Promise people they'll get to edit their chatbot interview afterward — does that make them open up more during the actual conversation?

This explores whether the promise of a later 'edit' step changes what people say to an interview chatbot while they're talking. The direct answer is that the corpus can't fully settle it. The closest study, with 188 participants, compared editing conditions after the interview. It tells us a lot about what happens once people start editing. It doesn't clearly show whether anticipating that step loosened their tongues during the conversation. What it does show is more surprising, and arguably more important: the editing safety net works very differently depending on how it's built.

When people edited their own transcripts by hand, they left a lot of identifying information in. When an AI flagged risky passages and suggested removals, significantly more personal information came out, and the transcripts stayed useful as research data Can AI suggestions help people remove their own personal information?. The gain came from two places: people noticed disclosures they hadn't registered as risky, and accepting a suggestion took less effort than spotting and rewriting things themselves. That fits a wider pattern. Writers who are handed AI-generated text change it only about 23% of the time, and their edits are usually tiny Do writers actually edit AI-generated text before publishing?. People rarely do much self-editing unless something prompts them to.

The striking finding is that comfort and protection came apart. Manual editors felt just as confident about their privacy as people who had AI help, even though they left far more identifiable detail behind Why do people feel safe sharing when they haven't actually removed sensitive information?. That is why your question matters. If a later edit step reassures people, it could encourage freer sharing during the interview. But the reassurance seems to come from simply having an edit step, not from how much it actually removes. A similar blind spot shows up with chatbot memory: most users never noticed that their assistant had stored or inferred private details until someone showed them, and then they all wanted editing and deletion controls Do users notice when LLMs infer and retain private information?.

The likely reason disclosure runs high in the first place has little to do with editing. Machines remove the social pressures that normally hold people back, such as saving face, managing impressions, and fear of being judged. That makes people more direct and more willing to share sensitive things Why do people share more openly with machines than humans? Do chatbots help people disclose more intimate secrets?. Chatbots that share emotions of their own also draw deeper disclosure back, much as people do with each other Do chatbots trigger human reciprocity norms around self-disclosure?. A related experiment offers a useful lesson: telling people they were talking to an AI changed nothing, but telling them what the chatbot was trying to do cut its persuasive effect roughly in half Does telling people they are talking to AI change how persuaded they become?. By analogy, saying 'you can edit later' may matter less than showing people concretely what in their answers is risky.

The takeaway: people already overshare to chatbots for social reasons, and an edit step mostly helps after the fact. It only does real work when it points out the risks for them. Left to themselves, people feel protected without being protected. Whether expecting to edit makes people share more in the moment is still an open question here, and a study comparing interviews with and without an announced edit step would answer it.