Beyond “Made with AI”: Visualizing Evidence Density to Improve Transparency

Summary

Generative AI has made polished prose inexpensive, creating what the authors call the “Fluency Trap”: users may trust fluent hallucinations while discounting accurate material after learning it was AI-generated. The paper argues that binary “Made with AI” labels disclose authorship but do not show which claims are supported by evidence. It proposes Provenance Density, an evidence-visualization interface that displays the density of verified claims within a text. In a user study of 81 participants, an idealized version produced a +4.15-point discernment gap between truth and fabrication, with a large effect size of d=1.82; participants given no signal showed no detectable discrimination. A technical audit of 200 samples found that retrieval density alone was insufficient, while a Consistency Veto provided most of the discriminative signal on dynamic queries. The authors conclude that effective transparency for AI-generated content should move beyond authorship disclosure toward visualization of evidentiary support.