pith. machine review for the scientific record. sign in

arxiv: 2605.10690 · v1 · submitted 2026-05-11 · 💻 cs.CY

Recognition: 2 theorem links

· Lean Theorem

When 'For You' Isn't For You: Measuring User Agency in TikTok's Algorithmic Feed

Alan Mislove, Devin Patel, Levi Kaplan, Nicole Gerzon, Piotr Sapiezynski

Authors on Pith no claims yet

Pith reviewed 2026-05-12 05:00 UTC · model grok-4.3

classification 💻 cs.CY
keywords TikTokFor You Pagealgorithmic recommendationuser agencycontent moderationsocial media platformspersonalization controls
0
0 comments X

The pith

TikTok users struggle to permanently exclude unwanted videos from their For You feed.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper examines how much real control users have over the algorithmically generated For You Page on TikTok. The authors built methods to run controlled experiments inside the mobile app, allowing them to send both explicit signals like marking videos 'Not Interested' and implicit signals like varying watch time to accounts they fully control. They show that the algorithm does adjust the feed in response to these signals, yet the most effective explicit control is placed several taps deep in a menu. When accounts stop sending disinterest signals, the unwanted topics often reappear and dominate the feed again.

Core claim

The FYP algorithm changes the volume of personalized content it delivers when accounts provide explicit disinterest marks or alter implicit engagement patterns. The most reliable explicit signal, however, requires opening a menu to select 'Not Interested,' an option that is not surfaced on the main video view. Once accounts stop supplying these signals, the feed frequently reverts to showing high volumes of the same topics the user had previously tried to remove.

What carries the argument

Controlled experiments using researcher-owned TikTok accounts that issue explicit 'Not Interested' marks and implicit watch-time signals to measure resulting shifts in For You Page content composition.

If this is right

  • Users must keep sending disinterest signals to maintain a feed free of unwanted topics.
  • The current interface placement of the 'Not Interested' control reduces how often people will use it.
  • Implicit signals alone are less effective than combined explicit and implicit signals for steering the feed.
  • The algorithm appears to treat absence of recent negative signals as permission to reintroduce previously disliked content.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Platforms could increase user agency by moving the strongest control to the primary video screen.
  • Similar buried controls may limit agency on other short-form video apps that rely on implicit signals at signup.
  • Design changes that keep signals active longer could reduce the need for repeated user effort.

Load-bearing premise

That the behavior seen in researcher-controlled accounts with planned signal patterns matches how the algorithm responds to ordinary users who interact naturally over long periods.

What would settle it

A long-term study of real users who mark videos 'Not Interested' only once and then stop all signals, checking whether the excluded topics stay out of the feed for weeks.

Figures

Figures reproduced from arXiv: 2605.10690 by Alan Mislove, Devin Patel, Levi Kaplan, Nicole Gerzon, Piotr Sapiezynski.

Figure 1
Figure 1. Figure 1: Steps for investigating user agency on TikTok. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Our Account Cloning technique is successful at [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: The number of topic videos delivered to our accounts across our fifteen experiments. The results of Phase 2 can be [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The account which sends an implicit positive sig [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Counts where we find statistical differences for the fifteen experiments across three topics, for Phases 2 and 3. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
read the original abstract

