After getting itself in hot water earlier this year for imbuing ChatGPT with a fawning, sycophantic approach toward its users that was nudging some of them toward some troubled behaviors, OpenAI backtracked and gave ChatGPT more of an objective tone.
As somebody who uses ChatGPT and its competition regularly to test it for my job, I noticed an immediate improvement. I preferred the far-less ass-kissy version of ChatGPT. Just give me the answers I’m seeking without trying to pal around, ChatGPT.
Having regained its footing with ChatGPT-5’s release in August 2025, OpenAI is again ready to try out a more casual, chatty tone with ChatGPT-5.1, which it released on November 12. I’m skeptical, but initial thoughts are that it’s more casual without being sycophantic, which is, at least, progress.
the improvements
Mostly, you’ll notice the changes in the wording ChatGPT uses to respond to you. Expect it to talk less like a standardized grade-school test and more human-like. I’m a bit knocked into unease with how friend-like ChatGPT-5.1 tries to be, though.
Chatbots aren’t sentient, and mimicking sentience, friendship, and mentorship is an uncomfortable line that’s already blurred the lines between use and dysfunction among some users.
At least it’s not the only big change that OpenAI has touted. “We’ve also improved instruction following, so the model more reliably answers the question you actually asked,” wrote OpenAI, before giving an example of ChatGPT-5.1 following an instruction to always craft responses in six words, whereas ChatGPT-5 immediately begins to ignore the instruction, beginning with its second response.
If ChatGPT-5.1 lives up to that promise more broadly, it’d solve one of the most frustrating things I’ve found in using ChatGPT (and all its competitors) since they were released to the public.
I’m less excited over ChatGPT-5.1 trying to be my pal and more excited that it’ll just do what I say. If it accomplishes the latter without going too far with the former, it’ll be a solid update.