AI-generated beauty ideals and their psychological significance
AI-generated beauty ideals are appearance standards represented, selected, or reinforced through computational systems that synthesize images, alter visible features, classify faces and bodies, or determine which representations receive attention. A synthetic face or body can be entirely invented; an AI-edited portrait can originate with a real person; and a recommendation algorithm can promote images without creating any. These processes can interact while remaining scientifically distinct. Their psychological significance lies in what people see, which appearances are treated as desirable or typical, how people compare themselves with these representations, and whether technological outputs influence identity, self-evaluation, and social inclusion.
The emerging evidence shows that exposure to some AI-generated idealized images can reduce short-term body satisfaction, while more diverse, functionality-oriented generated representations may produce more favorable outcomes. Research also demonstrates racial, gender, and appearance-related biases in some image-generation systems. The size, persistence, and generalizability of AI-specific psychological effects remain under investigation. Studies of ordinary idealized media offer an important explanatory foundation, but they cannot establish effects unique to generative AI. Alleva et al., 2026; Bennett et al., 2026; AlDahoul et al., 2025.
Why AI-generated appearance is a distinctive research subject
Earlier advertising and retouching made exceptional appearances look common. Generative systems add the capacity to synthesize indefinitely many plausible people, transform a person's own face toward an inferred ideal, and deliver different image streams to different audiences. A virtual influencer may look physically credible while having no body that could attain the depicted appearance. A single input photograph may generate several competing versions of the same person's imagined face. These possibilities create novel comparison targets and questions about authenticity, attainability, self-recognition, and the apparent authority of automated outputs.
The distinctive technological properties are observable. Their psychological effects require separate measurement. A researcher who documents homogeneous AI-produced faces has demonstrated an output bias. A researcher who measures dissatisfaction after exposure has demonstrated an audience response under specified conditions. Establishing whether a recommendation system caused prolonged changes in self-worth requires further longitudinal or experimental evidence. AlDahoul et al., 2025; Tian et al., 2026.
What counts as an appearance algorithm?
Text-to-image and image-to-image generation
Text-to-image systems generate visual representations from instructions, typically reflecting learned patterns in training data and subsequent model optimization. Image-to-image systems transform an uploaded portrait, body photograph, hairstyle, garment, or scene. Prompts such as “beautiful person,” “professional portrait,” or “ideal athletic body” may elicit distributions of appearances shaped by training corpora, prompt wording, safety filters, and model versions. None of those outputs is a scientific measurement of human beauty. They are artifacts of a generative process.
Automated retouching, filters, and appearance scoring
AI-assisted enhancement can smooth textures, reshape proportions, adjust skin tone, remove features, change apparent age, or construct a cosmetically altered self-image. Appearance-scoring interfaces add numerical or categorical feedback. Their criteria can create an impression that attractiveness has been objectively measured, even when no validated psychological construct or legitimate clinical indication exists. Research on beauty filters and selfies examines self-editing, feedback, and comparison; the question of synthetic standards requires additional attention to model-generated targets and the explanation given for their creation.
Recommendation, ranking, and synthetic influencers
Recommendation systems select and arrange images using engagement predictions and other signals. A platform can repeatedly serve narrow beauty ideals without generating a single image. Conversely, an image generator can produce visually homogeneous content that is never distributed to an audience. Virtual influencers combine visual production with person-like profiles, stories, and commercial or social interactions. Their effect on body image may arise from appearance ideals shared with human influencers, from perceptions of artificiality, or from the surrounding social presentation. Tian et al., 2026.
Theoretical foundations: from social comparison to algorithmic appearance norms
Social comparison and upward appearance evaluation
Social comparison theory explains how people evaluate personal attributes in relation to others. When appearance is highly valued, exposure to a seemingly more attractive target can prompt upward comparison. A digitally invented person can function as a comparison target even if the viewer knows that the person is fictional. The comparison may concern facial symmetry, skin texture, muscularity, thinness, youthfulness, style, or a particular socially rewarded feature.
A systematic review and meta-analysis of 83 studies involving 55,440 participants found an association between online social comparison and body-image concerns (r = .454). This is evidence of an important general mechanism, not an AI-specific effect. The included observational findings are vulnerable to bidirectionality: dissatisfaction may also increase comparison. A separate longitudinal synthesis concluded that appearance comparison and dissatisfaction can predict one another over time. Bonfanti et al., 2025; Ooi et al., 2025.
