The Privacy Sandbox initiative aims to create new technologies that both preserve people’s privacy online and give companies and developers the tools to build thriving digital businesses, which keeps the web open and accessible to everyone. The Privacy Sandbox is an ongoing collaborative effort to build alternatives to third-party cookies, by incorporating ideas from across the industry into new proposals that are tested, publicly discussed, and iterated upon.
One of the Privacy Sandbox proposals being tested in Chrome is called the Topics API. Topics proposes a new, privacy-preserving signal to help indicate a user’s interests without requiring third party cookies.
With the Topics API, a user’s browser would gather topics (for example, "Country Music," "Make-Up & Cosmetics," or "Vegetarian Cuisine") associated with that user based on their browsing activity during a period of time known as an epoch, currently proposed to be one week. The topic selected for each epoch would be randomly selected from the user's top 5 topics, with random noise added for privacy, for that time period.
How Google Ads and Display & Video 360 use Topics to help deliver relevant ads
When available, Google Ads and Display & Video 360 will receive Topics signals in eligible bid requests from publishers or their supply-side platforms (SSPs). Google Ads and Display & Video 360 will then use a combination of the following privacy-preserving signals to serve interest-based ads:
- Contextual information
- Topics API
- Publisher first-party ID identifiers (for example, Publisher Provided and Exchange Provided IDs)
- New machine learning insights using the signals above
How Google Ads and Display & Video 360 are testing the Topics API
In mid-2022, Google Ads and Display & Video 360 began using Topics signals to inform ad delivery on a small percentage of interest-based ad campaign volume. The earliest stages of testing focused on functional testing to determine whether, for example, Google Ads and Display & Video 360 can successfully take and analyze Topics signals.
Later, Google Ads and Display & Video 360 teams moved to utility testing to learn how effectively a combination of privacy-preserving signals such as contextual information, the Topics API, and first-party identifiers such as Publisher Provided IDs, can inform relevant ad delivery. Testing a more holistic solution that includes other relevant signals (versus testing Topics alone) is more representative of the solutions and performance levels we expect in our ads platforms after Chrome phases out third-party cookies.
This utility testing included a second iteration of the Interest-Based Ads alpha, which ran in Q4 2022, with a set of global advertisers across many industry verticals who met strict eligibility criteria. They were able to compare audience campaign performance when relying on a combination of new privacy first signals, including simulated Topics (due to limited traffic in Chrome’s Origin Trials at the time), versus third-party cookies to inform ad delivery.
Our teams also ran a separate experiment, in which we aimed to understand how Google’s interest-based audience (IBA) solutions perform when they rely on a combination of privacy-preserving signals. These signals included contextual information, the Topics API from the Privacy Sandbox and first-party identifiers such as Publisher Provided IDs, in comparison to third-party cookies. This experiment ran on a small slice of global traffic using interest-based audience solutions like affinity, in-market, custom audiences, and demographic segments; and included both Google Ads campaigns and Display & Video 360 line items, across industries.
The results were promising: the experiment showed that when using IBA solutions with privacy-preserving signals on the display network, Google Display Ads advertiser spending on IBA — as a proxy for scale reached — decreased by 2-7% compared to third-party-cookie-based results. For conversions per dollar, as a proxy for return on investment, the decrease was 1-3%. Finally, we also observed that click-through rates remained within 90% of the status quo. And we observed similar performance for Display & Video 360. To accompany the results, we published a detailed discussion of the experiment methodology.
In 2024, a new experiment indicated that the Privacy Sandbox APIs, in combination with other signals, can contribute to advertiser and publisher performance recovery. For example, using spend as a proxy for the potential scale, we observed 89% recovery for GDA and 86% for Display & Video 360. It’s important to remember that these results — while encouraging — should not be considered as an unequivocal indicator of Google’s IBA performance in a world where less observable data is available.