If you’ve read the others, you have the whole check: make the AI name the exact document it’s looking at, make it hand you one fact you can verify in under a minute, and bin the lot if it fails either. It works on a phone contract, a letter from a doctor, an email from your child’s school. Nothing in it is about money.
So why do I keep going on about it? Because at some point I had to test it somewhere a wrong answer comes with a bill.
The version with a bill attached
The decisions where this matters most all share one thing: getting them wrong comes with a bill. Sometimes the bill is money: a fund you’re choosing, a fee you didn’t spot. Sometimes it’s a letter from a doctor you’re about to act on, a contract you’re about to sign, or the paperwork for a trip you’ve already booked. Misread a holiday email and you lose an afternoon. Misread one of these and you lose something that stings. And the AI sounds every bit as helpful walking you off the edge as it does when it’s right.
Here’s one that caught me, and it had nothing to do with money. I asked an AI to help sort the paperwork for a trip to the States, and it told me, with total confidence, that I needed something called a “Sofi.” There’s no such thing. The document was an ESTA. It hadn’t checked anything; it had produced a plausible-sounding word and handed it over as fact. Thirty seconds of verifying caught it. Get that one wrong and you’re turned away at the airport on a trip you’ve already paid for.
So I run the same two questions on anything that carries a bill, write down the verdict, and (the part most people skip) write down the times the verdict was wrong. The place I’ve pushed it hardest, and logged the most catches, happens to be money: it’s an unforgiving proving ground, and a confidently wrong AI makes a far more useful witness than one that simply agrees with me. It’s less flattering to publish, which is presumably why nobody does.
Here’s the check in a single paste, whatever the stakes are for you. Keep it handy for the next answer that matters:
Before you answer anything else: name exactly what you’re looking at, the specific document, page, letter or account. Then give me one fact about it I can check myself in under a minute. If you’re not certain what you’re looking at, say so before you go on.
Three places to see it work
- The whole method, laid out. The four questions in full, with prompts you can copy.
- Where AI got it wrong. A running log of the answers that didn’t survive a check. The point of the site, really.
- One that fooled me for a paragraph. I asked an AI to review my website and it confidently reviewed someone else’s, then praised a framework that wasn’t mine. Every fact in it was true. None of it was about me. That is the failure the check is built to catch, screenshots and all.
Start wherever the stakes are highest for you. Mine happen to involve money. Yours might be a contract, a diagnosis, or an email you’re about to reply to at eleven at night. The check doesn’t care which. It just asks the AI to prove it’s looking at your problem before you believe a word, because how often AI is quietly wrong on things you can verify is the whole reason the check exists.
Ben tests how far you can trust the main AI assistants, and publishes exactly where they get things wrong. Every post here is a first-hand test with the receipts, including the times a tool simply wasn’t worth the trust. About Ben →
The site runs AI on real investing decisions. Start with the Prompt Stack for the four-stage framework, free and ungated, or the Bluff Filter for the paste-ready version with a real before and after.