Who is most vulnerable to harm from short-form video algorithms?

Why adolescents are the most documented victims of algorithm-driven harm

The clearest evidence points to adolescents as a primary vulnerable group. In a 2023 investigation, researchers created automated accounts registered as 13-year-olds and found that TikTok's algorithms fed them tens of thousands of weight-loss videos—including content promoting extreme calorie restriction and the so-called 'corpse bride diet'—within just a few weeks of joining the platform [1]. This isn't limited to one platform: the same study reviewed a critical mass of public health and neuroscience research linking algorithm-driven extreme content to increases in poor body image, eating disorders, and suicidality among youth [1]. The economic incentive is staggering—social media platforms earn nearly $11 billion annually from ads targeted at users aged 0 to 17, which the researchers argue drives platforms to keep these harmful algorithms in place [1].

A separate 2026 literature review on children's and adolescents' education adds that the developing brain may be especially vulnerable to addiction, and that already-marginalized groups like left-behind children (those separated from parents) are affected more severely but often overlooked [5]. Both studies agree that adolescents are not just incidental users but are actively targeted by algorithms that exploit their developmental stage.

How shyness and stress amplify addiction risk in young adults

Beyond adolescents, college students with specific personality traits are especially vulnerable. A 2021 study of 896 Chinese college students found that perceived stress was strongly linked to short-form video addiction, and that this link was significantly stronger for individuals with higher levels of shyness [4]. The mechanism works through what researchers call 'self-compensation motivation'—stressed individuals use short videos to escape negative feelings, and shy people are more likely to fall into that trap [4]. This means that a shy, stressed college student is not just a heavy user but is caught in a feedback loop where the algorithm's endless stream of personalized content feeds their need for escape, making addiction more likely.

This finding dovetails with research on 'information cocoons'—where algorithms trap users in a narrow bubble of similar content. A 2022 study of a major short-form video platform tracked new users for a year and found that users with narrow preferences are more likely to become trapped in these cocoons [2]. For shy or stressed individuals, who may already have limited social outlets, the algorithm's personalization can create an even tighter, more isolating loop.

A hidden vulnerability: blind and low-vision viewers are routinely excluded

A less obvious but well-documented vulnerable group is blind and low-vision (BLV) viewers. A 2024 study found that short-form videos are often inaccessible to BLV viewers because of rapid visual changes, on-screen text, and music or meme-audio overlays that don't convey visual information [3]. In a formative study, 7 BLV viewers who regularly watched short-form videos reported frequently skipping such inaccessible content—meaning they are excluded from a major source of information and entertainment [3]. The researchers built a system called ShortScribe that provides hierarchical visual summaries, and in a user study with 10 BLV participants, those using ShortScribe reported higher comprehension and gave more accurate summaries of video content [3]. This shows that the harm here is not just annoyance but real information inequality: algorithms serve content that BLV users cannot fully engage with, and current platform designs do not address this.

About These Sources

This answer is built on 5 peer-reviewed studies — published from 2021 to 2026, 2 from 2024 or later, 1 in Q1 journals, collectively cited 140 times — selected as the most relevant from 5 studies that passed quality screening, drawn from 47 papers retrieved from a database of over 500 million.

Sources used in this answer

Related research questions