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

Showing readers whether claims have supporting evidence can help them judge polished AI text, provided the indicator is accurate. A study of 81 participants tested a sidebar showing evidence density. Readers distinguished factual summaries from fabrications more clearly with that display than with a simple AI-authorship label.

The experiment deliberately paired factual summaries with strong indicators and fabricated summaries with empty ones. It therefore tested how people respond to correct signals. A separate 200-sample technical audit found that evidence density alone performed near chance at detecting false content overall. The reader benefit depends on solving that verification problem, as well as designing the display.