What we know
Major platforms rank by predicted engagement. Popular ADHD TikTok content is often clinically misleading. Echo-chamber structure varies by platform in other domains.
Algorithmic amplification
The feed is not a library
Recommendation systems predict what will hold attention. That can concentrate related mental-health material. Concentration is not, by itself, a change in self-concept.
How we know it
Engineering explainers, content analyses, and network studies. Almost no studies log the actual recommended mental-health corpus a young person received.
What else could explain it
User search and homophily can cluster content without sophisticated ranking. Creators, not only models, supply the distribution.
What we don't know yet
Whether algorithmic repetition of diagnostic content matters more than total screen time, and whether it changes self-diagnosis.
A loop that is easy to draw and hard to test
- Watch a mental-health video
- Engage
- The system recommends related content
- Greater exposure
- Greater engagement
- Increasingly specialized content
- Potential changes in perceived relevance to self
Steps 1–6 are mechanistically plausible given how rankers work (Narayanan, 2023). Step 7 is a psychological hypothesis. This page will not smuggle 7 in as if it were 3.
Mechanisms to keep separate
Engagement-based recommendation
Systems that predict what will be watched, liked, shared, or dwelt on, then show more of it.
Not the same as: user search, chronological social graph
Well described as engineering practice; weakly linked to psychiatric outcomes.
Repeated exposure
The same class of content returning until it feels personally relevant or normal.
Not the same as: one-off viewing, total screen time
Plausible mediator; seldom measured.
Personalization
User-specific ranking that can narrow the information diet.
Not the same as: mass viral broadcast
Both broadcast virality (FTLB-related videos) and personalization may matter—different predictions.
Echo chambers
Network structures in which people mostly encounter agreeing information and people.
Not the same as: mere high use, a single viral video
Documented in political content more than in mental-health identity content.
Feedback loops
Behaviour changes the feed, which changes behaviour.
Not the same as: static confounding
Core hypothesis of Path D. Direct clinical tests are missing.
Influencer authority
Perceived expertise conferred by reach, relatability, or claimed lived experience rather than clinical training.
Not the same as: peer support between non-celebrities
Yeung et al. found healthcare professionals were a minority of popular ADHD TikTok creators.
Short-form video
Compressed, affective, imitable clips optimized for immediate recognition.
Not the same as: long-form narrative, text communities
May favour checklist-style diagnosis and highly visible symptoms.
Emotional contagion
Automatic spread of affect through mimicry and synchrony.
Not the same as: disorder contagion, functional symptom modelling
Established as a lab phenomenon; not a theory of ADHD or DID.
Behavioural contagion / modelling
Adoption of observed behaviours, including some forms of self-harm or symptom display.
Not the same as: emotional contagion, malingering
Relevant to self-harm and possibly FTLB; condition-specific.
Mass sociogenic illness
Rapid spread of illness-like symptoms within a social group, historically face-to-face, shaped by culturally available models.
Not the same as: infection, fabrication
Historical construct applied, controversially, to some social-media-era presentations.
Mass social-media-induced illness
A proposed contemporary form in which the “group” is a distributed audience of a digital model.
Not the same as: classic local MSI, ordinary destigmatization
Term used in FTLB papers and commentaries. Still a proposed description, not a validated diagnosis.
Peer effects
Influence of friends’ behaviour and norms, online or offline.
Not the same as: algorithmic effects, celebrity modelling
Adolescent development makes peer effects generally plausible; identification is empirically hard.
What this page will not claim
It will not claim that recommenders create psychiatric disorders. It will not treat “the algorithm” as a homunculus with intent. It will not use political echo-chamber papers as if they measured ADHD identity. It will note that short-form video, influencer authority, and quantified feedback are candidate amplifiers of whatever content already exists—helpful, misleading, or mixed.
Linked claims
- Engagement ranking can loop exposureModerate evidence
- Popular ADHD content is often inaccurateModerate evidence
- Average screen-time effects are smallStrong evidence