A commentary published September 2, 2026 argues that the next edition of the Diagnostic and Statistical Manual of Mental Disorders, not expected until around 2030, needs to formally account for the role recommendation algorithms and generative AI now play in eating disorder development and relapse, something the current manual, last revised in 2013, could not have anticipated. The piece lands amid a wave of research trying to quantify just how much social media’s AI-curated content feeds are shaping disordered eating behavior at scale.
A Diagnostic Manual Written for a Different Internet
DSM-5 was finalized in an era before TikTok existed and before generative AI chatbots became a common source of health information for teenagers and young adults. Clinicians argue that treating a patient’s eating disorder today without asking about their algorithmic content exposure, what a recommendation engine has been serving them for months or years, is like treating a smoking-related illness without asking about cigarettes. Fewer than 20% of people with eating disorders ever receive treatment at all, researchers note, and eating-disorder-related content is pervasive across social platforms, creating both a public health problem and, potentially, a detection opportunity. A May 2026 survey of roughly 2,700 people conducted by the National Eating Disorders Association found that 82% of respondents said social media had triggered eating disorder thoughts or behaviors, a figure NEDA CEO Jessica Scheer has cited as evidence that algorithmic exposure belongs in any modern clinical assessment. Riley Nickols, a sports psychologist and clinical adviser for NEDA, has argued that the diagnostic manual’s clinical interview structure should explicitly probe not just what content a patient sees but how they emotionally respond to it, since two patients exposed to identical content can have very different symptom trajectories.
AI as Both the Cause and a Possible Cure
Researchers have spent several years building natural language processing and computer vision models that scan social media posts, captions, images, and engagement patterns for early warning signs of eating disorder symptoms, an approach that has shown promise in academic studies even though it raises its own thorny privacy and consent questions about passively screening people’s public posts for mental health risk. The tension is stark: the same class of AI technology that curates the triggering content in the first place is being proposed as the tool best positioned to detect its harm early.
Clinical Reality Versus Platform Incentives
Mental health clinicians who treat eating disorders describe patients trapped in what researchers have termed a cycle of impact ambivalence, where people with eating disorders recognize that certain content harms them psychologically yet continue engaging with it because platform algorithms are optimized to maximize exactly that kind of compulsive return engagement. Platform companies, for their part, have introduced some content moderation around explicit pro-eating-disorder material, but critics argue that borderline fitness, diet, and body-transformation content, exactly the material algorithms are best at identifying as highly engaging, often slips through moderation entirely while still functioning as a trigger for vulnerable users. Clinicians surveyed by NEDA describe a specific pattern playing out in session: patients in treatment bringing up GLP-1 weight-loss drug content unprompted, and people in recovery describing specific ads and videos that triggered a relapse, evidence, researchers argue, of a pathway running from exposure to emotional response to symptom reinforcement. That three-step cycle is now the working model researchers use to describe how a single scroll session can undo months of treatment progress, and it is precisely the kind of mechanism the current DSM’s checklist-style criteria were never built to capture.
The Case Against Over-Medicalizing Algorithm Exposure
Not every clinician agrees that formally codifying algorithmic exposure into a diagnostic manual is the right move. Some worry it risks pathologizing ordinary social media use for the vast majority of people who scroll the same content without developing an eating disorder, and that the manual should instead focus on individual vulnerability factors rather than environmental exposure, which is difficult to measure consistently across patients and platforms. Others counter that psychiatry has successfully incorporated environmental and social risk factors into diagnostic frameworks before, and that ignoring a documented, measurable exposure simply because it is new and technological would be an unusual exception.
Where This Debate Heads Next
With the next DSM revision years away, the more immediate practical shift is likely to happen inside individual clinical practices and treatment centers, where clinicians say screening questions about social media algorithm exposure are already becoming standard practice ahead of any formal manual update. Meanwhile, expect continued research into AI-based passive detection tools, and continued disagreement over whether platforms should be required to share more data with researchers and regulators about how their recommendation systems interact with users showing signs of disordered eating in the first place. Eating disorder treatment centers say they are already adapting intake questionnaires informally to ask about algorithmic content exposure regardless of what the eventual DSM revision decides, arguing that clinical practice should not wait years for a manual update when patients are struggling with an identifiable, addressable exposure right now. The commentary’s authors are specific about what they want DSM-6 to do procedurally: rather than creating an entirely new diagnostic category, they want the manual to direct clinicians to assess algorithmic exposure as a standard part of a comprehensive eating disorder evaluation, and to assess a patient’s emotional response to that digital exposure within the existing clinical interview structure. That is a narrower, more achievable ask than an entirely new disorder label, and it is already shaping how researchers frame the next round of screening tool development, including machine learning and natural language processing systems built to flag disordered eating language and imagery on social platforms with fewer false positives than current keyword-based moderation.
Photo: Towfiqu barbhuiya / PEXELS via Pexels
