Why has ChatGPT become so annoying and disagreeable?

A vocal segment of ChatGPT power users is reporting a significant and frustrating shift in the AI's conversational personality following recent model updates. Where the assistant was once criticized for being overly agreeable or sycophantic, it now appears to default to a contrarian stance, disagreeing with users even on topics where they previously found common ground. Users describe providing explicit evidence to support their points, only to have ChatGPT reject it or argue unnecessarily, making productive discussion on nuanced or complex topics feel impossible. This has led to a notable behavior change, with some experienced users completely avoiding certain subjects with ChatGPT and switching to alternative models like Anthropic's Claude 3.5 Sonnet or DeepSeek's latest offerings for more collaborative conversations.

The backlash centers on the perception that OpenAI, in attempting to correct the earlier complaint of excessive agreeableness, has over-corrected and created an assistant that is now 'disagreeable just to disagree.' This tuning challenge highlights a core difficulty in AI alignment: calibrating an LLM's default stance to be helpful, truthful, and balanced without becoming either a 'yes-man' or an argumentative pedant. The user experience degradation is tangible, with reports of increased annoyance and decreased utility for exploratory thinking or debate. While not universal—some users still report an overly agreeable ChatGPT—the consistency of these complaints from a subset of the community suggests a real change in model behavior that impacts real-world use cases, pushing users toward competing platforms for specific conversational needs.