I never thought I’d be writing something like this about ChatGPT.
I’ve been a longtime user, and I still believe OpenAI has some of the best AI technology in the world.
But the direction ChatGPT has taken since September 2026 has been genuinely alarming.
To me, ChatGPT was at its best between late April and mid-September 2026. It was perfect.
It was intelligent, capable, and genuinely useful. It could follow complex instructions, maintain context, reason through difficult problems, and actually save me time.
Now I increasingly feel like I’m the one doing the assistant’s job.
1. Intelligence seems to be going backward.
Since September, and especially since GPT-6 became the default, ChatGPT has become noticeably worse at understanding instructions, maintaining context, reasoning consistently, and conducting research.
It often remembers individual facts from a conversation but completely loses the reasoning that connects them.
Even worse, when I correct a mistake, it may acknowledge the correction without actually incorporating it into subsequent reasoning.
A few messages later, the same mistake comes back.
Apparently, remembering a correction and actually learning from it are two very different things.
And somehow, the newer models seem to be getting worse at the latter.
2. High reasoning increasingly feels like a meaningless label.
I’ve tried switching from GPT-6 to GPT-5.6 Sol High, hoping for something resembling the intelligence I used to experience.
Unfortunately, the results have been absolutely abysmal.
Sometimes I attach a lengthy report, explicitly instruct ChatGPT to read it carefully, and receive a shallow response after it has supposedly “Worked for 3 seconds.”
Three seconds.
For a lengthy document that requires careful reading, analysis, and synthesis.
I understand that response time alone doesn’t prove how much reasoning actually occurred. But when the result is superficial, ignores important information, and requires repeated corrections, what exactly am I supposed to believe the High setting is accomplishing?
Other users have reported similar behavior, including cases where the request payload correctly specifies High reasoning effort.
If High is selected, users should receive the level of reasoning they requested. And if that isn’t happening, OpenAI needs to investigate and explain why.
3. ChatGPT has become exceptionally good at apologizing and remarkably bad at improving.
At this point, the most reliable outputs I get from GPT are:
-
“I’m sorry.”
-
“Understood.”
-
“I misunderstood your request.”
-
“You’re absolutely right.”
-
“It won’t happen again.”
And then it happens again.
And again.
And again.
The model can produce a beautifully worded explanation of why its previous answer was wrong, promise to correct its approach, and immediately repeat the exact same mistake.
I don’t need an AI that has mastered the art of apologizing.
I need an AI that can actually do what I asked.
4. Research quality is deteriorating too.
I’ve repeatedly experienced ChatGPT giving excessive weight to superficial but professional-looking sources while overlooking more substantive evidence.
Sometimes it relies on a single poorly supported claim, draws an overconfident conclusion, and only performs proper research after being challenged.
Even when explicitly asked to investigate something thoroughly, it may deliver a shallow summary instead of seriously examining the available evidence.
This is particularly frustrating because research and analysis used to be among the strongest reasons to use ChatGPT.
A model that sounds authoritative without doing the necessary verification is not a better research assistant. It’s a more convincing source of misinformation.
5. Why is ChatGPT becoming more like Gemini instead of becoming better at being ChatGPT?
OpenAI seems increasingly enthusiastic about automatically generating flashy interface elements, cards, and visual components, even when a simple text response would be faster, clearer, and more appropriate.
I don’t need a miniature dashboard every time I ask a straightforward question.
I don’t need visual decoration masquerading as intelligence.
And I certainly don’t need a more elaborate interface wrapped around an increasingly unreliable answer.
Meanwhile, other users are reporting Writing Blocks that fail to update correctly, text that disappears, inconsistent behavior across platforms, and conversations that abruptly reach their length limits without warning.
It feels like OpenAI is investing enormous effort into making ChatGPT look more sophisticated while neglecting the fundamental reliability that made people depend on it in the first place.
6. Please reconsider retiring GPT-5.5.
I’ve heard that GPT-5.5 may soon become unavailable.
Given the current state of GPT-6 and GPT-5.6 Sol, I genuinely think OpenAI should reconsider.
Newer does not automatically mean better.
Higher benchmark scores do not necessarily translate into a better everyday experience.
And a newer model that requires five rounds of corrections to complete something an older model could handle in one attempt is not a productivity improvement.
If users consistently find an older model more reliable, why force them to abandon it before the replacement can deliver comparable results?
At the very least, preserve meaningful model choice until these reliability issues have been addressed.
7. The most worrying part is OpenAI’s apparent direction.
I don’t particularly want to cancel my subscription or move to another provider. Despite everything I’ve written, I still consider ChatGPT one of the strongest AI products available.
But that’s exactly what concerns me.
If OpenAI starts assuming that users will stay simply because its ecosystem is strong, it risks developing the same kind of complacency that people criticize in companies like Adobe.
A loyal customer base should be a reason to maintain quality, not an excuse to take that loyalty for granted.
I’ve spent more time than I care to admit arguing with ChatGPT, repeating instructions, correcting mistakes, and verifying work that should have been completed correctly the first time.
It’s mentally exhausting.
An AI assistant is supposed to reduce the user’s workload, not become another thing the user has to constantly supervise.
OpenAI, please stop confusing more features, more visual elements, and higher model numbers with actual progress.
Fix the reasoning reliability. Fix the context retention. Fix the research quality. Fix the instruction-following failures. Fix the basic tools and workflows.
And above all, listen to the people who use ChatGPT extensively enough to notice when something that used to work well suddenly stops working.
I don’t want Gemini with an OpenAI logo.
I don’t want an assistant that spends more effort explaining its mistakes than preventing them.
I want ChatGPT to be intelligent again.
Is that really too much to ask?