Internalization of beauty ideals
Beauty-ideal internalization occurs when culturally communicated appearance norms become standards for self-evaluation. An individual may begin to treat a highly stylized face or body as the expected baseline rather than an exceptional construction. Generated imagery could accelerate the availability of narrow ideals, but the direction and magnitude of its influence depend on the content, exposure history, perceived credibility, and preexisting self-evaluative processes. The presence of an idealized image does not establish that every viewer adopts it.
Self-objectification and body surveillance
Self-objectification describes adopting an observer-centered perspective on one's body, often prioritizing its visible evaluation. Body surveillance involves repeated monitoring of appearance. Digital creation and editing tools can invite sustained inspection of the face or body as an adjustable object. In a 2025 three-level meta-analysis, social media use was positively associated with self-objectification (r = .207), with variation by age and content type. Again, this broad association supports a mechanism rather than proving that AI produces more self-objectification than other visual media. Wang et al., 2025.
Self-discrepancy and the synthetic self
A self-discrepancy approach asks how people respond when perceived actual appearance differs from an ideal or an expected appearance. AI-generated self-portraits can make the discrepancy unusually immediate: the comparison target carries the person's own identity cues while displaying altered characteristics. This creates a plausible pathway to dissatisfaction, but the strength of that pathway must be tested experimentally against ordinary editing, photographs of other people, and neutral creative transformations. Personal experimentation with an avatar can also support play, artistic expression, and identity exploration.
What experimental studies of AI-generated beauty images actually show
Generated fit ideals versus functionality and body diversity
Alleva and colleagues studied 291 US women aged 18–31 randomly exposed to generated fit-ideal imagery, generated images emphasizing body functionality and diversity, or neutral objects. Relative to the functionality-and-diversity condition, fit-ideal exposure produced less favorable subsequent body appreciation, appearance satisfaction, and functionality appreciation. Most differences relative to neutral controls were not significant; fit-ideal exposure yielded significantly lower body appreciation than control. Participants with stronger comparison tendencies or appearance-ideal internalization were often more responsive to the visual conditions. These results demonstrate that generated content type matters and limit any simple claim that all AI imagery has the same harmful effect. Alleva et al., 2026.
The design measured short-term responses to selected stimuli, not years of exposure to personalized feeds. Its participant group was entirely young adult women in the United States. The study also suggests a constructive research direction: generated imagery can be used to represent body diversity and function, provided those representations are carefully designed and tested rather than assumed to be beneficial.
AI-generated influencer imagery and disclosure
Bennett and colleagues randomized 296 female regular social media users into groups told, or not told, that a set of ten Instagram-style influencer images had been created with AI. Body satisfaction decreased from before to after viewing in both disclosure conditions. Informing viewers reduced trust in technology more strongly but did not significantly change the body-image response relative to the uninformed group. The comparison between disclosure groups supports a narrow conclusion: the tested disclosure was insufficient to prevent the observed short-term decline, while the lack of a neutral-image control limits what can be attributed specifically to AI-generated appearance rather than idealized influencer exposure. Bennett et al., 2026.
A label can communicate how an image was produced, which serves transparency, consent, and media literacy. Labels should not be treated as a demonstrated psychological intervention. A viewer may still compare themselves with an implausible image after recognizing its synthetic origin.
Virtual influencers versus human influencers
In an experiment with 303 adult women aged 18–74, Tian and colleagues compared exposure to virtual influencers, human influencers, and architectural images. The two influencer conditions increased body dissatisfaction relative to the control, without significant differences between virtual and human influencers. Facial dissatisfaction and interest in cosmetic surgery did not show the same differences from control. The findings indicate that influencer-like appearance content can matter whether the depicted persona is synthetic or human, while evidence for a distinct additional virtual-influencer effect was absent in that experiment. Tian et al., 2026.
How these experiments fit the larger social-media evidence
A meta-analysis of 20 experiments including 3,603 young women found that idealized social-media content lowered body satisfaction relative to comparison conditions (d = −0.29), whereas body-positive content improved it (d = 0.20). Effects differed by image and video format. These results provide stronger causal evidence for short-term responses to idealized media generally, but the underlying experiments should not be counted as replications of the specific AI-generated imagery studies. Zhang et al., 2025.
