By skyforbes I’ve told ChatGPT multiple times to stop using em dashes because they just don’t look natural to me. Every time it confidently says, “Got it, I won’t use them again,” and then two sentences later… there it is again.
That got me thinking about less visible instructions. If it can’t stick to a simple, obvious rule, what happens with ones that are harder to notice?
Things like tone, framing, or how it interprets small details in a prompt. For example, a factual constraint that quietly shifts.
How much of our instructions are actually being followed, and what might ChatGPT be quietly ignoring?
Would love to hear if others have run into this. What’s something ChatGPT just refuses to do for you, even after you’ve asked it multiple times?
I swear it’s like the AI version of “I promise I’ll change.”