Early access: Pro is $1 while we build the tips library. When paid plans start, you'll get 14 days to keep Pro or move to Free, and early members who stay get 50% off for life. How it works
AI Daily Tips
Subscribe
All tips

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.

AI Daily Tips Editorial

3 min to applyFree · 2 days 6h left

Issue
Level
Beginner
Time to apply
3 min

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

  1. Get your result from the first chat: a research summary, a list of findings, a set of numbers.
  2. Open a fresh chat (or a different AI tool entirely).
  3. 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.
  4. 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

Independent verifier
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]
Open inChatGPTClaude

More tips

All tips