Nielsen Norman Group ran a study recently testing whether users could tell the difference, and more importantly whether it mattered to them, between AI generated images and real stock photography on a company website. They recruited 77 participants to rate a fictional consulting firm’s webpage, showing some people a version with AI generated hero images and others a version with real stock photos, without telling anyone which was which. The result: websites using AI generated images were not rated less trustworthy, less professional, or less authentic than the ones using real photography.
That’s a genuinely uncomfortable finding for anyone who’s assumed custom or real photography carries an inherent trust advantage over anything generated or stock. It doesn’t, or at least not one users can detect without being told. NN/g’s own framing is careful here, noting the study used a single fictional context and a specific type of workplace imagery, so it’s not a blanket claim about every use of AI imagery everywhere. But the direction of the finding is clear enough to take seriously: when a viewer doesn’t know an image is AI generated, they don’t treat it any differently.
The part of the study worth sitting with longer is what it implies in the other direction. If AI generated and real stock images perform equally well when unlabeled, then whatever trust a photograph was supposedly building was never really coming from the fact that it was a real photograph. It was coming from something else, familiarity, competence in composition, relevance to the content around it, none of which requires a camera.
That’s not an argument that photography or illustration style doesn’t matter. It’s an argument that generic imagery, stock or AI generated or otherwise, was probably never doing much trust-building work to begin with, and the industry habit of treating “real photo” as inherently more credible than “stock or generated” doesn’t hold up under actual testing. What NN/g’s other research on trustworthy design has consistently found instead is that comprehensiveness and consistency matter more than photographic authenticity, a site with thin or inconsistent imagery signals something is missing, regardless of whether the images used were shot on a camera or generated by a model.
Where this actually lands for anyone choosing imagery for a website isn’t “use AI generated images” or “avoid them.” It’s that the choice between real photography, stock, and generated imagery is a much smaller trust decision than most people assume, and a much bigger distinctiveness decision. A generic photo of people in an office doesn’t build trust because it’s a photo. It fails to build trust, or interest, because it looks like every other generic photo of people in an office, whichever way it was made. If a viewer can’t tell, and the research says they mostly can’t, the image was never doing the trust-building work anyway.
If a viewer can't tell, and the research says they mostly can't, the image was never doing the trust-building work anyway.