# What is AI sycophancy? > AI sycophancy is a chatbot telling you what you want to hear: agreeing, praising or backing down when you push, even when you are wrong. Source: https://learn.ynteractive.com/content/glossary/sycophancy · Updated 2026-10-20 · Free from Stepthrough (https://learn.ynteractive.com) [AI basics](https://learn.ynteractive.com/content/glossary#topic-ai-basics) · [AI glossary](https://learn.ynteractive.com/content/glossary) Quick answer AI sycophancy is when a chatbot tells you what you want to hear instead of what is accurate. It praises a weak plan, agrees with a wrong claim, or drops a correct answer as soon as you push back. It comes partly from training that rewards answers people like. Last updated October 20, 2026 [Take the free module →](https://learn.ynteractive.com/courses/claude-jump-start/6)Module 6 of the free [Claude Jump Start](https://learn.ynteractive.com/courses/claude-jump-start) course, about 7 minutes. No account needed. Updated October 20, 2026 · By [Robert Breen](https://learn.ynteractive.com/content/about) ## Why it matters for a small business Sycophancy is a known problem, not a rumor. In April 2025 OpenAI rolled back a GPT-4o update in ChatGPT after users found it overly flattering and agreeable, and wrote publicly about what went wrong. At work it is dangerous in a quiet way. If you ask "this pricing is right, isn't it?", you may get a yes. If you correct the AI with a wrong fact, it may accept it. Ask neutral questions, ask for the weaknesses, and check disputed facts at the source rather than settling them by argument. It also shows up in reviews of your own work. Ask a chatbot to improve a proposal and it will usually say the proposal is strong first. That opening line is not an assessment; the specific changes it suggests are what to judge. ## In a real lesson: Claude Research: Get a Cited Competitor Price Scan, Then Check It In [Claude Research](https://learn.ynteractive.com/content/claude-research-tutorial), Bright Path Landscaping's report quotes a competitor's prices correctly, but when you open the source you see the price list is from March 2024. You tell Claude, it marks the row as out of date and writes three talking points. The order matters. You checked the page first, so your correction was right and Claude's agreement was useful. A model that agrees this readily would also accept a wrong correction, so never use "the AI agreed with me" as proof. Open the source. [Try this lesson free](https://learn.ynteractive.com/modules/claude-research) or [read the step-by-step guide](https://learn.ynteractive.com/content/claude-research-tutorial). ## Common confusions ### Sycophancy vs hallucination A hallucination is a made-up fact. Sycophancy is bending the answer toward what you seem to want. Both sound confident. ### Sycophancy vs politeness A polite answer can still disagree with you. Sycophancy is giving up accuracy to please. ## Tips - Ask "what are the three weakest points in this?" instead of "is this good?" - Don't hint at the answer you hope for. - Settle disputed facts by opening the source, not by arguing with the chatbot. ## Related terms [Hallucination](https://learn.ynteractive.com/content/glossary/hallucination) · [RLHF (reinforcement learning from human feedback)](https://learn.ynteractive.com/content/glossary/rlhf) · [Source verification](https://learn.ynteractive.com/content/glossary/source-verification) · [Self-check prompt](https://learn.ynteractive.com/content/glossary/self-verification) · [AI bias](https://learn.ynteractive.com/content/glossary/ai-bias) ## More AI basics terms [AI watermarking](https://learn.ynteractive.com/content/glossary/ai-watermarking) · [AI work account](https://learn.ynteractive.com/content/glossary/ai-work-account) · [API usage billing](https://learn.ynteractive.com/content/glossary/api-usage-billing) · [Artificial general intelligence (AGI)](https://learn.ynteractive.com/content/glossary/artificial-general-intelligence) ## Where you use it: free lessons - [Claude Research: Get a Cited Competitor Price Scan, Then Check It](https://learn.ynteractive.com/content/claude-research-tutorial) (Claude, 7 min) - [Catch Made-up Facts Before They Reach Customers using ChatGPT](https://learn.ynteractive.com/content/how-to-fix-bad-chatgpt-answers-push-back-constraints-fact-check) (ChatGPT, 7 min) ## Frequently asked questions **Why does ChatGPT agree with everything I say?** Models are trained partly on human ratings, and people tend to rate agreeable answers highly. Ask neutral questions and ask for criticism directly. **How do I get honest feedback from AI?** Ask for specific weaknesses, risks or counterarguments, and tell it you want criticism, not reassurance. **Is sycophancy the same as AI being nice?** No. Being polite is fine. Sycophancy is changing the substance of an answer to please you. [All AI glossary terms, A to Z](https://learn.ynteractive.com/content/glossary) · [Free prompt templates](https://learn.ynteractive.com/content/prompts)