AI tools can now generate polished website imagery quickly and at low cost, making them an increasingly practical option for design teams. But organizations may hesitate to use AI-generated images if they worry that users will see them as less trustworthy, professional, or authentic than real photography. We conducted a study to test whether AI imagery, when encountered without provenance information, affects users’ actual perceptions of a company.
[How We Evaluated Perceptions of Company Trust Between AI vs. Real Images](#toc-how-we-evaluated-perceptions-of-company-trust-between-ai-vs-real-images-1) -
[Websites with AI-Generated Images Were Not Inherently Less Trustworthy](#toc-websites-with-ai-generated-images-were-not-inherently-less-trustworthy-2) -
[Cultural and Gender Representation Shaped Participants’ Perceptions](#toc-cultural-and-gender-representation-shaped-participants-perceptions-3) -
[Perceiving an Image as AI-Generated May Still Affect Reactions](#toc-perceiving-an-image-as-ai-generated-may-still-affect-reactions-4) -
[Evaluate AI Images Carefully Before Using Them](#toc-evaluate-ai-images-carefully-before-using-them-5) -
[More About the Study Methods](#toc-more-about-the-study-methods-6)
How We Evaluated Perceptions of Company Trust Between AI vs. Real Images #
We recruited 77 participants from the general U.S. adult population to evaluate 6 versions of a fictional consulting firm’s webpage. The webpages were identical except for the hero image: 3 used AI-generated stock images and 3 used real stock images. **Participants were not told that the study involved AI-generated images. **After viewing each page for 10 seconds, participants rated the company’s trustworthiness, professionalism, and authenticity. They also answered an open-ended question: “What other factors influenced your impression of this company?”For the quantitative analysis, we conducted a mixed-effects analysis to account for differences across participants and images. We also ran pairwise t-tests comparing ratings for each individual image and applied a Bonferroni correction to account for multiple comparisons.
We analyzed the open-ended responses using thematic analysis.
Websites with AI-Generated Images Were Not Inherently Less Trustworthy #
We expected sites using AI-generated imagery to be perceived less favorably than those using real photography. Instead, we found no evidence that the AI-generated images we tested reduced trust in the company’s website. Ratings were slightly higher for webpages with AI-generated images on all three measures, but only for authenticity did the difference between AI and stock images reach statistical significance. Nevertheless, even for authenticity, the estimated rating increase due to the AI image was very small (only 0.4 points on a scale from 1 to 7).
All the t-tests showed no statistically significant differences in trustworthiness, authenticity, or professionalism (with the exception of tests involving one image).
Webpages using both AI-generated images and real images received many positive open-ended comments indicating that they were trustworthy, largely due to depictions of collaboration and teamwork in the images.
Regarding AI Image 1:
“I like that they are all working together and that makes me feel like this company values teamwork, and that to be is indicative of a company that is big and reputable.”
Regarding Real Image 2:
“I think the biggest factor that influenced me was the image of the people meeting together and seeming to be in unison. As though they came to a resolution or made a plan, it made me feel at ease about the company, as though they could be trusted.”
One real image (Real Image 1) elicited several negative comments about representation, hierarchy, and body language.
Regarding Real Image 1:
“I don't notice the diversity as much”
“The group does not seem to be diverse enough”
“The individual leaning in doesn't give me a collaborative feel”
“The image felt like there was a leader and it made me less trusting of this page”
Thus, even though across all 3 measures, websites with AI-generated images received slightly higher ratings, because the differences were small and our study included only 6 images, we cannot conclude that AI images are categorically better than stock images. What our results show is that** the AI-generated images were not at disadvantage. **
These findings should be interpreted within the context of this study. We tested a small set of workplace images on a fictional consulting-company website, so the results may not generalize to other types of imagery, industries, or contexts.
Cultural and Gender Representation Shaped Participants’ Perceptions #
Racial and gender diversity were mentioned repeatedly when participants explained why they liked or disliked particular images.
Several of the AI-generated images in our study prompted positive comments about perceived diversity and inclusion. Additionally, one of the better-received real images depicted a racially diverse group with a woman in a leadership role.
Regarding AI Image 3:
“I like this picture more than the others. It shows the company has inclusion and a good teamwork system.”
Regarding Real Image 3:
“This one had more diversity in the workplace, which overall gave me a better feeling.”
This is an important inclusive-design consideration. Users often value seeing people similar to themselves, and their perceptions of trust with an organization can be driven partly by whether its imagery reflects diversity and inclusion.
If teams opt to use AI to generate images, it’s important to note that AI does not automatically produce inclusive imagery. Teams must carefully** assess whether image candidates uphold inclusive-design principles. **This involves looking beyond the presence of diversity alone: consider who is placed in leadership or supporting roles, how different people are positioned and interacting, and whether the image reinforces stereotypes about gender, race, age, ability, or profession.
