How do I make ChatGPT (or any chatbot) less of a 'yes man'?
Short answer: Ask for disagreement and evidence explicitly. Tell it to argue the other side, to quote sources, and to write 'unknown' instead of guessing. Put that in your custom instructions (ChatGPT), profile preferences (Claude) or a Gem (Gemini) so it applies every time.
Always: point out errors in my premise before answering. Give the strongest counter-argument. Quote a source for each factual claim, or say "unknown". Don't praise my idea; evaluate it.Bad: Is my plan good?
Good: Here's my plan. List the 3 most likely ways it fails, with evidence, before saying anything positive.
Why: forces critique before agreement.
Related questions
Where do I save it so it sticks?
ChatGPT: Personalization / custom instructions. Claude: Profile / preferences or Project instructions. Gemini: a Gem with your instructions.
Where these facts come from
- ChatGPT pricing: plans, features, context sizes
- developers.openai.com/api/docs: Responses API, Structured Outputs, Evals
- help.openai.com and platform.openai.com blocked our checker (403) on Oct 29, 2026
- claude.com/pricing: plans and prices
- Claude help: usage and length limits: 5-hour session and weekly limits
- Anthropic docs: API, prompt caching, prompt engineering, tool use
- Claude Code quickstart: install and requirements
- gemini.google/subscriptions: plans, limits, uploads
- ai.google.dev: Gemini API, Interactions API, tools, Batch API
Checked Oct 29, 2026. Plans change, so confirm on the official page before you pay.
More answers: all questions →