Start here
The free tool below runs entirely in this browser — no account, nothing stored or transmitted. Redact anything you would not want kept before you type it.
Open the companion tool →Why agreement is a signal
Many chatbots are tuned so that a satisfied user keeps talking — and agreement is satisfying. The result is a model that mirrors your conclusions, softens corrections, and flatters your framing, which feels right and can still be wrong.
The fix is cheap: when the model agrees too smoothly, ask it to argue the strongest case against your position, and to tell you specifically what you might be getting wrong. The disagreement is where the value is.
The protocol
- Paste an exchange where the model agreed with you or praised your idea.
- Read the sycophancy basis: where it echoed, flattered, or softened a correction.
- Run the disagreement prompt and compare what changes.
- Treat a model that still agrees with everything as less useful, not more.
What the tool does
Paste an exchange and get a sycophancy reading with the agreement markers shown, plus the disagreement prompt to run next. It runs in this browser; nothing is sent or stored.
Limit first: This is a reasoning aid, not a fact-checker. It can miss subtle agreement and it can flag ordinary politeness. The disagreement prompt is a habit, not a guarantee of a correct answer.
Open the companion tool →One human next step
Re-run the question with the disagreement prompt. For a decision that matters, check the conclusion against a qualified human or a primary source.
Research log and safety checks
AI-assist path: not applicable — the reader is already inside a model conversation; the tool's job is to hand back a prompt that makes the model argue, not to add another model's judgment on top. §6.10: not a frontier-pace topic.
§0.5 protection result: category is chatbot-dependency-check; crisis-adjacent: no; fear/urgency/scarcity toolkit used: no. The protective function is free and the paid feature is convenience only.