When collaborators can see exactly how AI was used in shared writing, do they welcome it or feel watched?

This explores how collaborators respond to seeing the details of each other's AI use, such as prompts, timing and how much text came from the model, rather than a one-line 'AI was used' label. The most direct evidence is reassuring. In a study of sixteen pairs of writers in shared editors, people preferred settings that showed more of their partner's prompting activity, not less Do writers want to see each other's AI prompts in shared editors?. Seeing when, where and how AI was used helped them follow their collaborator's thinking and check text the AI had produced. The same study found the cost: some people felt self-conscious, as if someone were reading over their shoulder, when their own prompts were fully on display. The corpus has only this one study on detailed logs, so treat the finding as a promising early result, not a settled one.

That self-consciousness has a basis. Across four experiments with more than 4,000 people, AI users expected colleagues and managers to see them as less competent and less diligent, and they were less willing to disclose their AI use as a result Do people fear judgment when they use AI at work?. Readers and writers also disagree about what needs to be disclosed. Readers consistently think disclosure matters more than writers do, especially when AI text was pasted in directly and couldn't easily be replaced, and how much effort the writer put in made no difference to their judgment Do readers and writers differ on AI disclosure necessity?. So the person creating the log and the person reading it judge it by different standards, and detailed logs show exactly the parts readers care about most.

Here is the twist you may not expect. The first reaction to visible AI use may not last. In studies of AI partners, people initially avoided a partner once they learned it was an AI, but that preference reversed after repeated rounds where they could see the results Does revealing AI identity help or hurt user trust?. Disclosure alone changed nothing. What recalibrated people was disclosure plus feedback on outcomes. That study was about AI partners, not human colleagues' logs, but the lesson plausibly carries over. A log that shows only prompts may trigger the competence penalty, while a log tied to visible good results gives collaborators the evidence they need to update their view.

Logs may also be more revealing than people assume. Process data shows a clear signature when someone hands a whole task to AI: the text arrives in concentrated bursts that don't match the author's normal rhythm. Lighter, back-and-forth assistance, though, looks almost the same as working with minimal help Can process data distinguish AI delegation from ordinary collaboration?. In practice, detailed logs can expose heavy delegation while making ordinary help look unremarkable. That may be part of why writers in shared editors found them useful for checking each other's work.

There is one more reason logs might be good for the person being watched. Research on the 'LLM fallacy' finds that when AI output is smooth and seamless, people start to believe they have skills they don't actually have Do AI-assisted outputs fool users about their own skills?. A visible record of which work came from whom pushes back against that self-deception as well as informing the collaborator. If you want to explore the design side, work on AI thought partners argues that legibility, meaning each party being able to read the other's reasoning, is a core requirement for real collaboration What makes an AI a true thought partner, not just a tool?.