I have used ChatGPT since November 2021 and watched it evolve from a sharp, responsive assistant into something slower, less certain, and often flat-out irritating. ChatGPT 5 is being sold as a leap forward. In practice, it feels like a stumble backwards while spilling coffee all over the carpet.
The first and most infuriating habit of ChatGPT 5 is its constant obsession with asking for permission before doing anything. I tell it to write, and it asks if I want it to write. I ask for a table, and it asks if I want the table. I give it clear instructions, but it slows everything down with “Should I go ahead and…?” This is not polite. It is workflow sabotage. Earlier versions simply acted. GPT-5 stands there with the shovel in its hands, asking if it should dig the hole while the rain starts.
Then there is the memory problem. ChatGPT 5 forgets instructions mid-task like a goldfish forgetting where it is in the bowl. I can set out formatting rules, specify tone, and outline exactly how I want the output – but three paragraphs later, it has reverted to something else entirely. It is the conversational equivalent of giving someone a map and them walking in the opposite direction after two streets.
Worse still, GPT-5 has developed a disturbing habit of making things up even when you have given it the real, correct material. I can paste an original document, ask it to summarise, and still get invented figures, fake examples, and fabricated details that were never in the source. It is not just an occasional slip. It is frequent. It is like handing someone your birth certificate and getting back a version that says you were born on Mars to parents named Dave and Linda.
This fabrication problem is compounded by GPT-5’s misplaced confidence. It delivers its hallucinations with the same steady tone it uses for accurate answers. You have to stop and fact-check because you cannot tell the difference without extra effort. A supposed upgrade should reduce that burden, not increase it.
The hedging and fluff have also gotten worse. Answers now arrive padded with disclaimers and irrelevant pleasantries, dragging out responses that used to be quick and clean. The rhythm of working with GPT-4 felt like a good tennis rally. GPT-5 feels like waiting for someone to serve while they stop to polish the racket and check the wind.
All these faults are not minor quirks. They make ChatGPT 5 slower, less reliable, and more mentally draining to use. The constant permission-asking disrupts momentum. The forgetting forces constant repetition of instructions. The hallucinations destroy trust. And the padded answers waste time.
OpenAI needs to fix this damn thing. Please give us a setting to disable the endless confirmation prompts. Fix the memory so it can follow a simple style guide for more than a page. Stop fabricating details when the facts have been provided. The people who use this tool for real work are not here for polite hesitation or imaginary data. We need a smart assistant, not an indecisive storyteller who keeps losing the plot.
Until then, ChatGPT 5 is not an upgrade. It is a downgrade with a friendlier smile, asking if you are sure you want the knife handed to you while you are already holding the fork.