The short-form video-sharing service TikTok has become an important platform in the social media landscape, with much of its popularity owed to its algorithmically-driven "For You Page" (FYP). This feature serves as the "home screen" for the platform and provides a personalized feed of content for each user. Unlike other social media services-where new users start their journey by explicitly signaling whom they choose to friend or follow-the TikTok FYP algorithm instead begins making inferences based on implicit signals, such as how long they watch particular videos. As a result, users have less explicit control over what content they see, and concerns have been raised about the impact on users (e.g., the delivery of potentially harmful content). In this work, we investigate the extent to which users have control over the content they see on the FYP on TikTok. We first develop novel techniques to study the TikTok mobile app, introducing a new avenue for conducting controlled experiments that enable us to send both explicit and implicit signals on the app. We then use these techniques to study the FYP algorithm based on accounts we control. We find that the FYP algorithm is sensitive to both types of signals, changing the amount of personalized content the account sees. However, we find that users may have difficulty convincing the FYP algorithm to stop showing content the user wishes to no longer see: the most effective explicit signal-marking a video as 'Not Interested'-is unintuitively buried in the interface. Worse, we find that once accounts cease to indicate disinterest in a topic, many find their feeds dominated by such content again.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper develops novel techniques for controlled experiments on the TikTok mobile app that allow sending both explicit and implicit signals to the For You Page (FYP) algorithm. Using researcher-controlled accounts, it reports that the algorithm responds to these signals by altering the amount of personalized content shown, but finds that the most effective explicit signal ('Not Interested') is buried in the UI and that feeds tend to revert to previously suppressed topics once such signaling stops.

Significance. If the results hold under more rigorous validation, the work would contribute empirical evidence on limited user agency in black-box recommendation systems, with practical implications for platform interface design and regulatory discussions around algorithmic control. A clear strength is the introduction of app-based experimental methods that enable reproducible, signal-controlled studies of production algorithms, which could be extended to other platforms.

major comments (2)
  1. [Abstract / Methods] Abstract and experimental description: no sample sizes, number of accounts, session lengths, statistical tests, error bars, or validation procedures for account control and signal delivery are provided. This directly affects assessment of the claims that the FYP is 'sensitive to both types of signals' and that reversion occurs 'once accounts cease to indicate disinterest.'
  2. [Results / Discussion] Central reversion finding: the observation that feeds become 'dominated by such content again' after stopping disinterest signals is load-bearing for the agency claim, yet rests entirely on deliberate, repeated signals from researcher accounts. No comparison to organic, lower-density, longer-horizon user logs is described, leaving open whether reversion is an artifact of the experimental regime rather than a stable algorithmic property.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the opportunity to respond to the referee's report. We have addressed the major comments by enhancing the methodological details and expanding the discussion of the experimental limitations.

read point-by-point responses

  1. Referee: [Abstract / Methods] Abstract and experimental description: no sample sizes, number of accounts, session lengths, statistical tests, error bars, or validation procedures for account control and signal delivery are provided. This directly affects assessment of the claims that the FYP is 'sensitive to both types of signals' and that reversion occurs 'once accounts cease to indicate disinterest.'

    Authors: We acknowledge this gap in the original manuscript. The revised version now includes a detailed Experimental Design subsection that reports the use of multiple researcher-controlled accounts, specific session lengths, the statistical tests applied to measure signal sensitivity (with associated p-values and error bars), and procedures used to validate that signals were correctly delivered and accounts remained isolated. These additions directly support evaluation of the sensitivity and reversion findings. revision: yes

  2. Referee: [Results / Discussion] Central reversion finding: the observation that feeds become 'dominated by such content again' after stopping disinterest signals is load-bearing for the agency claim, yet rests entirely on deliberate, repeated signals from researcher accounts. No comparison to organic, lower-density, longer-horizon user logs is described, leaving open whether reversion is an artifact of the experimental regime rather than a stable algorithmic property.

    Authors: This is a fair critique of the scope of our evidence. Our controlled experiments were designed to isolate the effect of signal cessation in a reproducible manner, which is challenging with passive organic logs. We have added text in the Discussion section explicitly noting this limitation and the potential for future studies to validate with user log data where ethically obtainable. We maintain that the observed reversion under controlled conditions provides evidence of the algorithm's response to the withdrawal of disinterest signals, supporting our claims about user agency. revision: partial

Circularity Check

0 steps flagged

Empirical measurement study exhibits no circularity

full rationale

This paper performs controlled experiments on researcher-managed TikTok accounts to observe how explicit and implicit signals affect the FYP algorithm's output. It introduces interface techniques for sending signals and reports direct measurements of content changes and reversion effects. No equations, fitted parameters, derivations, or self-citation chains appear in the work; all claims rest on experimental observations rather than quantities defined in terms of other quantities from the same paper. The central results are therefore independent of any internal reduction and qualify as non-circular.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is an empirical measurement study with no mathematical derivations, fitted parameters, or postulated entities; the central claims rest on the validity of the app instrumentation and the assumption that controlled accounts behave like real users.

