When the model misbehaves

10 recurring failures, each with what causes it, the fixes published research actually supports, and the prompts to paste. No tricks, no banned-word lists, and an explicit note where a popular fix does not hold up.

ChatGPT makes things up Confident, well-formatted, and wrong. Citations that do not exist, numbers that were never in the source. 589 techniques ChatGPT is too verbose Four paragraphs of preamble before the one sentence you needed. 537 techniques ChatGPT loses the thread in long documents Accurate about the first pages and the last, vague about everything between. 421 techniques ChatGPT refuses a reasonable request A legitimate task gets read as a policy problem and declined. 212 techniques ChatGPT will not stick to the format you asked for You asked for JSON and got JSON wrapped in an apology and a markdown fence. 197 techniques ChatGPT ignores your instructions You state a rule, it holds for two messages, then quietly drops it. 117 techniques The same prompt gives ChatGPT a different answer every time Run it twice, get two different answers, with no way to tell which one to trust. 76 techniques ChatGPT output sounds like AI wrote it Grammatically perfect, structurally tidy, and recognisable from the first line. 72 techniques ChatGPT forgets what you told it earlier An hour in, it contradicts a decision you made at the start. 61 techniques ChatGPT gives generic answers Technically responsive, completely unusable. The same answer anyone would have got. 55 techniques