Do writers want to see each other's AI prompts in shared editors?

The paper tests four levels of prompt visibility inside a shared text editor: sharing nothing, sharing a placeholder that indicates AI use, sharing details about how the resulting text was generated, and sharing everything, including how the prompt was formulated, in real time. Sixteen participants wrote persuasive essays in pairs under all four. The abstract reports "a strong preference for techniques that share more information about prompting activities" and concludes that collaborative editors "should share more information among writers on when, how, and where AI is used." The framing matters: tools like Google Docs and Overleaf already broadcast cursor movements, selections, and edits, but according to the introduction they do not share the prompt that was issued or where its output was used.

The discussion attributes the preference to awareness, and it splits that awareness into three parts. Seeing prompts being formed in real time told participants what a collaborator was doing at that moment. Seeing the prompt together with the original text selection showed how the resulting text was formed, which participants said helped them understand each other's thought process, built trust, and gave them chances to learn from each other. A comment marking AI-generated text signaled where AI was used, which mattered because participants wanted to indicate, and to know, which parts of the document needed additional verification.

The preference is not unanimous, and the paper says so. Some participants held that formulating a prompt should be "private," felt uncomfortable when everything was shared, and worried that their prompts would be judged. The authors also note that awareness of others' prompts "may have influenced each other's thinking in ways they did not appreciate," which some found distracting and annoying. The result is a tradeoff between awareness and self-consciousness, and the introduction anticipates it by noting that shared activity in editors can distract writers and make them feel self-conscious.

This sits alongside notes that treat the AI's contribution as hard to see. Do writers actually edit AI-generated text before publishing? describes AI text traveling onward with little revision. A marker showing where AI was used gives a co-writer a reason to check that text before it travels further. Do users truly own the AI-generated content they produce? describes a gap between claimed and felt authorship, and prompt sharing is one channel through which a collaborator could see the process behind a claim. This paper does not test that link. Can we measure prompt quality independent of model outputs? treats prompts as measurable, which fits the worry that visible prompts become judgeable.

The excerpt is silent on several things. It reports no effect sizes, no measures of essay quality or verification behavior, and nothing on whether trust actually changed. The benefits are what participants said they perceived. Sixteen participants writing persuasive essays in pairs is a small sample on one task, so the excerpt does not show that the preference holds for other writing tasks, larger groups, or working teams. What it supports is narrower: when writers in this setting could see more of a collaborator's prompting, most preferred it, and the cost they named was privacy and judgment rather than lost information. That points toward graded or opt-in disclosure, not an all-or-nothing default, though the excerpt does not test such designs.