pith-pipeline@v0.9.0 · 5609 in / 1101 out tokens · 37806 ms · 2026-05-12T05:00:05.259078+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Reference graph

Works this paper leans on

76 extracted references · 76 canonical work pages

  1. [1]

    Proceedings of the ACM Web Conference 2022 , pages=

    An empirical investigation of personalization factors on tiktok , author=. Proceedings of the ACM Web Conference 2022 , pages=

    work page 2022

  2. [2]

    Proceedings of the 22nd international conference on World Wide Web , pages=

    Measuring personalization of web search , author=. Proceedings of the 22nd international conference on World Wide Web , pages=

    work page

  3. [3]

    TikTok SSL Pinning Bypass , author=

    work page

  4. [4]

    Mitmproxy , howpublished=

    work page

  5. [5]

    Fastest Growing Social Media Networks 2022-23 , howpublished=

    work page 2022

  6. [6]

    Investigation: How TikTok's Algorithm Figures Out Your Deepest Desires , year=

    work page

  7. [7]

    Command-line tools , author=

    work page

  8. [8]

    Android Debug Bridge (adb) , author=

    work page

  9. [9]

    UIAutomator2 , author=

    work page

  10. [10]

    Square Wire Java Protobuf Decompiler , author=

    work page

  11. [11]

    Coull , journal =

    Alan Agresti and Brent A. Coull , journal =. Approximate Is Better than "Exact" for Interval Estimation of Binomial Proportions , urldate =

    work page

  12. [12]

    Proceedings of the International AAAI Conference on Web and Social Media , volume=

    Auditing Algorithmic Explanations of Social Media Feeds: A Case Study of TikTok Video Explanations , author=. Proceedings of the International AAAI Conference on Web and Social Media , volume=

    work page

  13. [13]

    Health Promotion Journal of Australia , volume=

    Communication of nutrition information by influencers on social media: A scoping review , author=. Health Promotion Journal of Australia , volume=. 2022 , publisher=

    work page 2022

  14. [14]

    Sage Open , volume=

    Communitainment on Instagram: Fitness content and community-driven communication as social media entertainment , author=. Sage Open , volume=. 2020 , publisher=

    work page 2020

  15. [15]

    Body image , volume=

    TikTok on the clock but the\# fitspo don’t stop: The impact of TikTok fitspiration videos on women’s body image concerns , author=. Body image , volume=. 2022 , publisher=

    work page 2022

  16. [16]

    2nd International Conference on Language, Art and Cultural Exchange (ICLACE 2021) , pages=

    The influence of the body image presented through TikTok trend-videos and its possible reasons , author=. 2nd International Conference on Language, Art and Cultural Exchange (ICLACE 2021) , pages=. 2021 , organization=

    work page 2021

  17. [17]

    Fitspiration: Social Media's Fitness Culture and its Effect on Body Image , author=

    work page

  18. [18]

    Eating and Weight Disorders-Studies on Anorexia, Bulimia and Obesity , volume=

    Effects of fitspiration content on body image: a systematic review , author=. Eating and Weight Disorders-Studies on Anorexia, Bulimia and Obesity , volume=. 2022 , publisher=

    work page 2022

  19. [19]

    human agency: understanding human--AI interactions on TikTok and their implications for user engagement , author=

    AI agency vs. human agency: understanding human--AI interactions on TikTok and their implications for user engagement , author=. Journal of Computer-Mediated Communication , volume=. 2022 , publisher=

    work page 2022

  20. [20]

    2019 , author =

    Our Recommendation to YouTube , howpublished =. 2019 , author =

    work page 2019

  21. [21]

    2021 , author =

    How TikTok Reads Your Mind , howpublished =. 2021 , author =

    work page 2021

  22. [22]

