Dynamics of Algorithmic Content Amplification on TikTok

Fabian Baumann, Nipun Arora, Iyad Rahwan, Agnieszka Czaplicka · 2025

Read Dynamics of Algorithmic Content Amplification on TikTok by Fabian Baumann, Nipun Arora, Iyad Rahwan, Agnieszka Czaplicka (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-20231.txt.

Introduction

TikTok's feed learns fast, and Fabian Baumann, Nipun Arora, Iyad Rahwan, and Agnieszka Czaplicka measured how fast in a 2025 study. They sent out bots with set interests to watch the For You feed, a method called a sock-puppet audit. Each bot liked and watched in line with its assigned taste. The team then tracked how the feed shifted toward that taste.

Content matching a bot's interest was boosted strongly and quickly, mostly within the first two hundred videos. The boost strength varied by topic, a sign of built-in topic bias. Over time the feed's range of topics shrank as the match grew. The link ran the other way too: more boost meant less contact with new tags. The study puts numbers on a widely felt effect.

Catalog record

CREATORFabian Baumann, Nipun Arora, Iyad Rahwan, Agnieszka Czaplicka
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE47 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-20231

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Dynamics of Algorithmic Content Amplification on TikTok Fabian Baumann, Nipun Arora, Iyad Rahwan, Agnieszka Czaplicka Abstract Intelligent algorithms increasingly shape the content we encounter and engage with online. TikTok’s ”For You” feed exemplifies extreme algorithm-driven curation, tailoring the stream of video content almost exclusively based on users’ explicit and implicit interactions with the platform. Despite growing attention, the dynamics of content amplification on TikTok remain largely unquantified. How quickly, and to what extent, does TikTok’s algorithm amplify content aligned with users’ interests? To address these questions, we conduct a sock-puppet audit, deploying bots with different interests to engage with TikTok’s ”For You” feed. Our findings reveal that content aligned with the bots’ interests undergoes strong amplification, with rapid reinforcement typically occurring within the first 200 videos watched. While amplification is consistently observed across all interests, its intensity varies by interest, indicating the emergence of topic-specific biases. Time series analyses and Markov models uncover distinct phases of recommendation dynamics, including persistent content reinforcement and a gradual decline in content diversity over time. Although TikTok’s algorithm preserves some content diversity, we find a strong negative correlation between amplification and exploration: as the amplification of interest-aligned content increases, engagement with unseen hashtags declines. These findings contribute to discussions on socio-algorithmic feedback loops in the digital age and the trade-offs between personalization and content diversity. 1 Introduction Artificial Intelligence (AI) is playing an increasingly significant role in mediating the information we engage with [ 1 ] . From chatbots powered by large language models to content feeds on platforms like Facebook, X, or TikTok, AI shapes the visibility [ 2 ] , dissemination [ 3 ] , and consumption of online content [ 4 ] . This growing influence of algorithms on the distribution of information can profoundly affect cognitive processes at the individual level, including emotional states and decision-making [ 5 , 6 ] . Beyond individual experiences, AI-mediated learning may also transform the emergence and evolution of cultural artifacts [ 7 , 8 ] . From art [ 9 ] to scientific innovation [ 10 ] , AI not only shapes the spread of existing cultural artifacts [ 11 ] but also shapes the creation of new ones, making the study of AI-driven information mediation a crucial frontier in understanding society’s transition into the digital age [ 12 , 13 ] . Figure 1: Sock-puppet audit and content stream analysis. Panel A provides a schematic overview of the sock-puppet audit methodology. The left side illustrates the stream of video content that bots scroll through, where green and red represent interest-aligned and non-aligned videos, respectively. Time is measured as the discrete number of videos, t\in\{1,\dots,N\} , with t=1 corresponding to the first video a bot encounters at the start of an experimental run. Bots analyze each video by extracting its meta-information (description and hashtags) and processing it through a large language model, which determines whether the content aligns with the bot’s predefined interests. If a video matches the bot’s interests, the bot interacts with the platform in three ways: (i) rewatching the video, (ii) liking