Research

Research Spotlight

Algorithmic Pathways to Extremity and Online Sexual Harms Among Adolescents

What We Looked At

This review examined how algorithmically curated social media platforms shape adolescents’ exposure to sexual content, with particular attention to processes by which users are progressively directed towards more explicit or harmful material. It draws on research into recommender systems, online radicalisation and extremism, adolescent development, and platform design, as well as findings from a 2026 Ofcom-commissioned report by Protect Children examining perpetrator pathways to child sexual abuse material (CSAM). The review also considers recent investigative reporting and whistleblower disclosures regarding decisions made by major social media companies about how their recommendation systems are configured.

Direct research into the role of algorithms in shaping exposure to harmful sexual content among adolescents specifically remains limited. Much of the relevant literature approaches the question from adjacent areas, including political radicalisation, online misogyny, and extremism, which provide conceptual frameworks transferable to understanding sexual content escalation.

What the Research Found

Social media platforms rely on recommender systems optimised primarily for engagement, measured through metrics such as watch time, clicks, and dwell time. These systems operate through feedback loops in which user interactions inform future recommendations, gradually narrowing and intensifying content exposure (Zou et al., 2019). Rather than users accessing harmful content in a single step, exposure typically unfolds incrementally through chains of recommendations — a process described in the literature as algorithmic “pathways.” Even minimal engagement signals, such as pausing on a video or rewatching a clip, can be interpreted by an algorithm as an indicator of interest and used to feed the user progressively more extreme material.

Unlike dedicated pornography platforms, mainstream social media rarely presents explicit material outright. Instead, sexualised content is often embedded within influencer culture, aesthetics, and performative self-presentation, and this “borderline” content is algorithmically amplified because it drives engagement. Weiss, Koban and Matthes (2025) describe a “gateway” effect whereby initial encounters with relatively soft sexualised content lower the threshold for engagement with more explicit material, accelerating progression. This process is often subtle and difficult for users to detect; recommendation systems embed sexualised content within otherwise non-sexual feeds rather than requiring users to search for it (Regehr et al., 2025). Meilani et al. (2023), reviewing adolescent social media use, found that individuals engaging with more than four platforms daily were 6.8 times more likely to access pornographic content.

Adolescents are particularly susceptible to these feedback loops. Developmental research highlights heightened reward sensitivity, increased novelty-seeking, and still-developing impulse control as features of adolescence that make young people more responsive to reinforcing platform mechanics (Walker et al., 2017). Adolescence is also a critical period for identity formation and sexual development, meaning that early or repeated exposure to sexualised content may play a formative role in shaping expectations and behaviour. Limited media literacy among some young people further constrains their ability to critically interpret algorithmically curated content (Protect Children, 2026).

Repeated exposure to escalating content can produce desensitisation, a reduction in emotional response to material that was previously experienced as shocking, which may in turn sustain or accelerate further escalation as users seek more intense stimulation. Daneback et al. (2018) found that adolescents became less bothered by sexual content they had previously found novel or disturbing, with results suggesting a normalisation of sexually explicit material over time. The Protect Children (2026) report found that 39% of respondents reported being exposed to harmful or sexual content through algorithmic recommendations without having actively searched for it, supporting the view that escalation pathways can be initiated by passive exposure rather than deliberate user intent.

Gamification is a further mechanism through which harmful content exposure is sustained and intensified. Research by Regehr et al. (2025) found that five days of platform usage could produce an exponential increase in the volume of misogynistic content appearing on a user’s algorithmically generated feed. In March 2026, a US court found Meta and YouTube liable for deliberately designing addictive products that led to harm in a young user (Wheale, 2026).

Internal documents and whistleblower disclosures indicate that commercial incentives have directly shaped platform decisions about harmful content. Investigative reporting by Spring and Radford (2026) revealed that Meta and TikTok made decisions allowing more harmful content into users’ feeds after internal research demonstrated that outrage drove higher engagement. The same reporting found that Instagram invested in 700 staff to expand its Reels product while refusing requests for two additional specialist staff to work on child protection. Senior managers at Meta reportedly pressured staff to permit more “borderline” harmful content in feeds in order to remain competitive with TikTok, with stock price performance cited as the rationale. These findings directly contradict the framing, sometimes advanced by technology companies themselves, that exposure to harmful content reflects gradual, user-driven progression rather than platform design choices.

What This Means in Practice

The evidence indicates that escalation pathways towards extreme sexual content are shaped by an interaction of user behaviour, developmental vulnerability, and deliberate platform design decisions optimised for engagement over safety. Young people are not simply exercising individual choices to access harmful content; they are often passively directed towards it by systems whose configuration reflects commercial priorities.

The implication for intervention is that approaches focused solely on educating young people about the risks of seeking out harmful content are insufficient. Because escalation pathways are often initiated without deliberate user intent, the platform architecture itself is a primary target for change. Greater algorithmic transparency, stronger obligations on platforms to design recommendation systems that do not amplify harmful content, and independent audit mechanisms are each implicated by the research.

Education and media literacy remain relevant, particularly in developing young people’s understanding of how recommendations are generated and how their engagement signals are used against their interests. However, the evidence base for specific educational interventions that successfully reduce algorithmic escalation is currently limited, and any programme should be designed for evaluation from the outset. The research also highlights a significant gap in longitudinal data tracking how adolescents’ real-world interactions with social media shape their exposure trajectories over time.

References