Lina Kluy, Barbara Deml · International Conference on Human-Agent Interaction (HAI) 2026 · 2026
DOI: 10.1145/3841580.3845668
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Disclosure of AI limitations, knowledge, and underlying architecture is a transparency mechanism that helps users calibrate expectations and trust-related behavior. However, it is unclear how users respond to such disclosures. We qualitatively analyzed 336 LLM disclosure segments from publicly accessible naturalistic human-LLM conversations. We coded LLM disclosure types, along with their preceding human triggers and subsequent human responses. For human responses, we found that explicit acceptance of LLM disclosure is rare, regardless of the trigger or disclosure type. Instead, users tend to explore further or redirect to a new topic. When users object, they do not object to the disclosure’s informative value but to its repetitive, formulaic delivery. We discuss implications for designing and operationalizing the efficiency of LLM disclosures as transparency mechanisms.
arXiv (Cornell University) 2026 · 0 citations
No comments yet — start the discussion below.