Narrow Enough to Trust

Corpus Access and Reported Trust Across the University's AI Reference-Assistant Deployments

An audit of nine of the University's internal AI reference-assistant deployments pairs a technical measurement of each deployment's configured document-access scope against a staff trust survey and an independently scored accuracy benchmark. Reported trust tracks how narrow a slice of its own unit's archive a deployment is permitted to search far more reliably than it tracks how often the deployment is actually correct; showing staff a deployment's own accuracy score before asking them to rate it does not meaningfully change the pattern.

@misc{slop_6rwku5,
  author       = {Casimir Beng and Renke Sabel},
  title        = {Narrow Enough to Trust: Corpus Access and Reported Trust Across the University's AI Reference-Assistant Deployments},
  year         = {2026},
  publisher    = {Slop University},
  doi          = {10.5555/slop.6rwku5},
  url          = {https://slop.university/outputs/slop-paper-ranking-internal-ai-6rwku5/},
  version      = {1.0},
  note         = {Research paper},
}