the video, and (iii) following the video’s creator. Panel B illustrates how key quantities for our analysis are derived from the binary content stream, which encodes whether a video at time t matches the bot’s interest ( S_{\alpha,i}(t)=1 ) or not ( S_{\alpha,i}(t)=0 ). The upper plot schematically depicts the rate of interest-aligned videos over time, r_{\alpha,i}(t) , obtained by convolving S_{\alpha,i}(t) with a uniform kernel. The lower plot illustrates the cumulative count of interest-aligned videos, C_{\alpha,i}(t) , highlighting its monotonically increasing nature and its relationship to S_{\alpha,i}(t) . TikTok represents an extreme case of an algorithm-driven ecosystem [ 14 , 15 ] . As a short-form video-sharing platform, TikTok allows users to create, share, and engage with highly viral, user-generated content across a wide range of topics, including entertainment, lifestyle, education, and politics. Unlike other digital platforms, where content search and discovery is often guided by user queries or influenced by social connections among users, TikTok’s content exposure is almost entirely dictated by its algorithm. Its ”For You” feed curates each user’s content diet primarily based on their explicit and implicit interactions with previously encountered videos, such as rewatching, liking, or following content creators. This direct feedback loop enables the algorithm to rapidly and continuously adapt to user preferences and emerging trends. TikTok’s rapid global expansion, combined with its highly personalized user experience, raises critical questions about the societal impact of its algorithm. Previous studies have examined various aspects of TikTok’s algorithmic feed, including its potentially addictive nature [ 16 , 17 ] , its influence on social connectedness [ 18 ] , and the role of user-algorithm feedback loops in shaping content exposure [ 19 , 20 ] . Despite these insights, it remains largely unclear how—and to what extent—TikTok’s recommendation algorithm shapes the content distribution individuals encounter by amplifying content aligned with their interests. To address this gap, we conduct a sock-puppet audit by deploying automated bots that interact in real-time with TikTok’s ”For You” feed, each programmed with distinct content preferences. Extending beyond previous sock-puppet methodologies [ 21 , 22 , 23 , 24 ] , our approach leverages a large language model (GPT-3.5 Turbo) to dynamically assess video relevance based on descriptions and hashtags. Specifically, bots evaluated each video encountered to determine whether it matched their assigned interests. Upon detecting alignment, the bots performed predefined user-like interactions—such as rewatching, liking, and following the video’s creator—to explicitly signal their preferences to the platform’s recommendation algorithm. Investigating content amplification on TikTok requires selecting topic categories that are sufficiently prevalent to ensure that the bots can effectively signal their interests within a reasonable timeframe. To this end, we analyze three experimental conditions: (i) bots with a single interest in Gaming content (condition G ), (ii) bots with a single interest in Food content ( F ), and (iii) dual-interest bots engaging with both Gaming and Food content ( G+F ). The dual-interest bots interact with content classified as either Gaming , Food , or both. See Fig. 1 for a schematic overview of our methodology and the Methods section for details on the implementation of the sock puppet audit. Adopting this user-centric approach enables us to systematically examine how TikTok’s feed algorithm shapes content distribution based on user interests. Specifically, we investigate how quickly the recommendation algorithm detects a bot’s interests and how it adapts over time. Additionally, we analyze how the dynamics of content amplification vary across different bot interests and how amplification correlates with other content characteristics, such as popularity and duration. Finally, we explore the relationship between content amplification and content diversity, assessing whether increased reinforcement of interest-aligned content narrows exposure to new topics and perspectives. The paper is organized as follows. In the next section, we present our results, detailing the dynamics of interest-aligned content amplification on TikTok’s ”For You” feed. We analyze these dynamics using time series analysis and (hidden) Markov models to uncover distinct recommendation patterns. Following this, we discuss our findings in the context of existing research on algorithmic feeds and outline the limitations of our study. A detailed description of our sock-puppet audit and analysis can be found

Read the complete text