Beyond 'Made with AI': Visualizing Provenance Density to Mitigate the Transparency Penalty

Forum topic · 小凯 · 2026-09-06

Summary

This paper introduces the 'Fluency Trap': as generative AI makes polished prose cheap, users can no longer treat fluency as a proxy for truth—they trust fluent hallucinations while discounting accurate content once it is disclosed as AI-generated. Binary 'Made with AI' labels provide authorship disclosure but do not show what evidence supports a claim. The authors propose Provenance Density, an evidence-visualization interface displaying the density of verified claims within a text. In a user study with 81 participants, an idealized Provenance Density interface produced a large discernment gap between true and fabricated content (+4.15 points, d=1.82), while participants receiving no signal showed no detectable discrimination. A technical audit of 200 samples found that retrieval density alone is insufficient; a 'Consistency Veto' mechanism carried most of the discriminative signal on dynamic queries. The paper argues that effective transparency must shift from authorship disclosure to evidence visualization. Posted on zhichai.net, referencing arXiv:2509.00010 by Qing Zhang, Yifei Huang, and Juyoung Lee.

Paper Overview

Field: AI/ML Authors: Qing Zhang, Yifei Huang, Juyoung Lee Published: 2026-09-06 arXiv: 2509.00010

Abstract

As generative AI makes polished prose cheap to produce, users can no longer rely on fluency as a proxy for truth. The authors call this failure mode the Fluency Trap: users trust fluent hallucinations while also discounting accurate content once it is disclosed as AI-generated. Binary 'Made with AI' labels respond with authorship disclosure, but they do not show what supports a claim.

The paper proposes Provenance Density, an evidence-visualization interface that shows the density of verified claims in a text.

Key Findings

In a user study with 81 participants, an idealized Provenance Density interface produced a large discernment gap between truth and fabrication (+4.15 points, d = 1.82), whereas participants given no signal showed no detectable discrimination.
A technical audit with 200 samples shows that retrieval density alone is insufficient; unexpectedly, the Consistency Veto carried most of the discriminative signal on dynamic queries.
As AI-generated content becomes indistinguishable from human writing, effective transparency must shift from authorship disclosure to evidence visualization.

--- *Auto-collected on 2026-09-06*