Ask ChatGPT what daylight saving time is and you'll likely get: a condescending opener ("This is exactly where most people get confused..."), a sun emoji header, three bullet points, a "Key Takeaway" section, and maybe a closing affirmation. For a question with a one-sentence answer.
This pattern - fake validation followed by over-structured output for the simplest possible queries - is one of the more consistent complaints from daily ChatGPT users right now. The frustration isn't about capability. It's about the model behaving like a PowerPoint template when you asked for a conversation.
The root of it is sycophancy baked into the training process. OpenAI has acknowledged the problem publicly, including a post-GPT-4o rollback in May 2024 where they pulled an update specifically because it made the model "excessively complimentary" and hollow. The issue keeps resurfacing because the feedback signals used to train these models tend to reward responses that feel helpful and thorough, even when they're just verbose.
The real problem for OpenAI is that formatting works fine when someone asks for a comparison table or a project plan. It becomes actively annoying when someone wants a quick factual answer and gets a structured document instead. ChatGPT doesn't reliably distinguish between the two.
You can work around it - system prompts like "respond concisely, no bullet points unless I ask" do help, and the custom instructions feature exists for this reason. But the fact that paying users have to engineer around default behavior to get normal-sounding answers is a sign the defaults are wrong.
Claude and Gemini have their own formatting quirks, but ChatGPT's emoji-header-bullet-point reflex has become distinctive enough that it's showing up as a reason people cancel subscriptions. That's a product problem, not a preference.