P3.031: TikTok’s Algorithm Unwrapped: A Systematic Analysis of Video Content Shown to New TikTok Users

Background: TikTok is a social media platform with 42.5 million adolescent users (AU). This widespread adoption among AU has raised concerns about the platform's content delivery mechanisms and their potential impact on AU. Of particular concern is TikTok's recommendation algorithm, which curates videos on the “For You Page”, the main place where AU consume content. Given the impressionable nature of AU, understanding how TikTok tailors content to different AU demographics is crucial for assessing risks. This study aims to systematically examine the videos TikTok recommends to new AU based on the age and gender information provided at account creation. It seeks to evaluate the effectiveness of safety measures for AU and identify disparities in content exposure across AU groups.

Methods: 24 new TikTok accounts were created with 6 ages (13-18) and 4 gender options (male, female, nonbinary, none). The first 15 videos on the For You Page were analyzed. The same amount of time was spent on each video with no content interaction to maintain algorithm consistency. Afterwards, the following data were collected on each video: 1) gender/age of video creators; 2) type of content; 3) sensitive content (ST); 4) controversial topics (CT); 5) engagement metrics. Data was analyzed in R.

Results: 360 videos (15,055,051,000 views) were analyzed. Video creator ages ranged from >18 (n=68), 18-22 (n=31), 23-27 (n=71), and < 28 (n=69). 118 were male, 140 female. Account gender did not have a significant effect on age, gender, or content of videos recommended. Similar content was shown to accounts aged 13-17, with differences for 18-year-old accounts. All 13-17 accounts were shown at least 1 educational video promoted by TikTok encouraging critical thinking regarding dangerous online trends. 18-year-old accounts had more videos with ST and CT shown to them. Further testing revealed that spending more time on videos with ST/CT caused the TikTok algorithm to recommend more videos with the same ST/CT. Figure 1 shows breakdown of videos shown to each account. Engagement metrics were relatively consistent among the accounts of all ages and genders.

Conclusion: Our study demonstrated that simply lingering on ST/CT will cause TikTok’s algorithm to recommend more ST/CT in the future, emphasizing the danger of even a few videos with ST/CT being initially shown to AU. While these analyses reveal that TikTok has implemented safety measures for AU under 18, TikTok should consider extending these protective measures to all age groups as some AU may misrepresent their age when signing up. Further research is needed to assess the long-term effectiveness of these safety features.