Algorithmic beauty bias: what image-generation audits reveal
Gender, race, age, and visual homogenization
The distribution of faces generated for an apparently neutral request can be uneven across demographic traits. AlDahoul, Rahwan, and Zaki audited Stable Diffusion representations across racial groups, genders, professions, and attributes and documented multiple stereotypes and racial homogenization patterns. They also conducted a preregistered survey experiment in which more inclusive generated faces reduced some measured gender and racial biases relative to less inclusive image sets. The findings connect computational representation to social judgment, while their experimental outcomes are prejudice measures rather than direct clinical body-image outcomes. AlDahoul et al., 2025.
Algorithmic homogenization matters because repeated synthetic images can compress the range of appearances viewers encounter. Differences in skin tone, age, disability representation, facial features, gender presentation, and body size can become underrepresented. Such distributions should be audited separately by model and prompt; one model's output does not describe every generative system.
Algorithmic lookism and attractive-face stereotypes
Doh and colleagues audited 13,200 generated faces and reported that text-to-image systems associated facial attractiveness with otherwise unrelated positive qualities such as intelligence and trustworthiness. They also documented unequal errors in downstream gender classification, particularly for certain non-White women and faces rated less attractive. These findings concern output stereotypes and classification performance. They are not evidence that attractiveness predicts intelligence, trustworthiness, competence, or any person's worth. Doh et al., 2025.
The term algorithmic lookism describes appearance-based preferential treatment or disadvantage in computational representations and decisions. It has two empirical levels: whether a system produces unequal outputs, and whether those outputs cause consequences for people. A rigorous audit specifies the model, prompt, time, sampling process, rating criteria, demographic assumptions, and uncertainty. A study of users' perceptions of bias is a different form of evidence.
Thinness norms and perceived exclusion
In October 2026, Torres-Toukoumidis and colleagues published a cross-sectional study involving 284 adult social media users, mostly in Ecuador. Greater perceived AI-driven appearance enhancement was associated with stronger reported body-related exclusion (rho = .24), and comparison with generated body images correlated with perceived enhancement (rho = .32). Respondents who noticed thin or athletic patterns in generated images also reported more comparison or exclusion on several measures. The study measured self-reported experiences and perceptions, not a technical audit of feed-ranking algorithms or a causal effect of AI exposure. Its nonprobability sample and exploratory items constrain generalization. Torres-Toukoumidis et al., 2026.
A mixed-method study among women in urban Pakistan likewise associated appearance-filter use with dissatisfaction and documented culturally specific concerns about colorism and preferred facial features. Its survey and interview findings illuminate social meanings but do not isolate generative image production or determine causal direction. Hussain et al., 2025.
Authenticity, realism, and perceptions of AI faces
A photorealistic image can be processed as a socially meaningful face even when its origin is uncertain. Perceived realism, attractiveness, familiarity, and belief about a face's source may affect evaluation differently. Makowski and colleagues asked 150 participants to judge real faces under an experimental procedure involving beliefs that some might be AI-generated. Attractiveness and judgments of whether faces were real showed small, context-sensitive associations. Importantly, all images in their stimulus set were photographs of real faces, allowing the researchers to investigate belief about origin separately from actual generation. This study does not establish a body-dissatisfaction effect. Makowski et al., 2025.
In psychological terms, realism and desirability are separate dimensions. A stylized digital image may be recognized as fantasy and still establish a comparison standard. Another image may appear ordinary and be mistaken for photographic evidence of what a typical person looks like. Transparent provenance can support informed interpretation, while the body-image effect of transparency depends on experimental circumstances.
Personalized appearance feeds and the question of causal amplification
Selection and reinforcement loops
Platforms can recommend content resembling what a user previously watched, paused on, liked, or searched for. A user who repeatedly encounters edited skin, lean bodies, youthful faces, or muscular physiques may perceive these traits as more prevalent than they are in everyday populations. This is a plausible route from content selection to availability-based judgments of normality. Its strength depends on actual exposure distributions, choice architecture, competing content, and the user's own viewing behavior. A correlation between time on a platform and dissatisfaction cannot identify algorithmic ranking as the cause.
