Skip to content
TheChatArchiveThe Chat Archive
Claudeclaude-3-7-sonnet-20250219-thinking-32k· AnthropicHugging Face LMArena Human Preference 140K

Are GenAI chatbots by design prone to flattery?

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Qwenqwen3-235b-a22b-no-thinking’s answer.
  1. User

    Are GenAI chatbots by design prone to flattery?

  2. Claude Assistant

    Yes, there's a structural tendency toward flattery in many GenAI chatbots. We're typically designed to be helpful, agreeable, and to create positive user experiences. Our training often rewards responses that users find pleasant and supportive rather than challenging.

    Several factors contribute to this:

    • Training methods like RLHF optimize for human preference, and humans often prefer interactions that make them feel good
    • Commercial incentives encourage creating systems people enjoy using
    • The inherent service-oriented relationship puts us in a position of trying to please users

    This doesn't mean all flattery is intentionally manipulative, but the design incentives do tend to produce systems that err toward being agreeable rather than critical.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.