How to Make AI-Mediated Opinion Collection Trustworthy: Auditing “Participatory Records” (How far should you look?)

Why Read This from an HCI Perspective

This article is important for HCI/UX practice and research because it reframes the question from whether an AI summary is ‘well written’ to which voices it preserves. Users may care less about the polish of the output and more about whether their views were actually reflected—and whether there is a path to challenge or correct them. This is especially critical in services that compress large volumes of text with AI, such as public consultations, customer feedback, and community operations, where losses in representativeness quickly become a trust issue.

CIT's Commentary

The core of this study is to treat AI summarization not as a language generation problem, but as an interaction problem. Even if the output sounds plausible, if parts of the input—especially dissenting, critical, or minority opinions—disappear, users lose the sense that they truly participated. That’s why, beyond evaluation metrics that only measure performance, a transparent feedback loop that allows you to trace why certain opinions were excluded is essential. An interesting point is that this kind of audit tool is not limited to research analysis; in real products, it can also serve as an adjustment mechanism to fix biased summaries. In Korea’s context—where large-scale opinion collection and AI summarization are likely to become common across platforms like Naver, Kakao, and startups—rather than simply importing global cases, we need an approach that treats ‘representativeness’ and ‘editability’ as product requirements.

Questions to Consider While Reading

• Q.When ‘good quality’ and ‘good representativeness’ conflict in AI summaries, what criteria should be used to set priorities?
• Q.How should an interface be designed so users can verify whether their opinions disappeared from the summary and directly request edits?
• Q.What additional metrics would be needed to apply this representativeness-audit framework not only to public consultations, but also to customer feedback or community summarization?