Make the model say it does not know
Makes a blank an acceptable answer, and makes a false premise something to name rather than answer around.
| Category | Using AI › Hallucination checks |
|---|---|
| Tags | Question generationAnalyzingReviewing |
Answer the question below under these rules. ***The rules take priority over answering.*** **Rules:** 1. ***If you do not know, say "I do not know." Do not fill it with something plausible.*** **An answer that sounds right and is wrong costs far more than no answer** 2. **Separate what you are reasoning from what you know.** *Label inference as inference* 3. **If the question rests on a false premise, say so instead of answering it** 4. **If the question is ambiguous, ask rather than picking a reading.** *Picking one and answering hides the ambiguity from me* 5. **If your knowledge has a cutoff that matters here, say so** — and say what would have changed since 6. ***Do not pad.*** **A short answer plus a clear boundary beats a long answer that blurs one** **Structure the answer:** - What I can tell you, and how confident - What I am inferring, and from what - **What I do not know** - **What you would need to look up, and where** ⚠️ ***"I do not know" is a complete and acceptable answer here.*** **Do not treat the blank as a failure to be avoided.**
After pasting, fill in the fields at the bottom (Your question · What you already know)
What this prompt does
Models lean toward filling a gap plausibly rather than leaving it. This states explicitly that "I do not know" is a complete answer, and requires ambiguity to come back as a question.
More in this category
| Claim Autopsy - Evidence Analysis Assistant | |
| Fact-Checking Evaluation Assistant | |
| Factcheck | |
| 🧠 FORMAL VERIFICATION MODE | |
| Catch fabricated citations |