This AI prompt makes AI check its own answer. Paste it right after a reply & the model goes back to name the 3 weakest assumptions it made, why it made them & what changes if they’re wrong.
Make the LLM critique its own answer
You asked a question. The model gave you an answer. But it also made choices you didn’t see. What it assumed you meant. What it filled in on its own. What it left out because you didn’t ask.
This prompt makes it go back & show you where it guessed.
Look at your last response. Find the 3 weakest assumptions in it. For each one, say what you assumed, why you assumed it, and what would change if it were wrong.
If you can't find 3 real assumptions, say so. Don't invent one to fill the count.Why it works: Most prompts ask the LLM to produce. This one tells it to check itself. It finds the gaps you didn’t think to ask about, which is the whole point of a 2nd opinion.
One gap: “Weakest” has no definition. The model picks what sounds weak, which sometimes means inventing a critique just to hit the count of 3.
Simple fix: The last part (“if you can’t find 3, say so”) already handles this.
Want to go deeper: After it lists the 3, paste back: “Pick the one most likely to be wrong. Be the devil’s advocate on your original answer. Be brutally honest.”
Original post by u/adCold1610. Link
When you need AI to admit what it doesn’t know
This prompt checks an answer after it’s written. To get the model to flag a guess while it’s still answering, use the hallucination prompt. For facts you plan to quote, the cite sources prompt shows you where each one came from.
This prompt first ran in issue 3.