Sharing your AI prompts with a co-writer helps most people, but why does it make some feel exposed?

This explores what it is about having your AI prompts visible to a co-writer that makes some writers feel exposed. The corpus has one direct study, and it is thin on the "why." In a study of sixteen writing pairs, Do writers want to see each other's AI prompts in shared editors? found that most writers wanted more prompt visibility. They valued knowing when, how and where their partner used AI, and it helped them verify AI-written text. But some found full sharing intrusive and self-conscious. So self-consciousness looks like the price of a feature most people like, not a reason to reject it. The study doesn't dig into its causes.

The adjacent notes suggest a few candidate causes. These are my inferences, not findings about co-writers. First, a prompt is a rough, one-sided artifact. How do prompts reshape the role of context in AI conversation? describes a prompt as a single static bundle of what you're saying, the context, and the role you're assigning. Human dialogue builds context cooperatively as it goes. A prompt log shows the half-formed instruction rather than the finished thought. Second, prompts can be judged. Can we measure prompt quality independent of model outputs? shows that researchers can score prompts on clarity, logic and more, independent of what the model returns. Once prompt quality is measurable, a partner reading yours could be grading it.

A visible prompt trail also shows how much of the text is yours. Do writers actually edit AI-generated text before publishing? found that writers edited AI-generated paragraphs only 23% of the time, and the edits kept 96% of the original. A prompt log lets a co-writer see how heavily you leaned on the tool. The stakes are higher because Does AI writing assistance change how readers perceive the writer? found that AI help shifts how readers see the writer, toward more confidence, extremism and perceived privilege. A collaborator who can see your prompts can check whether the polished voice on the page is really yours.

The design implication comes from a different corner of the collection. How can proactive agents avoid feeling intrusive to users? argues that welcome AI features need civility: respect for boundaries, timing and autonomy, not just intelligence. Transparency features seem to need the same care. Writers wanted more visibility, not maximum visibility, so the useful design question is who sees what, and when. It is not whether prompts are shared. The corpus doesn't yet include a study asking writers directly what makes them wince. The self-consciousness finding is one sentence in a paper about something else, and that gap is probably where the next good study sits.