pith. sign in

arxiv: 2604.16414 · v1 · submitted 2026-04-02 · 💻 cs.CY · cs.SI

How Do Terms of Service Influence Social Media User Dynamics from A Privacy Anxiety Perspective

Jingyuan Liu This is my paper

Pith reviewed 2026-05-13 21:34 UTC · model grok-4.3

classification 💻 cs.CY cs.SI
keywords privacy anxietyterms of servicesocial media dynamicsuser engagementmigration intentionsAI trainingcreator communitiesplatform governance
0
0 comments X

The pith

A Terms of Service update enabling default AI training on X activates privacy anxiety in creator communities that then spreads to users, lowering engagement and raising migration intentions.

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

The paper examines how an X Terms of Service change that sets AI training on user content as the default reduces users' sense of control over their data. This loss of control creates privacy anxiety that first appears among content creators and then moves outward through community interactions to broader user groups. Once anxiety rises, users post and interact less while showing stronger plans to switch platforms. The work frames this sequence as evidence of a deeper tension in how platforms can pursue AI development without eroding the trust needed to keep users active.

Core claim

Privacy anxiety is activated within creator communities and diffused across user groups through inter- and cross-community interaction. As anxiety escalated, engagement declined and migration intentions increased. These findings point to an unresolved dilemma in AI-driven platform governance: how user trust and autonomy can be sustained under conditions of concentrated power and data-dependent business models remains unclear.

What carries the argument

Privacy anxiety, treated as a structural outcome of reduced control over data use especially among content creators, serves as the linking mechanism between the TOS update and measurable drops in engagement plus rises in migration intentions.

If this is right

  • Privacy anxiety originates in creator communities before reaching other users through network interactions.
  • Higher anxiety levels directly correspond to reduced posting, liking, and other platform activity.
  • Anxiety escalation produces measurable increases in users' stated intentions to leave for other services.
  • AI-driven governance creates an inherent tension between data extraction and the autonomy users need to remain engaged.

Where Pith is reading between the lines

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

  • Platforms that make AI training opt-in rather than default could limit the spread of anxiety and preserve engagement.
  • The same anxiety dynamic is likely to appear on other major platforms that adopt similar default data-use policies.
  • Actual migration behavior after the update could be tracked to test whether stated intentions translate into real departures.
  • The findings connect to larger questions of whether users will tolerate continued data use for AI when control feels permanently reduced.

Load-bearing premise

That observed drops in engagement and rises in migration intentions are caused mainly by privacy anxiety from the TOS update rather than by other platform changes or outside events happening at the same time.

What would settle it

Longitudinal data on the same users showing no drop in engagement or migration plans once other platform events are statistically controlled would falsify the causal role assigned to the TOS-triggered anxiety.

Figures

Figures reproduced from arXiv: 2604.16414 by Jingyuan Liu.

Figure 1
Figure 1. Figure 1: Anxiety Score Calculation........................................................................... 17, 51 [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 6
Figure 6. Figure 6: Interaction Categorized by Different Anxiety Level [PITH_FULL_IMAGE:figures/full_fig_p036_6.png] view at source ↗
read the original abstract

This study examines how a Terms of Service update on X enabling default AI training on user content activated privacy anxiety and reshaped user behavior. Privacy anxiety is conceptualized as a structural outcome of reduced control over data use, particularly among content creators. The study finds that privacy anxiety is activated within creator communities and diffused across user groups through inter- and cross- community interaction. As anxiety escalated, engagement declined and migration intentions increased. These findings point to an unresolved dilemma in AI-driven platform governance: how user trust and autonomy can be sustained under conditions of concentrated power and data-dependent business models remains unclear.

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

1 major / 1 minor

Summary. The manuscript examines the effects of a Terms of Service update on the X platform that defaults user content to AI training. It conceptualizes privacy anxiety as arising from reduced data control, especially among content creators. The central claim is that this anxiety activates in creator communities, diffuses via inter- and cross-community interactions, and produces measurable declines in engagement together with increased migration intentions.

