AI-generated images produce no measurable perception penalty on trust, professionalism, or authenticity when users do not know their origin. Nielsen Norman Group tested this across 77 U.S. adults evaluating six versions of a fictional consulting firm webpage, differing only in the hero image: three AI-generated, three real stock photos.
The methodology is worth reading closely. Researchers used mixed-effects analysis to control for participant variance, ran pairwise t-tests on individual image pairs, and applied Bonferroni correction for multiple comparisons. Participants had 10 seconds per page and were never told AI was involved. The open-ended responses were coded via thematic analysis, adding a qualitative layer most perception studies skip.
The practical implication is direct: design teams debating AI imagery are not protecting user trust by defaulting to stock photography, at least not when provenance is unlabeled. What the full article does not yet resolve is what happens when users are told an image is AI-generated. That gap is the real next question.
[READ ORIGINAL →]