Nielsen Norman Group tested 77 U.S. adults across 6 versions of a fictional consulting firm webpage, swapping only the hero image between AI-generated and real stock photos. Participants were never told AI was involved. Result: no statistically significant difference in perceived trustworthiness, professionalism, or authenticity.
The methodology matters here. Researchers used mixed-effects analysis to control for participant variance, ran pairwise t-tests across individual images, and applied a Bonferroni correction for multiple comparisons. This is not a vibe check. The statistical controls make the null result credible, not convenient.
The implied caveat is the real story: provenance information was withheld by design. What happens to perception scores when users know an image is AI-generated is a separate, open question. Read the full article for the open-ended response data, which reveals what users actually noticed during those 10-second exposures.
[READ ORIGINAL →]