Significance. If the causal attribution can be secured, the work would add to the literature on privacy perceptions and platform governance by documenting a community-level diffusion mechanism and its behavioral consequences. It draws attention to tensions between AI-driven data practices and user autonomy, with potential relevance for platform policy and regulatory design.

major comments (1)
  1. [Empirical analysis / Results] The central claim requires that observed drops in engagement and rises in migration intentions are attributable primarily to TOS-triggered privacy anxiety rather than concurrent platform changes or external events. The empirical design relies on community interaction patterns and self-reported anxiety but provides no documented identification strategy (difference-in-differences, regression discontinuity at the rollout date, or matched control cohort) to isolate the effect. This is load-bearing for the attribution and must be addressed with explicit pre-trend checks or counterfactual construction.
minor comments (1)
  1. [Abstract] Abstract: the phrase 'structural outcome of reduced control' is used without a supporting definition or citation; a brief operationalization would improve clarity.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comments and the recommendation for major revision. The concern about causal identification is substantive and we address it directly below by outlining concrete additions to the empirical strategy.

read point-by-point responses

  1. Referee: The central claim requires that observed drops in engagement and rises in migration intentions are attributable primarily to TOS-triggered privacy anxiety rather than concurrent platform changes or external events. The empirical design relies on community interaction patterns and self-reported anxiety but provides no documented identification strategy (difference-in-differences, regression discontinuity at the rollout date, or matched control cohort) to isolate the effect. This is load-bearing for the attribution and must be addressed with explicit pre-trend checks or counterfactual construction.

    Authors: We agree that the attribution of behavioral changes to the TOS update and ensuing privacy anxiety requires a clearer identification strategy. In the revised manuscript we will add a difference-in-differences specification that exploits the known rollout date of the TOS change. Creator communities will serve as the treated group; matched non-creator cohorts and users on unaffected platforms will serve as controls. We will report pre-trend tests, event-study coefficients, and robustness checks that include controls for other contemporaneous platform events. These additions will appear in a new subsection on identification and will be accompanied by the corresponding tables and figures. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical claims rest on external observations without self-referential derivations

full rationale

The manuscript presents an empirical study of user behavior changes following a TOS update on X, framing privacy anxiety as a structural outcome observed through community interactions. No equations, models, fitted parameters, or derivations appear in the text. Claims about anxiety activation, diffusion, engagement decline, and migration intentions are stated as direct findings from the study rather than predictions derived from prior fitted inputs or self-citations. The central attribution to TOS-triggered anxiety is presented as an empirical result without reducing to definitional equivalence or imported uniqueness theorems. This is a standard non-circular empirical paper whose validity hinges on data collection and identification strategy rather than internal logical closure.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only; the central claim rests on the domain assumption that privacy anxiety functions as a measurable structural outcome of data control loss, with no free parameters or invented entities specified.

axioms (1)
  • domain assumption Privacy anxiety is a structural outcome of reduced control over data use, particularly among content creators.
    This definition underpins the activation and diffusion claims in the abstract.

pith-pipeline@v0.9.0 · 5385 in / 1156 out tokens · 52100 ms · 2026-05-13T21:34:22.905287+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

15 extracted references · 15 canonical work pages

  1. [1]

    Introduction 1.1 Background In October 2024, X (formerly Twitter) announced its upcoming update of Terms of Service (ToS), giving permission to machine learning training and generative AI models on user generated content. The terms explicitly state, ‘analyze text and other information you provide and to otherwise provide, promote, and improve the Services...

    work page 2024

  2. [2]

    value tensions

    Theoretical Framework and Related Work 2.1 Value Sensitive Design and Misalignment The framework of Value Sensitive Design (VSD) states that technology is not value-neutral but heavily implicated with moral and ethical values (Friedman et al., 2013). A core concern within VSD is "value tensions" or "misalignment,". Value tensions arise when platform desig...

    work page 2013

  3. [3]

    value misalignment

    When it comes to social media platform like X, "value misalignment" represent the logic of connectivity tends to override individual values such as privacy, autonomy, and user control (van Dijck, 2013). It also happens when incompatible logics produce structural misfits that challenge organizational legitimacy and trigger resistance among stakeholders (Kr...

    work page 2013

  4. [4]

    value artifacts

    As for ToS documents, they are not merely legal contracts but "value artifacts" that codify the platform's power to dictate how data is valued and used (Fiesler C. L., 2016). Platform values are encoded through language that expands licensing rights, legitimizes automated data scraping, and collects user content for AI model training. The VSD framework he...

    work page 2016

  5. [5]

    Studies on privacy anxiety emphasize its role as an affective condition shaped by uncertainty, power asymmetries, and lack of control over data use. (Acquisti, 2015)More recent research on platform migration further suggests that users do not leave platforms solely due to technical features, but because of perceived value conflicts and breakdowns in trust...

