Making Sense of Private Advertising: A Principled Approach to a Complex Ecosystem

Kyle Hogan, Alishah Chator, Gabriel Kaptchuk, Mayank Varia, Srinivas Devadas · 2025

Read Making Sense of Private Advertising: A Principled Approach to a Complex Ecosystem by Kyle Hogan, Alishah Chator, Gabriel Kaptchuk, Mayank Varia, Srinivas Devadas (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-2512-20583.txt.

Introduction

Kyle Hogan and co-authors model online advertising end to end and find its privacy story does not add up. Proposals for private advertising treat targeting and engagement measurement separately. Privacy guarantees that look fine for each piece in isolation can fail when the pieces compose, letting advertisers extract new information about their audiences.

The authors then prove a sobering limit: perfect privacy is impossible for any even minimally useful advertising ecosystem, because advertisers expect market research from the results. Leakage is inherent, so the question becomes what privacy should realistically mean. Building on the notion of sensitive data in context, they argue that advertising subsystems need fundamentally new designs. Only end-to-end thinking keeps the whole system's promises aligned with people's actual privacy desires.

Catalog record

CREATORKyle Hogan, Alishah Chator, Gabriel Kaptchuk, Mayank Varia, Srinivas Devadas
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE112 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-20583

Browse the full library · About the Tower of Babel

Full text

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

Making Sense of Private Advertising: A Principled Approach to a Complex Ecosystem Kyle Hogan, Alishah Chator, Gabriel Kaptchuk, Mayank Varia, Srinivas Devadas Abstract . In this work, we model the end-to-end pipeline of the advertising ecosystem, allowing us to identify two main issues with the current trajectory of private advertising proposals. First, prior work has largely considered ad targeting and engagement metrics individually rather than in composition. This has resulted in privacy notions that, while reasonable for each protocol in isolation, fail to compose to a natural notion of privacy for the ecosystem as a whole, permitting advertisers to extract new information about the audience of their advertisements. The second issue serves to explain the first: we prove that perfect privacy is impossible for any, even minimally, useful advertising ecosystem, due to the advertisers’ expectation of conducting market research on the results. Having demonstrated that leakage is inherent in advertising, we re-examine what privacy could realistically mean in advertising, building on the well-established notion of sensitive data in a specific context. We identify that fundamentally new approaches are needed when designing privacy-preserving advertising subsystems in order to ensure that the privacy properties of the end-to-end advertising system are well aligned with people’s privacy desires. 1. Introduction Behavioral advertising, in which people are preferentially shown advertisements that align with their interests and demographics, has become the financial backbone of the internet and the default business model for large swaths of the technology sector. This advertising ecosystem—and the user-tracking infrastructure that powers it—are widely known to be privacy invasive ( 50 ) , not only due to the collection of sensitive, personal data, but also because of the ways in which that data is used ( 36 ; 17 ; 95 ; 96 ) . A great deal of research has focused on the harms caused to people through the use of their data for the targeting of advertisements. Targeting algorithms are designed to maximize profit for the constituents of the advertising industry, not to serve the best interests of the viewer, and prior studies have shown that behavioral targeting is used to perpetuate harmful biases ( 54 ; 35 ; 88 ; 75 ; 3 ) , facilitate the spread of disinformation ( 15 ; 102 ; 34 ; 53 ) , and exploit sensitive information such as mental health data to target vulnerable populations ( 17 ; 24 ; 98 ; 6 ) . While targeting has been the most widely-researched component of behavioral advertising, matching ads to users is not the only functionality of the ecosystem. Advertisers expect to conduct market research using the relative success of their different advertisements. To do so, they demand metrics on how users responded to ads—did they buy something after viewing the ad? Subscribe to the advertiser’s mailing list? If so, which ad drove this engagement? Ad networks keep records of this information and submit it back to advertisers, allowing them to improve their understanding about the preferences of their consumers and refine future ad campaigns. Targeted advertising is not unique to digital advertising: print advertisements have long been targeting their ads towards specific groups of people using both the context in which the ad will be displayed (e.g., the magazine or billboard on which to advertise) as well as cues within the ad material itself (e.g., visuals, audio, etc.). A famous example is the 90’s Subaru campaign that was targeted towards the LGBT community by including, e.g., queer-coded license plates on the depicted cars ( 39 ; 65 ) . The shift to digital advertising is, therefore, not a change in type , but a change in magnitude and precision . Digital advertising, with its specific target audiences and accurate attribution of user behavior to associated ad views, allows advertisers to conduct market research that is far more invasive than was possible with print media, calling into question the ethics of campaigns that focus on sensitive audiences. This is the case not only due to the well-understood harms of collecting sensitive data for targeting ads ( 85 ) , but also because it is currently unclear what or how much information advertisers are able to learn about these sensitive audiences from the metrics released on their ad campaigns. In response, researchers and technology companies have proposed a shift towards “privacy-preserving advertising systems,” a collection of proposals ( 106 ; 107 ; 74 ; 13 ; 46 ; 8 ; 48 ; 92 ; 56 ; 84 ; 51 ; 52 ; 83 ; 90 ; 42 ; 100 ; 73 ) that aim to maintain the existing advertising business model while making the process of (1) targeting advertisements and (2) reporting metrics on advertisement efficacy in a privacy-preserving manner. Proponents of privacy-preserving advertising systems have made significant progress in refining the design of these systems, some of which have even been deployed in popular browsers ( 42 ; 69 ; 100 ) . Unpacking “privacy” within advertising. In this work, we take a step back to re-examine “privacy-preserving advertising systems” from first principles. This choice is motivated by the desire to better understand what is possible to achieve when adding privacy to advertising when, largely, the concept feels like an oxymoron. Ultimately, we find that this initial reaction is not far off: any useful behavioral advertising ecosystem must necessarily permit advertisers to extract information about end users, regardless of what privacy protections are in place. Leveraging the language of ideal functionalities, we give an implementation-independent modeling of privacy-preserving advertising that focuses on the minimal functionality required by the ecosystem. Notably, our model departs from prior work in that it focuses on the entire end-to-end advertising pipeline, with a particular emphasis on the ways that privacy-preserving targeting and privacy-preserving metrics interact with one another and form a feedback loop. By seeing privacy-preserving advertising in this way, we are able to identify real-world advertising use cases in which common notions of privacy for targeting and metrics fail to compose satisfyingly, which undermines natural privacy guarantees for the end-to-end system despite targeting and metrics protocols independently achieving reasonable standards for privacy. We emphasize that this composition failure is not merely the result of specific, ill-designed protocols from early research; instead, it is fundamental to the nature of targeted advertising itself. Looking further: we also examine how to consider privacy formally in the context of advertising. While perfect privacy may not be possible for advertising, we observe that it is also not necessarily required or even desirable. For instance, while contextual advertising (where ads are targeted only to the context in which they will be displayed, i.e., the website or article) can suffer from the same problems as behavioral advertising in the worst case, it is still strongly preferred as an alternative to behavioral advertising. While the privacy provided by contextual advertising may be imperfect, it is likely “good enough” ( 63 ; 99 ) . Hence, using the language of information leakage alone makes it difficult to distinguish between advertising systems that are widely considered “invasive” and those that are not. We additionally observe that a narrow focus on building mechanisms to regulate—without eliminating—information leakage, like differential privacy ( 37 ) , risk treating all

Read the complete text