What Counts as Harmful Eating Disorder Content? A Delphi Study Developing Consensus-Based Harm Identification Rules with Lived Experience and Professional Experts

Abstract Objective :We aimed to develop expert-informed harm identification rules to identify eating disorder (ED)-related harmful social media content for individuals at risk of or experiencing ED. Given the increasing influence of social media on ED risk, recovery and relapse, there is a need for systematic approaches to identify harmful content. To our knowledge, this is the first study to systematically develop consensus-based harm identification rules integrating both professional expertise and lived experience perspectives. Method :We employed a modified Delphi methodology, involving n=21 experts by profession (researchers, clinicians and professionals in the ED and body image field) and n=20 experts by lived experience (individuals with a history of ED). The process included four Steering Committee rounds alternating with three Delphi panel rounds (two primary and one refinement) to iteratively refine the rules. Agreement/consensus was assessed using predefined consensus thresholds, and qualitative feedback from open-ended responses was analysed descriptively to inform refinement. Results :We generated a total of N=67 preliminary rules, with n=45 retained. Rules capturing explicit harmful behaviours (e.g., thin-ideal promotion and unsafe weight-loss practices) achieved higher agreement, while context-dependent content (e.g., nutrition, exercise and recovery narratives) required iterative refinement. Qualitative feedback highlighted the importance of context, intent and user vulnerability in determining harm. Conclusion :Harm in ED-related content depends on both content and context. Explicitly harmful content represents clear harm, while other forms are more nuanced and shaped by user interpretation. The resulting rules provide criteria for identifying both explicit and nuanced harm, which could support automated detection, personalised content moderation, and policy to protect vulnerable users while preserving beneficial content.