    work page 2015

  6. [6]

    The choice of this design is driven not by methodological preference, but by the multi-layered nature of the research problem itself

    Methodology 3.1 Methodological Rationale This study adopts a mixed-methods research design grounded in methodological triangulation to examine the formation, amplification, and consequences of privacy anxiety following the X ToS update. The choice of this design is driven not by methodological preference, but by the multi-layered nature of the research pr...

    work page 2023

  7. [7]

    This process also generated an empirically grounded anxiety lexicon, which was subsequently used to quantify shift in anxiety expression

    Through iterative reading of posts and related replies and quotes, the analysis identified shared interpretive patterns into a community level. This process also generated an empirically grounded anxiety lexicon, which was subsequently used to quantify shift in anxiety expression. The structure of the lexicon is summarized in Table 3, with full details pr...

    work page 2023

  8. [8]

    Full technical specifications are provided in the Appendix C

    Community Pre (#posts) During (#posts) Post (#posts) Total Sensitive 2038 3336 2148 7522 Non-sensitive 99 120 105 315 Table 4: Number of Posts by Community and Phase 16 3.4.2 Metrics To quantify users’ reactions to the ToS update, we construct a set of metrics that operationalize anxiety expression, value orientation, engagement intensity, and privacy ris...

    work page 2038

  9. [9]

    Specifically, we conceptualized privacy anxiety as a value-induced risk signal, whose activation is explained through a Value Sensitive Design (VSD) perspective

    The Activation of Privacy Anxiety Based on an overall Social Amplification of Risk Framework, this chapter studies the starter stage of the risk process: how a platform governance change is interpreted as the risk signal in the first place. Specifically, we conceptualized privacy anxiety as a value-induced risk signal, whose activation is explained throug...

    work page 2023

  10. [10]

    scissor-like

    to one dominated by IP-negative (October 2024). A "scissor-like" pattern can be easily discovered regarding IP discourse topic. Specifically, IP-positive content, associated with user ownership and data protection consistently dropped by 20% by 2024-10. While IP-negative content, emphasizing data extraction surged aggressively and becomes dominant. 20 Thi...

    work page 2024

  11. [11]

    The Amplification of Anxiety Based on SARF, a social media platform functions not merely as a transmission channel, but as a complex web of ‘social stations’ where risk signals are decoded, processed, and recoded. Having established how privacy anxiety forms through value misalignment and collective interpretation, this chapter examines how that anxiety i...

    work page 2003

  12. [12]

    LuizaJarovsky

    Conclusions This study examines a platform-driven governance decision by X’s ToS update that reshaped community dynamics. The findings show that privacy anxiety was not uniformly experienced across users but was disproportionately concentrated among content creators whose labor is both highly visible and highly extractable. Through interaction networks an...

    work page 2018

  13. [13]

    We are the product

    1.00 = extremely anxious -1.00 = not anxious at all 0.00 = neutral or unclear anxiety Output a value such as: 0.27, -0.53, 0.91, etc. # Output format requirements: Output only one number, nothing else # Format: floating-point, two decimal places No explanation, no description, no text besides the number {data} 56 BIBLIOGRAPHY Acquisti, A. B. (2015). Priva...

    work page doi:10.1126/science.aaa1465 2015

  14. [14]

    (2010) Netnography: Doing Ethnographic Research Online

    10.23860/MGDR-2018-03-03-08 Kozinets, R.V. (2010) Netnography: Doing Ethnographic Research Online. Sage Publications, London. Kraatz, M. S., Flores, R., & Chandler, D. (2020). The value of values for institutional analysis. Academy of Management Annals, 14(2), 474–512. https://doi.org/10.5465/annals.2018.0074 Kraatz, M.S., & Block, E.S. (2008). Organizati...

    work page doi:10.23860/mgdr-2018-03-03-08 2018

  15. [15]

    Silent Listeners: The Evolution of Privacy and Disclosure on Facebook

    “Silent Listeners: The Evolution of Privacy and Disclosure on Facebook”. Journal of Privacy and Confidentiality 4(2). https://doi.org/10.29012/jpc.v4i2.620 58 Teigen, K. H. (1994). Yerkes-Dodson: A law for all seasons. Theory & Psychology, 4(4), 525–547. https://doi.org/10.1177/0959354394044004 Tufekci, Z. (2015). Algorithmic Harms beyond Facebook and Goo...

    work page doi:10.29012/jpc.v4i2.620 1994