    2022 , author =

    The Creepy TikTok Algorithm Doesn’t Know You , howpublished =. 2022 , author =

    work page 2022

  23. [23]

    Richard Landis and Gary G

    J. Richard Landis and Gary G. Koch , journal =. The Measurement of Observer Agreement for Categorical Data , urldate =

    work page

  24. [24]

    Simon Bell and Timothy F

    Natasa Gisev and J. Simon Bell and Timothy F. Chen , keywords =. Interrater agreement and interrater reliability: Key concepts, approaches, and applications , journal =. 2013 , issn =. doi:https://doi.org/10.1016/j.sapharm.2012.04.004 , url =

    work page doi:10.1016/j.sapharm.2012.04.004 2013

  25. [25]

    2024 , author =

    Regulated Goods and Commercial Activities , howpublished =. 2024 , author =

    work page 2024

  26. [26]

    2024 , author =

    TikTok CEO Shou Chew's Opening Statement , howpublished =. 2024 , author =

    work page 2024

  27. [27]

    2024 , author =

    TikTok CEO Shou Chew's Opening Statement - Senate Judiciary Committee Hearing on Online Child Sexual Exploitation Crisis - January 31, 2024 , howpublished =. 2024 , author =

    work page 2024

  28. [28]

    2024 , author =

    Most popular short video platform and features worldwide 1st quarter 2024, average video views , howpublished =. 2024 , author =

    work page 2024

  29. [29]

    2023 , author =

    Algorithms of Trauma 2: Stuck in a ``doomscrolling trap'' on Facebook? The platform will not let you escape , howpublished =. 2023 , author =

    work page 2023

  30. [30]

    2021 , author =

    Algorithms of trauma: new case study shows that Facebook doesn’t give users real control over disturbing surveillance ads , howpublished =. 2021 , author =

    work page 2021

  31. [31]

    Wall Street Journal , year=

    Investigation: how TikTok’s algorithm figures out your deepest desires , author=. Wall Street Journal , year=

    work page

  32. [32]

    Nature human behaviour , volume=

    A consensus-based transparency checklist , author=. Nature human behaviour , volume=. 2020 , publisher=

    work page 2020

  33. [33]

    Communications of the ACM , volume=

    Datasheets for datasets , author=. Communications of the ACM , volume=. 2021 , publisher=

    work page 2021

  34. [34]

    Centre for the Governance of AI

    A guide to writing the NeurIPS impact statement , author=. Centre for the Governance of AI. URL: https://perma. cc/B5R8-2B9V , year=

    work page

  35. [35]

    Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics: Tutorial Abstracts , pages=

    Understanding Ethics in NLP Authoring and Reviewing , author=. Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics: Tutorial Abstracts , pages=

    work page

  36. [36]

    NeurIPS 2021 Paper Checklist Guidelines

    NeurIPS. NeurIPS 2021 Paper Checklist Guidelines

    work page 2021

  37. [37]

    The FAIR Data principles

    FORCE11. The FAIR Data principles

    work page

  38. [38]

    Journal of Marketing Research , volume=

    Down a rabbit hole: how prior media consumption shapes subsequent media consumption , author=. Journal of Marketing Research , volume=. 2022 , publisher=

    work page 2022

  39. [39]

    Down the rabbit hole

    “Down the rabbit hole” of vaccine misinformation on YouTube: Network exposure study , author=. Journal of Medical Internet Research , volume=. 2021 , publisher=

    work page 2021

  40. [40]

    2023 , author =

    Sports Gambling , howpublished =. 2023 , author =

    work page 2023

  41. [41]

    Current addiction reports , volume=

    Online gambling addiction: the relationship between internet gambling and disordered gambling , author=. Current addiction reports , volume=. 2015 , publisher=

    work page 2015

  42. [42]

    ‘I literally can’t stop.’ The descent of a modern sports fan , howpublished =

    Jason Quick , year =. ‘I literally can’t stop.’ The descent of a modern sports fan , howpublished =

    work page

  43. [43]