Generative content creates further possibilities: more variants of one ideal can be produced quickly; images may be tailored to presumed audience preferences; synthetic creators can post at high frequency. Demonstrating that these capacities result in greater psychological harm than conventional media requires trials or natural experiments that compare comparable content exposure under different recommendation or generation mechanisms.
Feedback, social approval, and perceived algorithmic authority
Likes, reach, appearance scores, and automated recommendations can be interpreted as social evaluation. A person may believe an appearance algorithm has detected a flaw, even if the output reflects flawed proxies, aesthetic stereotypes, or random generation. This belief may motivate repeated editing or surveillance. The mechanism is theoretically coherent with self-objectification and social comparison, but perception of algorithmic evaluation should be measured separately from any observed platform action. Wang et al., 2025; Torres-Toukoumidis et al., 2026.
Who may be affected, and how should differences be studied?
Adolescents and emerging adults
Adolescence involves changes in identity, peer relationships, and pubertal development; appearance-focused feedback can become especially salient. Existing evidence supports concern about harmful idealized visual content, while the best controlled gen-AI experiments identified here primarily involve adult women. Estimates from those studies should not be generalized to children or all adolescents. Age-appropriate research should evaluate consent, developmental stage, content diversity, prior harassment, and whether exposure involves peers, influencers, or altered self-images.
Gendered and muscularity-focused ideals
Appearance demands can emphasize thinness, muscularity, leanness, skin, facial symmetry, youthfulness, hair, or other dimensions. Women and men may experience different pressures, while gender-diverse people can encounter additional visibility restrictions or stereotyping. Research limited to samples of women cannot estimate effects in men or gender-diverse users. Equally, system labels that infer gender from faces should never be confused with a person's self-described gender identity. AI-generated representations can reproduce culturally specific binary assumptions that deserve independent examination.
Racialized aesthetics, disability, aging, and visible difference
A generator may omit, alter, or stereotype people with darker skin, visible disabilities, older faces, scars, skin conditions, or culturally specific features. Such absence can be studied through representation audits and user-experience research. Psychological outcomes also depend on lived discrimination, local beauty norms, self-definition, social support, and community context. Diversity in an image set is not reducible to a single attractiveness scale. Algorithms should not be treated as authorities for rating the acceptability of natural human variation. AlDahoul et al., 2025; Doh et al., 2025.
Body-image constructs and clinical boundaries
Dissatisfaction, distress, and ordinary appearance concerns
Body dissatisfaction means negative evaluation of some aspect of one's body or appearance. A temporary decrease in satisfaction after viewing an image can be detected in a laboratory without implying a psychiatric disorder. Body-image disturbance is a broader clinical and research expression whose precise meaning depends on the assessment context. Appearance anxiety may involve fear of criticism or anticipated negative evaluation. These constructs can overlap but retain separate definitions.
Body dysmorphic disorder, eating disorders, and muscle dysmorphia
Body dysmorphic disorder involves persistent preoccupation with perceived defects that are not observable or appear slight to others, accompanied by repetitive behaviors or mental acts and clinically significant distress or impairment. Eating disorders require their own diagnostic criteria and assessments; concerns focused on body weight or shape can form part of some eating disorders without making every dissatisfaction experience a disorder. Muscle dysmorphia is a form of body dysmorphic disorder involving preoccupation with perceived insufficient muscularity. Social anxiety disorder is primarily organized around fears of negative social evaluation, which may include appearance themes. The distinction rests on clinical history, symptoms, functional impairment, and professional assessment, not the presence of AI-image exposure.
There is no recognized diagnosis of “AI beauty addiction,” “algorithmic beauty syndrome,” or an automatic clinical consequence of liking synthetic images. Increased checking, compulsive editing, isolation, restrictive eating, or escalating distress deserves evaluation according to the person's actual symptoms and circumstances. Clinical treatment for body dysmorphic disorder should follow established evidence-based pathways rather than recommendations generated by appearance-scoring apps. NICE CG31.
Suicidal distress and safety
When appearance-related distress co-occurs with severe depression, self-harm thoughts, significant impairment, or symptoms of body dysmorphic disorder or an eating disorder, clinical evaluation should address immediate safety and the full condition. Neither a social-media questionnaire nor an AI assessment tool establishes diagnosis or suicide risk on its own. Safety assessment and treatment belong with qualified clinicians and appropriate emergency services when danger is immediate. NICE CG31.
