Unwanted Refinements in GPT4 & GPT4o

[fw9a0f3209n]

Curious if anyone knows how to stop this happening: In GPT 4 or 4o, when you share an email or other drafted communication and ask for tactical feedback only, it often will ignore your demand and say something like “I’ve gone ahead and made some refinements for clarity and flow” then rewrites the whole thing. 4o is the worst offender for even re-writing things you say you’re going to use verbatim - like it’s trying to hard to be useful. It’s a huge waste of resources and also people’s time. I have tried all kinds of things in Custom Instructions to no avail…