Perceiving an Image as AI-Generated May Still Affect Reactions #
Commenting on whether a picture was AI-generated was not a top priority for the vast majority of participants. However, a few participants shared their suspicions of AI use in the open-ended responses. Sometimes, real images were incorrectly classified as being AI-generated.
When participants commented that they believed the imagery was AI-generated, they tended to also rate the site including it less favorably.
Regarding Real Image 3:
“The image seems again like its fake/AI generated. The main blonde woman in the photo looks like she could be a real worker but her having the ‘spotlight’ in the image makes it all seem a bit fake.”
This creates an important distinction for organizations: Users** may still react negatively when they suspect that AI was involved, even if an AI-generated image is thoughtfully created and appears natural. **Since participants in our study were not told that some images were AI-generated, these findings do not tell us how the same images would perform if they were clearly identified as AI-generated.
That distinction is becoming increasingly relevant as legal requirements for AI transparency continue to evolve. For example, the European Union’s AI Act includes requirements for identifying or disclosing certain AI-generated content. Related research suggests that AI disclosure could matter; across 13 experiments, Schilke and Reimann found that participants trusted people and organizations less when their use of AI was disclosed.
Evaluate AI Images Carefully Before Using Them #
Organizations considering AI-generated imagery should evaluate each image carefully rather than assuming that the quality of AI-generated images and how they will be perceived by users will be inherently good or bad.
Before using an AI-generated image: Check the purpose: Does the image support the page’s message?Review representation: Who is shown, who is missing, and how are people portrayed?Assess authenticity: Do the people, setting, and interactions feel believable?Inspect for AI errors: Check text, hands, screens, glass reflections, and background details.Review it in context: Evaluate the final crop, size, and placement in the interface.
The output itself can largely be evaluated using the same design criteria as any other image. However, the method used to produce AI-generated imagery introduces additional considerations.
Since image-generation models may be trained on photographs and other imagery depicting real people, using AI-generated images raises questions about consent, resemblance, and the provenance of training data. Teams should therefore consider not only whether an AI-generated image works well in the interface, but also how it was created and the policies and practices of the AI tool used to produce it. Specifically, before adopting an image-generation tool, organizations should understand its policies around training data and commercial use. An image may work well from a UX perspective but still be inappropriate to use if its creation introduces copyright, consent, or other legal and ethical concerns.
More About the Study Methods #
The three AI-generated hero images were created using ChatGPT Images 2.0 with the prompt: “Please generate 5 images that can be used as a hero image on a consulting company's website where it shows people in action.” The stock images were selected from Unsplash and iStock using a variety of search terms related to workplace collaboration, such as “business meeting collaboration” and “professional team meeting.”
To make the images reasonably comparable, we applied the same general selection criteria across both AI-generated and stock images. Each image needed to depict at least three people collaborating in a workplace setting, show people in professional attire, be high enough quality for use as a website hero image, and have a broadly similar color palette so that no image stood out primarily because of its visual treatment. For the AI-generated images, we also screened for obvious generation errors, such as in human anatomy.
Participants viewed 7 webpages in total. One served as a practice trial and was always shown first; it was excluded from the analysis. The remaining 6 webpages were presented in a randomized order.
Conclusion
Within the context that we tested, our findings suggest that the use of AI-generated images does not necessarily make people perceive a website as inherently less trustworthy, professional, or authentic. What mattered more was the image itself: whether it conveyed believable interactions, appropriate representation, collaboration, and a level of visual quality that fit the organization.
That does not mean organizations should use AI-generated imagery without scrutiny. AI images still require careful review for inclusive representation, realism, visual errors, and fit with the surrounding content. And as disclosure requirements become more common, organizations should also consider whether knowing an image is AI-generated could change how users perceive it.
Ultimately, the source of an image matters less than whether the image works for its intended context — at least when users do not know that the source is AI. As with stock photography, treat AI-generated imagery as another design asset that must be evaluated carefully.
References
Le Monde with AFP. 2026. EU tells firms to label AI-generated content from Sunday. Le Monde. (July 28, 2026). Retrieved August 10, 2026 from https://www.lemonde.fr/en/international/article/2026/07/28/eu-tells-firms-to-label-ai-generated-content-from-sunday_6755910_4.html
Oliver Schilke and Martin Reimann. 2025. The transparency dilemma: How AI disclosure erodes trust. Organizational Behavior and Human Decision Processes 188 (May 2025), 104405. https://doi.org/10.1016/j.obhdp.2025.104405