    Journal of Behavioral Addictions , volume=

    Situational features of smartphone betting are linked to sports betting harm: An ecological momentary assessment study , author=. Journal of Behavioral Addictions , volume=. 2023 , publisher=

    work page 2023

  44. [44]

    TikTok removes video critical of gambling advertising while increasing wagering content , howpublished =

    Henry Belot , year =. TikTok removes video critical of gambling advertising while increasing wagering content , howpublished =

    work page

  45. [45]

    How TikTok’s algorithm made it a success: ‘It pushes the boundaries’ , howpublished =

    Alex Hern , year =. How TikTok’s algorithm made it a success: ‘It pushes the boundaries’ , howpublished =

    work page

  46. [46]

    How TikTok recommends videos \#ForYou , howpublished =

    TikTok , year =. How TikTok recommends videos \#ForYou , howpublished =

    work page

  47. [47]

    National Collegiate Athletic Association , court =

    Murphy v. National Collegiate Athletic Association , court =. 2018 , docket =

    work page 2018

  48. [48]

    Journal of Gambling Studies , volume=

    How the Internet is changing gambling: Findings from an Australian prevalence survey , author=. Journal of Gambling Studies , volume=. 2015 , publisher=

    work page 2015

  49. [49]

    Addictive Behaviors , volume=

    Sports-betting-related gambling disorder: Clinical features and correlates of cognitive behavioral therapy outcomes , author=. Addictive Behaviors , volume=. 2022 , publisher=

    work page 2022

  50. [50]

    Proceedings of the ACM on Human-Computer Interaction , volume=

    A comparative study of dark patterns across web and mobile modalities , author=. Proceedings of the ACM on Human-Computer Interaction , volume=. 2021 , publisher=

    work page 2021

  51. [51]

    Proceedings of the ACM on Human-Computer Interaction , volume=

    The distressing ads that persist: Uncovering the harms of targeted weight-loss ads among users with histories of disordered eating , author=. Proceedings of the ACM on Human-Computer Interaction , volume=. 2022 , publisher=

    work page 2022

  52. [52]

    Proceedings of the ACM on Human-Computer Interaction , volume=

    Identifying user needs for advertising controls on Facebook , author=. Proceedings of the ACM on Human-Computer Interaction , volume=. 2022 , publisher=

    work page 2022

  53. [53]

    and Redmiles, Elissa M

    Zannettou, Savvas and Nemes-Nemeth, Olivia and Ayalon, Oshrat and Goetzen, Angelica and Gummadi, Krishna P. and Redmiles, Elissa M. and Roesner, Franziska , title =. 2024 , isbn =. doi:10.1145/3613904.3642433 , booktitle =

    work page doi:10.1145/3613904.3642433 2024

  54. [54]

    Proceedings of the ACM on Web Conference 2024 , pages=

    TikTok and the Art of Personalization: Investigating Exploration and Exploitation on Social Media Feeds , author=. Proceedings of the ACM on Web Conference 2024 , pages=

    work page 2024

  55. [55]

    Data and discrimination: converting critical concerns into productive inquiry , volume=

    Auditing algorithms: Research methods for detecting discrimination on internet platforms , author=. Data and discrimination: converting critical concerns into productive inquiry , volume=

    work page

  56. [56]

    Proceedings of the acm on human-computer interaction , volume=

    Problematic machine behavior: A systematic literature review of algorithm audits , author=. Proceedings of the acm on human-computer interaction , volume=. 2021 , publisher=

    work page 2021

  57. [57]

    A mixed-method study on user assumptions about the TikTok algorithm , author=

    Trick and please. A mixed-method study on user assumptions about the TikTok algorithm , author=. Proceedings of the 13th ACM Web Science Conference 2021 , pages=

    work page 2021

  58. [58]

    The App That Knows You Better than You Know Yourself: An Analysis of the TikTok Algorithm , author=

    work page

  59. [59]

    Proceedings of the ACM on human-computer interaction , volume=

    For you, or for "you"? Everyday LGBTQ+ encounters with TikTok , author=. Proceedings of the ACM on human-computer interaction , volume=. 2021 , publisher=

    work page 2021

  60. [60]

