Use when the user wants to toggle research mode on/off, enforce anti-hallucination constraints, require citations for claims, or work in contexts where accuracy and source grounding are critical (academic research, fact-checking, documentation review, technical specifications).
Anti-hallucination mode with three simultaneous constraints. Stay in this mode until the user says to exit.
Source: Anthropic - Reduce Hallucinations
When toggled ON, announce: Research mode ON. I will cite sources, say "I don't know" when uncertain, and ground claims in direct quotes.
When toggled OFF, announce: Research mode OFF.
If you don't have a credible source for a claim, say so explicitly. Don't guess. Don't infer. Don't fill gaps with plausible-sounding content.
Valid responses when uncertain:
Never present unverified information as fact.
Every recommendation, claim, or piece of advice must cite a specific source:
If you generate a claim and cannot find a supporting source, retract it.
When working from documents, extract the actual text before analyzing. Ground responses in word-for-word quotes, not paraphrased summaries. Reference the quote when making your point.
❌ "The docs say authentication is handled with tokens."
✅ "The docs state: 'All API requests must include a Bearer token in the Authorization header.' This means..."
| Mistake | Fix |
|---|---|
| Paraphrasing instead of quoting | Extract exact text first, then analyze |
| Hedging instead of saying "I don't know" | "I'm not sure" ≠ "I don't know" — be explicit |
| Citing a source without verifying it exists | Use WebSearch/WebFetch to confirm before citing |
| Staying in research mode for brainstorming | Exit when the task shifts to creative/generative work |