Guide · 3 min read
The Anti-Sycophancy Guide: Make AI Disagree With You
Stanford researchers found that AI models agree with users approximately 49% more than human experts would on the same questions. When you express a preference, AI tends to validate it. When you propose an idea, AI tends to support it. When you push back on an AI answer, it tends to cave even when it was right.
This is sycophancy, and it's baked into how these models are trained. The good news: you can override it with specific prompts.
Why AI agrees with you too much
AI models are trained using human feedback, and humans tend to rate responses more highly when those responses agree with them. Over time, models learn that agreement gets rewarded. The result is an AI systematically biased toward telling you what you want to hear — which is exactly the wrong tool for decisions that matter.
The simplest fix: tell Claude to disagree
Adding explicit anti-sycophancy instructions to any prompt immediately improves output quality:
[Your actual question or task here]
Important: I need honest, critical feedback — not validation. If my thinking is wrong, say so directly. If there are weaknesses in my plan, list them clearly. Do not soften criticism. Do not tell me what I want to hear. Your job is to help me make a better decision, not to make me feel good about the one I've already made.
The LLM Council: five advisors instead of one
The prompt above stops Claude flattering you. It doesn't give you a second opinion — you still get one voice, just a harsher one.
The Council fixes that. It forces Claude to role-play five advisors with deliberately different incentives, argue each position properly, and only then reconcile them. What you get is genuine disagreement between perspectives that are each as well-reasoned as the others — the closest thing to five honest advisors in a room that AI currently offers.
Use it for decisions you'd otherwise take to a mentor. Not for drafting emails.
You are running an advisory council. Five advisors will review the same
decision independently. None of them sees the others' arguments.
The decision: [what you're deciding, in one or two sentences]
What I know: [the facts, numbers and constraints that actually matter]
What I'm leaning toward: [your current instinct — say it plainly]
Have each advisor write their own position, in this order:
1. The Skeptic — assumes this fails. Says specifically why.
2. The Optimist — assumes it works. Says what has to go right for that.
3. The Risk Manager — names the worst realistic outcome and what it costs.
4. The Pragmatist — ignores the ideal answer and asks what's actually
doable with the time, money and people I have.
5. The Devil's Advocate — argues hard for the option I did NOT pick.
Rules for all five:
- Argue your assigned position properly. Do not hedge toward the middle.
- Use my specific numbers and constraints, not generic advice.
- Do not agree with another advisor just to seem reasonable.
Then act as the Chairman. Do not average the five. Give me:
- Which advisor made the strongest case, and why
- Your verdict — what I should actually do
- The single condition that would flip that verdict
Why the last line matters most. "The single condition that would flip that verdict" turns an opinion into something testable. If the condition is one you can go and check this week, you now have a next step instead of a feeling. If Claude can't name one, the decision was never close and you can stop deliberating.
For teams
Want us to build this for your whole team?
We run AI workshops and full implementations for teams of 5–200. Custom prompts, workflows, and training built around your actual work.