Build your immunity in ten minutes: ask your AI tool five detailed questions about something you know deeply — your hometown, your hobby, the TV show you can quote. Push into specifics: dates, names, episode details, prices.
Somewhere in those answers, odds are good you'll catch it — a small confident wrongness, stated in the same smooth voice as everything true. That catch is the lesson no warning paragraph can teach: you now know, from experience, exactly what it sounds like. (It sounds like everything else. That's the point.)
Prompt Pack: Accuracy and Review
[YOUR NORMAL REQUEST]... Afterward, list anything in your answer I should verify before relying on it — facts, figures, anything you might be uncertain about.
Here's a draft: [PASTE]. List every checkable factual claim in it — every date, number, name, and stated fact — as a checklist I can verify line by line. Don't assess them; just extract.
In your last answer, separate clearly: what came from the information I gave you versus what you added from general knowledge. I need to know which parts to verify.
Review this document for unstated assumptions about people — gender, family structure, holidays, food, physical ability, work schedules: [PASTE]. List what it quietly assumes and suggest neutral alternatives.
Draft [DOCUMENT, e.g., 'a job posting for an office coordinator'] using gender-neutral language and requirements phrased to welcome the widest qualified pool. Then note which phrases you avoided and why.
Take the position that this draft contains an error: [PASTE]. Hunt for it — contradictions, suspicious numbers, claims that don't follow. If you find nothing, say what you'd still verify first.
Before I use your answer: how confident are you in it, what would make you more certain, and where would a careful person double-check? Answer plainly.