Updates to the Topics taxonomy and filtering mechanisms, along with speed improvements and enhanced user controls.
Update: November 8, 2023
In June, we outlined several enhancements to the Topics API. We closed by reiterating our commitment to continue to listen to ecosystem feedback. Today, we are announcing a further enhancement to the Topics API, in response to that feedback.
Top topics selection
The initial Topics API proposal selected users' top five weekly topics based on the frequency by which users interacted with each topic, on participating websites. We received feedback that this resulted in the API often returning topics that are less useful for ad relevance such as "News" or "Arts & Entertainment". We explored many solutions, including allowing callers to set a priority list, ranking by inverse frequency on the web, by ad clicks observed by Chrome, and other approaches.
The most promising approach we've seen is to integrate Topics utility feedback from the advertising ecosystem directly. We have done so by introducing the concept of "high utility" buckets. Chrome places each of the 22 root topics (those without an ancestor) from the taxonomy into one of two buckets indicating higher or standard utility for the ecosystem overall. All descendants of the root topics inherit the same bucket assignment from their parent. The assignment of root topics to buckets is based on input about utility we received from companies across the ecosystem when crafting our improved taxonomy.
Considering the above, the updated top topics selection methodology is as follows:
- At the end of each epoch, Chrome converts participating hostnames from the user's browsing history into topics.
- First, topics are sorted by bucket, and then by frequency. That is, if two topics are in the same bucket but have different frequency, the higher frequency topic is sorted higher.
- Lastly, Chrome selects the top five as the user's top topics for that epoch, which are eligible to be shared with callers.
We expect the "high utility" bucket assignments of specific topics to evolve over time based on feedback from the broader ecosystem, which can be provided by creating an issue on the Topics repository on GitHub. The update will be available beginning this quarter (Q4 2023).
Update: June 15, 2023
Over a year ago, we announced the Topics API, a proposal for interest-based advertising. Topics is designed to enable websites to serve relevant ads in a privacy-preserving manner, without resorting to covert tracking techniques, like browser fingerprinting. Topics utilizes several techniques to preserve user privacy, including reducing data, noising data, excluding potentially sensitive topics, and processing data on-device. Combined, these changes make Topics a significant step forward for user privacy compared to third-party cookies.
When we first offered Topics, we were clear that this was an initial proposal, and we asked the ecosystem to provide input to help improve it. Since our announcement, we have been listening carefully to their suggestions. Today, we're excited to share some of the latest improvements to the Topics API. We believe these changes will make Topics even more useful to the digital advertising industry, without compromising user privacy.
Alongside the initial Topics API announcement, we proposed a taxonomy designed for testing. The taxonomy is the list of available topics that may be returned by the API. We repeatedly received feedback that the testing taxonomy did not represent topics the advertising industry cared most about, so today we're announcing an improved taxonomy.
When crafting this new taxonomy, we saw deep engagement from companies across the ecosystem, like Raptive (formerly CafeMedia) and Criteo. It removes categories we've heard are less useful, in favor of categories that better match advertiser interests, while maintaining our commitment to exclude potentially sensitive topics. We have added 280 commercially focused categories, like "Athletic Apparel", "Mattresses", and "Luxury Travel," while removing 160 categories including topics like "Civil Engineering" and "Equestrian" which don't add much commercial value for ad selection on most sites. The new taxonomy has 469 topics, compared to 349 for the previous version. We chose to limit the taxonomy's size, to protect against re-identification risk.
We expect the taxonomy to evolve over time, and for governance of the taxonomy to eventually transition to an external party representing stakeholders from across the industry. We encourage the ecosystem to review the latest taxonomy and provide feedback on the changes.
One of many privacy-preserving features of Topics is the per-caller filtering requirement. This feature ensures that callers can only receive topics that they've observed the user visit in the past, rather than provide the topics to any caller regardless of their level of interaction with the user. For example, if a caller observes a user visit a site about news, but not shopping, that caller cannot learn that the user is interested in shopping.
Consider the topic "Boots," which is fully expressed as "/Shopping/Apparel/Footwear/Boots." "Shopping" and "Apparel" are ancestors of "Boots." Chrome has updated the definition of "observation" to include all ancestors of a given topic. Previously, in order for a caller to observe "Shopping" or "Apparel" a caller must have observed a user visit a page with that topic. With this change, if "Boots" is observed, then all ancestors (such as "Shopping" and "Apparel") of that topic are recorded as observed as well.
This change increases the likelihood sites will receive topics information, without impacting the API's privacy since the topic's ancestors were already known to the caller.
With Topics, users can view and control how their cross-site data is used to personalize ads in a more intuitive and accessible manner compared to tracking mechanisms like third-party cookies. In fact, participants in user research conducted by Google reported a significantly better privacy experience and feeling of control when introduced to Topics user controls, compared to current third-party cookie controls.
Today we're announcing our plans to give users even greater control over which topics are associated with them. Specifically, users will be able to proactively block topics. This means users will be able to curate the set of available topics they are interested in by removing selected topics. This change, coming by early next year, will give users even more control over their privacy and make the Topics API even more user-friendly.
Last year, we announced support for Topics via headers, in requests initiated via Fetch and (temporarily) XHR . Recently, we announced that we plan to extend support to request headers for iframes. These changes will improve the performance of Topics, limiting potential negative impacts on developers and users.
We are excited about these updates to the Topics API and believe that they not only will make it more effective for advertisers and keep ads relevant for people, but still preserve privacy. Per-caller filtering updates and speed improvements are already available in Chrome 114. Taxonomy updates will be available in Q3 2023. User controls updates will be available by early next year. We are committed to continuing to listen to ecosystem feedback as we build new, more private technologies for the web.