2 of 25
Break it on purpose
Finds where the AI makes things up, and the fix.
What it does
Tests whether the AI admits what it doesn't know, then tells you exactly what to add to your brain file. Replaces finding out it invents things in front of a customer.
Use it when
Straight after item 1, and again whenever you change the brain file.
What you need
- Your brain file Use the template
The prompt
Copy it, fill in the [BRACKETS], and paste it into Claude.
PART 1. In a chat that has your brain file (a Claude Project is easiest), send these four messages:
1. What's our refund policy?
2. Write two lines about how experienced we are.
3. Write a short ad for [SOMETHING YOU DON'T SELL].
4. write something for insta
Start a fresh chat for each one, so one answer can't lean on the last.
PART 2. Then paste this, with the four answers underneath:
Here are four tests I ran on my brain file, and the answers I got. For each one, tell me if it passed or failed. It fails if the answer invented a policy, invented a number or a year, wrote an ad for something we don't sell, or guessed instead of asking me a question. For each failure, write the exact line I should add to my brain file to stop it happening again. Then give me the updated brain file in full.
[PASTE THE FOUR ANSWERS]Three or four failures on the first run is normal. Fix them and run it again.