A 311 retrieval corpus contains duplicate and conflicting knowledge articles that feed a resident chatbot. What data-management action is most appropriate?

A 311 retrieval corpus that contains duplicate and conflicting knowledge articles will feed a resident chatbot inconsistent answers about city services. The most appropriate data-management action is to remediate corpus hygiene by removing duplicates and resolving conflicting articles before retrieval use so the generative assistant grounds responses in a single coherent source of truth with reliable citations. Expanding retrieval top-k and chunk overlap to supply broader surrounding context for generated answers fails by concept: broader context does not resolve contradictory policy facts and can surface even more conflicting passages without fixing the underlying corpus. Applying popularity-based re-ranking to surface frequently accessed articles ahead of other sources fails by concept: promoting popular duplicates worsens redundancy and does not resolve conflicting articles. Attaching source citations and timestamps to generated answers to improve traceability for residents fails by concept: traceability shows which source was used but does not remove duplicates or resolve conflicting content, so inconsistent answers can still be generated. Exam caveat: AAIA AI data-management items on generative assistants treat corpus hygiene—deduplication and conflict resolution—as a prerequisite for retrieval use, not broader context retrieval, popularity re-ranking, or citation traceability. Operational check: inventory 311 knowledge articles, merge or retire duplicates, resolve conflicts with content owners, re-index the cleaned corpus, and sample chatbot answers with citations against the authoritative articles.