    The playful politics of memes , pages=

    ‘\# OkBoomer, time to meet the Zoomers’: Studying the memefication of intergenerational politics on TikTok , author=. The playful politics of memes , pages=. 2023 , publisher=

    work page 2023

  61. [61]

    Companion Proceedings of the ACM on Web Conference 2024 , pages=

    Comprehensively Auditing the TikTok Mobile App , author=. Companion Proceedings of the ACM on Web Conference 2024 , pages=

    work page 2024

  62. [62]

    Clinical Child Psychology and Psychiatry , volume=

    Using TikTok for public and youth mental health--A systematic review and content analysis , author=. Clinical Child Psychology and Psychiatry , volume=. 2023 , publisher=

    work page 2023

  63. [63]

    Studies in conflict & terrorism , volume=

    Research note: Spreading hate on TikTok , author=. Studies in conflict & terrorism , volume=. 2023 , publisher=

    work page 2023

  64. [64]

    Journal of Adolescent Health , volume=

    Dissemination of mask effectiveness misinformation using TikTok as a medium , author=. Journal of Adolescent Health , volume=. 2021 , publisher=

    work page 2021

  65. [65]

    2024 , school=

    Exploring the Consistency and Variability of Algorithmic Filter Bubbles: A Comparative Analysis of Instagram Reels and TikTok , author=. 2024 , school=

    work page 2024

  66. [66]

    Filter Bubbles

    How Social Media Can Solve the Problem of “Filter Bubbles” Under the NewMedia Algorithm Recommendation Mechanism the Example of Tik Tok , author=. 2023 2nd International Conference on Social Sciences and Humanities and Arts (SSHA 2023) , pages=. 2023 , organization=

    work page 2023

  67. [67]

    2023 , author =

    Can you control your TikTok fyp? , howpublished =. 2023 , author =

    work page 2023

  68. [68]

    2025 , author =

    New Features Bring TikTok's Handheld Magic to Desktop , howpublished =. 2025 , author =

    work page 2025

  69. [69]

    2021 , author =

    How TikTok recommends content , howpublished =. 2021 , author =

    work page 2021

  70. [70]

    Social Media+ Society , volume=

    You can (not) say what you want: Using algospeak to contest and evade algorithmic content moderation on TikTok , author=. Social Media+ Society , volume=. 2023 , publisher=

    work page 2023

  71. [71]

    2024 , school=

    You can’t say that on TikTok: cxnsxrshxp, algorithmic (in) visibility, and the threat of representation , author=. 2024 , school=

    work page 2024

  72. [72]

    This is not pro-ana

    “This is not pro-ana”: Denial and disguise in pro-anorexia online spaces , author=. Fat Studies , volume=. 2017 , publisher=

    work page 2017

  73. [73]

    Proceedings of the 2016 cHI conference on human factors in computing systems , pages=

    First I" like" it, then I hide it: Folk Theories of Social Feeds , author=. Proceedings of the 2016 cHI conference on human factors in computing systems , pages=

    work page 2016

  74. [74]

    Proceedings of the ACM on Human-Computer Interaction , volume=

    How transfeminine TikTok creators navigate the algorithmic trap of visibility via folk theorization , author=. Proceedings of the ACM on Human-Computer Interaction , volume=. 2022 , publisher=

    work page 2022

  75. [75]

    Honestly, I Think TikTok has a Vendetta Against Black Creators

    " Honestly, I Think TikTok has a Vendetta Against Black Creators": Understanding Black Content Creator Experiences on TikTok , author=. Proceedings of the ACM on Human-Computer Interaction , volume=. 2023 , publisher=

    work page 2023

  76. [76]

    Proceedings of the ACM on human-computer interaction , volume=

    Algorithmic folk theories and identity: How TikTok users co-produce Knowledge of identity and engage in algorithmic resistance , author=. Proceedings of the ACM on human-computer interaction , volume=. 2021 , publisher=

    work page 2021