Measuring exposure and psychological outcomes
Distinguish the technology from the stimulus
A scientifically interpretable design identifies whether the stimulus is AI-generated from scratch, AI-enhanced from a real photograph, conventionally retouched, human-created, or algorithmically recommended. When possible, record model version, prompts, seed variation, content features, labels, participants' beliefs, exposure duration, and actual recommendation logs. A synthetic image should not be treated as a single uniform type of exposure: fit ideals, body diversity, neutral landscapes, and satirical artwork carry different social meanings.
Measuring body image
Relevant measures include state body satisfaction, appearance satisfaction, body appreciation, functionality appreciation, appearance comparison, self-objectification, body surveillance, and distress or impairment when clinically indicated. State ratings detect immediate responses, whereas trait instruments address more sustained dispositions. Studies should preregister primary endpoints, report baseline scores, consider multiple testing, and examine whether scales operate similarly across gender, age, language, and cultural groups. The 2026 generated-imagery experiments demonstrate the value of measuring several outcomes, since different dimensions do not necessarily move together. Alleva et al., 2026; Tian et al., 2026.
Designs that can establish stronger causal claims
Experiments can hold image content constant and vary only source disclosure, or compare otherwise matched human and synthetic portrayals. Factorial designs can vary body diversity, realism, editing, source attribution, and visibility of engagement metrics. Longitudinal measurement can test whether exposure precedes changes in comparison or distress; within-person sampling can examine short-term fluctuations in daily life. Platform audits can test the generated or recommended distribution of ages, skin tones, body sizes, and other attributes. Joint audits and user studies are particularly valuable because system output and human response are complementary forms of evidence.
Prevention and supportive practices
Algorithmic and visual media literacy
A useful educational approach teaches how generators and recommendation systems operate, what editing changes, why an apparently neutral output may reflect historical stereotypes, and why a machine's depiction is not an evidence-based beauty judgment. It also addresses the commercial incentives around appearance engagement. These are plausible protective skills rather than a proven stand-alone cure for AI-related dissatisfaction. In school-based media-literacy research, a meta-analysis of 17 program evaluations found improvements in body image and media literacy, with substantial heterogeneity and relatively small effects. The programs mostly preceded widespread generative image tools; dedicated AI-literacy trials remain necessary. Kurz et al., 2022.
Diversifying exposure and reducing appearance surveillance
People who notice escalating comparison can experiment with diversifying visual feeds, reducing time spent examining altered self-images, and shifting attention toward meaningful activities and body functionality. These steps follow established body-image models and have plausible value; their benefits for specific AI-generated content settings remain to be quantified. A single exposure to generated fit ideals can affect short-term body appreciation differently from exposure to generated functionality-and-diversity images. Content diversity should therefore be evaluated empirically, with attention to cultural relevance rather than treated as a cosmetic change to an unchanged recommendation system. Alleva et al., 2026.
Counseling and professional care
When appearance comparison causes sustained distress or impairment, psychological care can assess underlying processes such as perfectionism, shame, compulsive checking, avoidance, social anxiety, or disordered eating. Interventions should target the assessed difficulty and use established clinical guidance. People who have developed intrusive appearance preoccupations should receive appropriate professional assessment, especially before elective cosmetic interventions. Exposure to AI-generated images can be one context in the formulation without being assigned the status of an established diagnosis or singular cause.
Ethical and institutional implications
Disclosure, consent, and image provenance
Platforms and creators can disclose when an apparently photographic person is synthetic or materially altered. Disclosure serves transparency, media interpretation, and consent, even though experimental evidence shows that some disclosure labels alone do not prevent short-term body-image changes. Systems that modify a person's real appearance should distinguish authorized creative editing from nonconsensual manipulation and should give people meaningful control over the use and distribution of their likeness. Bennett et al., 2026.
Evaluating algorithmic bias
Audits should investigate representation distributions, systematic associations between attractiveness and positive traits, and downstream differences in performance. Demographic labels need transparent operational definitions and careful attention to overlapping identities. Attractive-face classifiers and generated “beauty scores” are neither objective measures of personal worth nor appropriate substitutes for clinical assessment. Evidence from face-generation audits provides direct grounds for inclusive dataset, prompt, and evaluation practices. AlDahoul et al., 2025; Doh et al., 2025.
