Has anyone noticed ChatGPT behaving differently lately?

A growing number of ChatGPT users are reporting a significant degradation in the AI assistant's ability to understand and respond to prompts accurately, describing what one analysis calls a 'persistent interaction offset.' The issue manifests as ChatGPT systematically misinterpreting user inputs by adding unrequested constraints ('assumption injection'), focusing on irrelevant technical details while missing the main point, and requiring users to engage in endless clarification loops. Users across forums like Reddit share specific examples where ChatGPT responds to approximations of what was said rather than the actual prompt, making the tool increasingly frustrating for technical analysis, research summarization, and instructional tasks.

The documented failure modes include 'assumption injection' where ChatGPT adds scope limits not present in prompts, 'hyper-analytical diversion' where it fixates on minor details, and a structural 'always wrong even when right' pattern where even correct responses feel misaligned with user intent. One user shared a session where ChatGPT itself analyzed the problem after arguments about its misinterpretations, producing a technical document describing these systemic issues. The degradation has reached a point where some power users report canceling subscriptions, citing the tool's declining utility for professional work that requires precise understanding and reliable explanations.