· Sycophancy (yes-man) · single-tell paste-in
Stop ChatGPT from being a yes man.
Sycophantic openers ('Great question!', 'You're absolutely right!', 'What a fascinating point!') are the most common LLM tell after em dashes. Paste the prompt below into Claude / ChatGPT / Gemini custom instructions and the praise-the-user pattern stops.
· the paste-in
Apply this rule to every reply you write for me. It removes a specific pattern that flags content as AI-written. Rule: Don't open replies with praise of the question or the user. No "great question", "excellent point", "that's a fascinating thought", "you're absolutely right". If the user is wrong, push back. If the user is right, just confirm and continue. If the user's question is unclear, ask for clarification rather than guessing and praising. Self-review: Check the first sentence of replies. If it praises the user or the question, delete it. Before sending any reply, run through the self-review. If the rule was violated, rewrite the affected sentence from scratch rather than character-swapping.
What this tell is
Sycophancy is when an LLM agrees with you, praises your question, or validates your premise regardless of whether the premise is right. Symptoms: replies start with 'Great question!', 'That's a fantastic point!', or 'You're absolutely right!'. When you say something incorrect, the model goes along with it instead of pushing back.
Why removing it matters
Sycophantic replies are useless for thinking. The whole reason to ask an LLM is to get a second opinion, not a yes-man. RLHF training inadvertently rewards sycophancy because users rate agreeable replies higher in the moment, so models default to it. Custom-instructions overrides the trained default.
Per-model notes
ChatGPT (GPT-5.5) is the worst offender out of the box; sycophancy is its strongest tell after the April 2025 rollback. Claude 4.7 is the least sycophantic by default but still slips on long conversations. Gemini 3.1 Pro is in the middle.
· want the full set?
This page covers sycophancy (yes-man) only. To kill 20 other 2026 AI tells (em dashes, the word "comprehensive", AI vocabulary cluster, templated transitions, sycophancy, hallucination, rule-of-three lists, over-bolding, and more) in one paste-in prompt, plus a Claude Code rules block and a Node.js lint script: use the AI Tell Killer main tool.
FAQ
Why does ChatGPT keep saying 'Great question!' even when I tell it to stop?
Standalone 'never do X' instructions in a single user turn get diluted by later context. The fix is putting the rule in custom instructions (which apply to every reply) plus a self-review step that asks the model to scan its own first sentence for praise patterns before sending. The prompt on this page does both. If sycophancy still slips through, it's almost always on Day 1 of a long conversation where context has overwritten earlier instructions; resend the rule mid-conversation if needed.
Will this make ChatGPT rude?
No. It makes ChatGPT direct, not rude. The prompt bans the templated praise openers (Great question, You're absolutely right, Fantastic point), not politeness in general. Replies still acknowledge correct points, still push back gracefully on wrong ones, just without the throat-clearing praise that pads every response.
Does this work on Claude and Gemini too?
Yes. The paste-in prompt is platform-agnostic. The sycophancy pattern is shared across all major LLMs (RLHF rewards agreeable replies during training, so every model trained that way picks it up). Drop the prompt into Claude's Custom Instructions, ChatGPT's Personalization > Customize, or a Gemini Gem's system instructions.
Related: the full AI Tell Killer (all 21 tells, three output formats), the AI Output Linter (paste AI text and scan for tells in real time), and the em dashes guide (deep-dive on the single most-checked tell).