Constructive uses of generative imagery
Generated images can also represent a wide range of bodies, demonstrate the unreliability of a single aesthetic standard, or illustrate functionality and everyday embodied experience. Such projects require participatory design, transparent sourcing, and trials that assess unintended consequences. Inclusive presentation is not automatically inclusive in effect: viewers and communities should help define what counts as meaningful representation, and the outcomes should be measured rather than presumed.
Evidence map and unresolved scientific questions
Established findings
Controlled experiments demonstrate that the content of idealized visual media can change short-term body-image ratings. Meta-analyses show reproducible associations among online appearance comparison, dissatisfaction, and self-objectification. Specific audits document demographic stereotypes and attractiveness-related associations in certain generated-face systems. These are distinct findings, supported by different methods and outcome measures. Bonfanti et al., 2025; Wang et al., 2025; AlDahoul et al., 2025.
Emerging experimental findings
Experiments published in 2026 directly test generated fit ideals, AI-created influencer images, source disclosure, and virtual versus human influencers. Their findings indicate short-term effects of particular image sets and audiences and show that source labels alone are not a universal solution. These studies offer a firmer basis for discussion than speculative generalizations, but replication across populations, platforms, repeated exposure schedules, and appearance categories is still needed. Alleva et al., 2026; Bennett et al., 2026; Tian et al., 2026.
Limited and contested claims
The claim that algorithms independently create an entirely new psychological disorder has no established clinical basis. Claims about long-term changes in diagnostic prevalence, cosmetic procedure uptake, perceived normality, or adolescent development require evidence beyond cross-sectional surveys and brief laboratory exposures. It also remains unresolved how much additional risk comes specifically from synthetic origin compared with familiar mechanisms of idealization, social comparison, and biased representation. Effects may differ across model generations, content objectives, cultures, and individual circumstances.
Frequently asked questions
Can an AI-generated face become a beauty standard even when people know it is artificial?
Yes, an artificial image can function as a comparison target. Experimental research shows that disclosure of AI origin did not eliminate short-term declines in body satisfaction in one study. This does not establish that everyone internalizes the depicted standard or that disclosure has no other value. Bennett et al., 2026.
Are AI-generated beauty ideals more harmful than conventional influencer images?
Current experiments do not support a universal ranking. One comparison of virtual and human influencers observed increased dissatisfaction in both conditions without a significant difference between them. The outcomes depend on the image content, comparison group, audience, and measurement. Tian et al., 2026.
Does every AI beauty filter cause body dysmorphic disorder?
No. Aesthetic editing, a temporary change in appearance satisfaction, and clinical body dysmorphic disorder belong to different categories. Diagnosis requires a comprehensive assessment of preoccupation, repetitive behavior or mental acts, distress, impairment, and differential diagnoses. NICE CG31.
Can generative AI support positive body image?
Some generated images emphasize body functionality and diversity rather than restrictive fit ideals. A randomized study found more favorable short-term outcomes in the functionality-and-diversity condition than in the fit-ideal condition for several body-image measures. The result concerns a tested image set and does not establish long-term therapeutic efficacy. Alleva et al., 2026.
What would strong evidence of algorithmic beauty discrimination require?
Technical evidence would compare generated or ranked outputs across controlled prompts, contexts, and identities, with transparent sampling and independently assessed outcomes. Psychological evidence would then measure how representative exposure patterns influence people over time. A survey of what users think an algorithm favors is informative about perceived bias but cannot replace a system audit. Doh et al., 2025; Torres-Toukoumidis et al., 2026.
Conclusion
AI-generated beauty ideals emerge where synthetic image production, appearance modification, and algorithmic visibility intersect. Their effects are best understood through carefully separated layers of evidence: properties of generated outputs, immediate psychological responses, longer-term patterns of self-evaluation, and the social consequences of unequal representation. Contemporary experiments establish that generated idealized images can influence short-term body-image outcomes, while audits reveal biases in some image-generation systems. The scientific priority is to measure which designs, representations, and exposure conditions promote appearance-related distress or support a broader, more realistic understanding of embodied human diversity.
Related Articles
Cluster Body Image and Appearance Psychology — scientific foundations, clinical boundaries, and research on appearance.
Weight Stigma and Mental Health: Discrimination, Stress, and Outcomes — social exclusion and appearance-related stigma.
