the worst case is a draft that says "I need to check that with Jee." Which is fine. Which is, in fact, correct.

Fix — one rule per fear. Ask yourself: what's the worst thing a wrong message could do to this business? Whatever comes to mind first, write the rule that prevents it. Then the next one. Five rules is plenty. Mine are things like never claim a customer we don't have and never promise a date I haven't confirmed.

The habit that makes all four self-healing

The wrong-draft rule. A draft got something wrong in a way that annoyed you? Don't just fix the draft. Ask: what line in the Brain would have prevented this? Add it.

That mistake now never happens again. Do it for two weeks and the document stops being something you maintain and starts being something that maintains itself.

The 20-Second Judge — how to catch the wrong draft in the first place.

What never to paste into an AI chat — what belongs in the Brain, and what doesn't.