Cognitive biases and emotional triggers make AI-generated images highly deceptive

Emotionally charged AI-generated images may encourage social media users to respond before they critically assess whether the images are real, according to a study published in Computers in Human Behavior.

AI-generated images can now be produced quickly and cheaply, making it possible to create large amounts of realistic-looking material for social media. Some images are paired with emotional captions and distributed through networks of automated pages or content farms that seek attention, advertising revenue or user interaction.

Images may be particularly effective at attracting attention because they are processed quickly and can trigger emotional reactions before people examine details closely. A picture of an elderly couple, a neglected child or an apparently extraordinary achievement may prompt sympathy, anger or admiration. Those reactions can make users less likely to stop and ask whether the scene is authentic.

Márk Miskolczi of Corvinus University of Budapest in Hungary analyzed public Facebook material collected between March 1 and May 1, 2025. The sample consisted of 146 AI-generated images and 8,922 user comments.

Miskolczi selected 12 public Facebook pages that regularly posted suspected AI-generated images. The pages covered themes such as nostalgia, rural life, elderly people, spirituality, everyday hardship and family relationships. Images were first examined using an eight-step manual system that looked for features such as distorted anatomy, unusual facial details, floating objects, impossible reflections, malformed writing and unnatural textures. Images were also checked with an online AI detector, and only those receiving an AI-probability score of at least 60% were included.

Free daily newsletter

An initial database contained 11,547 comments. After comments believed to have been produced by automated accounts were removed, 9,082 remained. The researcher then excluded 160 comments considered incomplete or meaningless, leaving 8,922 comments for qualitative analysis. The comments were categorized, or coded, in stages to identify recurring topics, emotional responses and possible mental shortcuts involved in users’ reactions.

Five broad themes emerged in the images: emotion and nostalgia; empathy and compassion; extraordinary achievements and hobbies; birthdays and social problems; and religion and spirituality. The most common comment category was birthday wishes, with 1,980 comments (22.19%). Inspiring messages and encouragement accounted for 1,802 comments (20.20%). Religious content accounted for 1,490 comments (16.70%).

Many users responded as though the images represented real people or events. They expressed sympathy for older adults, shared memories of childhood, offered congratulations and posted religious blessings. The analysis suggested that several mental shortcuts, known as cognitive biases, may have contributed. Confirmation bias may lead people to accept images that fit their existing beliefs. Anchoring may make the first emotional reaction more influential than later inspection. Familiar themes may feel trustworthy, while large numbers of supportive comments may create the impression that the content has been socially verified.

The researcher found that images involving older people, nostalgic memories and apparent neglect often produced strong emotional reactions. Outrage-inducing content could also encourage comments while drawing attention away from visual inconsistencies. By contrast, some images involving inter-ethnic conflict or neglected birthdays attracted more skepticism or less engagement.

Miskolczi summarized: “This study examined how AI-generated images [AIGIs] on Facebook exploit users’ vulnerabilities, revealing their remarkable potential for digital deception… Emotional engagement was often accompanied by cognitive biases which together reduced users’ critical reflection.”

Some limitations are to be noted. The study was limited to public Facebook content, which may not reflect platform-specific or culturally embedded differences in how users on other networks engage with such images. Additionally, the manual frameworks used to detect AI images and automated comments have not yet been externally validated. Furthermore, the analysis inferred cognitive biases from the content of comments rather than measuring them directly in experiments. It also cannot establish exactly how many users genuinely believed the images were real, how many were responding ironically or whether the images changed their views or behavior.

The study, “The illusion of reality: How AI-generated images (AIGIs) are fooling social media users,” was authored by Márk Miskolczi.