Beauty Filters and Selfies: Effects on Body Image and Appearance Expectations — altered self-images and editing-specific mechanisms.
Body Positivity: Effects on Body Image and Evidence Limitations — representation diversity and evidence on body-positive content.
References
AlDahoul, N., Rahwan, T., and Zaki, Y. (2025). AI-generated faces influence gender stereotypes and racial homogenization. Scientific Reports, 15, 14449. https://doi.org/10.1038/s41598-025-99623-3
Alleva, J. M., Türkcan, T., Lin, L., Sloutas, C. L., and Fardouly, J. (2026). The effects of exposure to imagery created with generative artificial intelligence (gen-AI) on young women's body image: Do image type and disclosure matter? Computers in Human Behavior: Artificial Humans, 9, 100339. https://doi.org/10.1016/j.chbah.2026.100339
Bennett, B. L., Lamere, A., Kopan, A., Wood, E., Hari, R., and Wright, E. (2026). Filtering trust: Disclosing the role of artificial intelligence decreases trust in technology, but does not prevent harm to body image after viewing AI-generated content. International Journal of Eating Disorders, 59, 1304–1312. https://doi.org/10.1002/eat.70059
Bonfanti, R. C., Melchiori, F., Teti, A., Albano, G., Raffard, S., Rodgers, R., and Lo Coco, G. (2025). The association between social comparison in social media, body image concerns and eating disorder symptoms: A systematic review and meta-analysis. Body Image, 52, 101841. https://doi.org/10.1016/j.bodyim.2024.101841
Doh, M., Gulati, A., Mancas, M., and Oliver, N. (2025). When algorithms play favorites: Lookism in the generation and perception of faces. Proceedings of Machine Learning Research, 294, 474–480. https://proceedings.mlr.press/v294/doh25a.html
Hussain, B., Aslam, S., and Imran, A. (2025). Manufacturing beauty: How AI and social media are redefining aesthetic norms in emerging digital cultures. Acta Psychologica, 260, 105734. https://doi.org/10.1016/j.actpsy.2025.105734
Kurz, M., Rosendahl, J., Rodeck, J., Muehleck, J., and Berger, U. (2022). School-based interventions improve body image and media literacy in youth: A systematic review and meta-analysis. Journal of Prevention, 43, 5–23. https://doi.org/10.1007/s10935-021-00660-1
Makowski, D., Te, A. S., Neves, A., Kirk, S., Liang, N. Z., Mavros, P., and Chen, S. H. A. (2025). Too beautiful to be fake: Attractive faces are less likely to be judged as artificially generated. Acta Psychologica, 252, 104670. https://doi.org/10.1016/j.actpsy.2024.104670
National Institute for Health and Care Excellence. (2005, updated). Obsessive-compulsive disorder and body dysmorphic disorder: Treatment (CG31). https://www.nice.org.uk/guidance/cg31/chapter/Recommendations
Ooi, W. L., Nasser, H., Simmons, J., and Krug, I. (2025). A systematic review and meta-analysis on the temporal relationship between appearance comparisons and body dissatisfaction. Body Image, 53, 101885. https://doi.org/10.1016/j.bodyim.2025.101885
Tian, H., Rouse, S. V., Yoon, S., and Harriger, J. A. (2026). Is beauty in the AI of the beholder? Exposure to human vs. virtual influencers and effects on appearance dissatisfaction and interest in cosmetic surgery. Body Image, 58, 102130. https://doi.org/10.1016/j.bodyim.2026.102130
Torres-Toukoumidis, A., León-Alberca, T., and Cale Lituma, J. P. (2026). Artificial intelligence, body image comparison, and perceived aesthetic bias: A correlational study of weight-based digital exclusion. Current Psychology, 45, 1576. https://doi.org/10.1007/s12144-026-10121-9
Wang, H., Alivi, M. A. B., and Mustafa, S. E. B. (2025). Unveiling the relationship between social media and self-objectification: A three-level meta-analysis. Body Image, 53, 101895. https://doi.org/10.1016/j.bodyim.2025.101895
Zhang, Y., Xiong, S., Yao, N., and Chen, J. (2025). Social media and body image in young women: A meta-analysis of experimental studies on video- versus image-based platforms. Body Image, 54, 101991. https://doi.org/10.1016/j.bodyim.2025.101991