Reckless Designs and Broken Promises: Privacy Implications of Targeted Interactive Advertisements on Social Media Platforms

Julia B. Kieserman 1 , Athanasios Andreou 1 , Laura Edelson 2 , Sandra Siby 3 , Damon McCoy 1 · 2026

Read Reckless Designs and Broken Promises: Privacy Implications of Targeted Interactive Advertisements on Social Media Platforms by Julia B. Kieserman 1 , Athanasios Andreou 1 , Laura Edelson 2 , Sandra Siby 3 , Damon McCoy 1 (2026) 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-2603-03659.txt.

Introduction

Julia B Kieserman and four others expose a privacy leak hiding inside social media ads. TikTok, Facebook, and Instagram let advertisers aim campaigns at touchy traits. Those ads are interactive by default: anyone can comment or react. Their 2026 paper shows that this design choice lets advertisers view the profiles of all who interact, in effect naming the people who match the touchy targeting tests.

That behavior flatly contradicts the platforms' promises to keep user data hidden from advertisers. A reaction meant as a casual tap becomes an unintentional disclosure of exactly the trait the ad targeted. The authors close with design changes that would warn users about the consequences of interacting and give them a clear view before the tap. That would restore the promised protection without banning aimed ads outright.

Catalog record

CREATORJulia B. Kieserman 1 , Athanasios Andreou 1 , Laura Edelson 2 , Sandra Siby 3 , Damon McCoy 1
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE9925
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.03659

Browse the full library · About the Tower of Babel

Full text

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

Reckless Designs and Broken Promises: Privacy Implications of Targeted Interactive Advertisements on Social Media Platforms Julia B. Kieserman 1 , Athanasios Andreou 1 , Laura Edelson 2 , Sandra Siby 3 , Damon McCoy 1 Abstract Popular social media platforms TikTok, Facebook and Instagram allow third-parties to run targeted advertising campaigns on sensitive attributes in-platform. These ads are interactive by default, meaning users can comment or “react” (e.g., “like”, “love”) to them. We find that this platform-level design choice creates a privacy loophole such that advertisers can view the profiles of those who interact with their ads, thus identifying individuals that fulfill certain targeting criteria. This behavior is in contradiction to the promises made by the platforms to hide user data from advertisers. We conclude by suggesting design modifications that could provide users with transparency about the consequences of ad interaction to protect against unintentional disclosure. I Introduction Many social media companies own and operate advertising tools. With a wealth of users, and user data, they are well-equipped to help advertisers reach large numbers of people targeted across a diverse set of demographics and interests. There are known issues with targeted advertising, including that they exhibit discriminatory behavior, rely on sensitive user attributes, and target individual users [ 1 , 9 , 4 ] . Nevertheless, social media companies claim that they protect user privacy from advertisers. For example, TikTok claims that they share only “aggregated statistics and insights with users of our analytics services” [ 10 ] . Similarly, Faceook promises not to tell “advertisers who you are or sell your information to anyone” [ 6 ] . However, the design decisions made by platforms may impact their ability to uphold these privacy claims. TikTok and Meta, the focus of this study and two advertising platforms with significant user bases, have designed their infrastructure such that ads are interactive by default. Much like user-generated content, users can comment on or react to (e.g., “like,” “love”) an ad. This leaves a record of engagement which, if made visible to advertisers, would create a privacy loophole; since advertisers already know the potentially sensitive nature of the targeting criteria they have selected (e.g., household income, dependents’ age), this would allow them to learn the usernames of individuals that fit this criteria. To understand if this is possible, we ran advertisements on TikTok and Meta (Instagram and Facebook). We found that neither platform takes precautions to obscure the individual usernames of those who interact with ads from advertisers. TikTok shares the username of individuals who comment on (but not react to) ad content directly through the advertising interface. Meta shares the usernames of both users who comment and users who react to an ad. This means that users who interact with targeted ads are inadvertently identifying themselves to advertisers as individuals who meet specific targeting criteria. In summary, we demonstrate that social media platforms leak the personal data of users who interact with advertisements directly to advertisers, in contradiction to their own documentation. Platform Person Who Commented Visible Person Who Reacted Visible Interactivity Configurable Meta (Facebook + Instagram) ✓ ✓ ✻ TikTok ✓ ✓ TABLE I: Summary interaction data visibility and configurability across platforms ✻ The documentation suggests it is available in a limited capacity but it was not available to the papers’ authors. II Background Many social media platforms provide an infrastructure and user interface for advertisers to run marketing campaigns. These platforms make a set of design choices, and expose certain configuration options to advertisers, which dictate how users experience ads. As a baseline, advertisers typically need to specify a budget and creative copy (i.e., photo or video advertisements). After the initial setup, advertisers can use a provided user interface to view aggregated metrics about who viewed their advertisements, including demographics (e.g., age, gender, geography) and device information (e.g., iOS, Android). Two of these design choices, when combined, have significant privacy implications: targeting and interactivity. Targeting Platforms like Meta and TikTok allow advertisers to choose from a set of platform-defined targeting criteria to determine what characteristics about users should make them eligible to view an ad. A subset of the targeting options are potentially sensitive including age and gender (TikTok and Meta), household income (TikTok and Meta), U.S. postal code (Meta), age range of dependents (Meta), and interest and behavior-based targeting (TikTok and Meta). Targeted advertising alone can be privacy invasive; prior work has shown it can facilitate discriminatory targeting [ 9 ] , use sensitive user attributes [ 1 ] , and, with the right combination of attributes, target individual people [ 4 ] . Further, despite calls for user ad transparency and the introduction of features like Meta’s “Why Am I Seeing This Ad” [ 7 ] , users still have no way to adequately understand why they were targeted for a particular ad [ 3 ] , and thus what sensitive attribute groups they belong to. Interactivity Ad interactivity, which allows users to comment on an ad, is a default behavior in both Meta and TikTok. It is configurable in TikTok, although the user interface hides the toggle and the accompanying text description discourages advertisers from disabling this feature (Appendix Figure 2 ). Although the ability to disable comments in the Meta ads interface is reported to exist in a limited capacity, it was not available to our business account at the time of the study [ 8 ] . There are ways to limit interactions through the post, but not disable them entirely. III Case Studies: TikTok and Meta To find out if user information is visible to advertisers, we ran advertisements on TikTok and Meta using their UI business tools. Our advertisements were targeted to users over 18 in the United States. TikTok We ran our TikTok advertisement (Figure 1 ) on December 18-20, 2025, where it was viewed 11,667 times. Two users commented on the advertisement and one ”liked” it. Through the TikTok advertising interface, we were able to view the name and profile image of the users that left a comment. We were not able to view the username of the individual who ”liked” it. Fig. 1: Advertisement run through TikTok’s campaign platform, which targeted users over 18 in the United States. Meta We ran our Meta advertisement (nearly identical to Figure 1 ) on January 6-9, 2026, across Facebook and Instagram. It was shown to approximately 13,755 accounts on Facebook and 17,546 on Instagram. On Facebook, it received 4 reactions, all of which were visible to advertisers with a username and profile photo. On Instagram, it received 11 comments and 14 reactions. We were able to view the username and photo for 10 out of 11 comments and the username, account name, and profile photo for 8 of the 14 reactions. We do not know exactly why some of the comments/reactions were hidden, but we verified that the visible reactions and comments came from a mixture of public and private accounts. As our ad was not shared on either platform, we do not know how that action might impact data visibility but see it as an area for future work. Ethics Our study was reviewed and approved by our Institutional Review Board (IRB).

Read the complete text