As generative AI makes polished prose cheap to produce, users can no longer rely on fluency as a proxy for truth. Researchers identify this failure mode as the "Fluency Trap," where users trust fluent hallucinations while simultaneously discounting accurate content once it is disclosed as AI-generated.
Introducing Provenance Density
The study argues that binary "Made with AI" labels focus solely on authorship disclosure but fail to show the underlying support for specific claims. To address this, the researchers propose "Provenance Density," an evidence-visualization interface that displays the concentration of verified claims within a text.
User Study and Technical Audit
In a user study involving 81 participants, an idealized Provenance Density interface produced a significant discernment gap between truth and fabrication (+4.15 points, d=1.82). In contrast, participants provided with no signal showed no detectable ability to discriminate between the two.
A technical audit of 200 samples further refined these findings:
- Retrieval density alone proved insufficient for reliable verification.
- The "Consistency Veto" was found to carry most of the discriminative signal on dynamic queries.
As AI-generated content becomes indistinguishable from human writing, the researchers conclude that effective transparency must evolve. The focus should move away from simple authorship disclosure and toward the visualization of verifiable evidence.
