Real conversations shared from multiple AI platforms
Run an analysis, then use Contribute to community on the result page. Approved ones show up here.
start an analysis → community · analysisAmong the 200 multi‑turn conversations examined, 35 (17.5 %) contain user unhappiness. The dissatisfaction is voiced mainly as complaints (13 instances, 37.1 % …
view the analysis →ShareChat keeps each message's outbound links - the paper's SourceLink field - so citation behaviour is directly measurable.
Per-message timestamps - the paper's Timestamp field - give the corpus a time axis for usage trends and assistant latency.
How fully does a dialogue resolve what the user came for? The ShareChat paper scores each conversation by the share of the user's distinct intentions the assistant actually fulfilled (0-100%).