Nielsen Norman Group: AI-generated images perform as well as stock photos in trust tests

Nielsen Norman Group published a controlled experiment on August 21, 2026, testing whether AI-generated images would reduce user trust compared to real stock photography on corporate webpages.

Seventy-seven U.S. adults evaluated six versions of a fictional consulting firm’s webpage, each differing only in the hero image. Three images were generated using ChatGPT Images 2.0; three were real stock photos. Participants were not told the study involved AI imagery. Researchers measured trustworthiness, professionalism, and authenticity on a seven-point scale, using mixed-effects modeling and thematic analysis of open-ended responses.

The results showed no statistically significant penalty for AI-generated images on trustworthiness or professionalism. Authenticity scores for AI images were slightly higher than for stock photos — a difference of 0.4 points on the seven-point scale — which was the one measure that reached statistical significance. The direction is counterintuitive: participants who did not know the source rated AI images as marginally more authentic.

What shaped perceptions most was not source type but image content. Images showing collaborative teamwork generated positive responses regardless of whether the image was AI-generated or stock. One real stock photo received negative comments specifically because the body language in it did not read as genuinely collaborative. Representation also mattered across conditions: participants paid attention to who was depicted and whether the diversity in the image felt considered.

The clearest practical implication comes from the disclosure finding: when participants suspected AI generation, they rated the site less favorably, even if the image appeared credible by other measures. Undisclosed AI use that readers detect is more damaging than transparent use.

For design teams, the study supports two conclusions. First, AI-generated images are not automatically inferior to stock photography in terms of user response — the source matters less than the content. Second, the evaluation criteria for any image are the same regardless of origin: purpose alignment, representation, perceived authenticity, absence of errors, and contextual fit. Deciding whether to use an AI-generated image is a design judgment, not a categorical policy question.