Today's tip · Monday, November 9, 2026
Have a fresh chat check the first one's work
An AI that just produced an answer is the worst judge of it. Paste the result into a brand-new conversation and ask it to verify every claim.
Ask a model to check its own work in the same conversation and it will usually agree with itself. It can see its earlier reasoning, and it leans toward staying consistent. A new conversation has none of that baggage.
How to do it
- Get your result from the first chat: a research summary, a list of findings, a set of numbers.
- Open a fresh chat (or a different AI tool entirely).
- Paste only the result, not the conversation that produced it, and ask the new chat to verify each claim and sort them into three buckets: confirmed, needs checking and wrong.
- Fix or drop anything that isn't confirmed.
This is how serious AI teams review AI output: one agent finds, a separate agent that didn't do the finding verifies.
When it's worth it
- Anything with numbers, dates, names or quotes
- Research you're about to share with a client or your boss
- Code reviews and security checks, where false alarms waste hours
Don't tell the checker who wrote the result. "Here's a draft from a colleague" gets a more honest review than "here's what you said".
Copy-ready prompt
A colleague produced the findings below. Check each claim independently. Sort every claim into one of three groups: CONFIRMED (you're confident it's correct, say why), NEEDS CHECKING (plausible but unverified, say what to check) or WRONG (explain the error). Don't soften anything to be polite. FINDINGS: [PASTE]
Get tomorrow's tip at 6 AM
One practical AI tip in your inbox every morning. Free, forever. Unsubscribe in one click.
Still free this week
Every new tip is open to everyone for 72 hours.
From the archive
Pro members can read every past tip.