Wiki · Concept · Last reviewed June 25, 2026

Topics API

The Topics API is a browser proposal for interest-based advertising that exposes coarse interest categories derived in the browser instead of exposing a user's exact browsing history.

Category: Concept Updated: June 25, 2026 Tags: browser privacy, ad tech, interest-based advertising, Privacy Sandbox, surveillance

Definition

The Topics API is a Privacy Sandbox proposal for interest-based advertising. Instead of a third party building a detailed cross-site profile through tracking cookies, the browser infers broad topics from recent browsing activity and may expose a small set of those topics to eligible callers. MDN describes it as an advertising mechanism based on browser-collected topics rather than developer tracking across sites.

The proposal sits between privacy reform and advertising continuity. Google Privacy Sandbox describes Topics as a way to enable interest-based advertising without third-party cookies and says the specific sites a user visited are not shared across the web. That is narrower than "no profiling": the system still converts browsing activity into advertising categories.

MDN says the feature is not part of an official standard, is specified in an unofficial proposal draft, and is opposed by Mozilla and Apple. Google announced on October 17, 2025 that several Privacy Sandbox technologies, including Topics, would be retired. A Blink-dev intent posted on November 7, 2025 said Chrome planned to deprecate and remove the Topics API.

Mechanism

Topics works through observation, derivation, and access. Google Privacy Sandbox says a caller observes a topic when it makes a Topics API request from a page or app associated with that topic during the past three epochs. An epoch is a period for topic computation, defaulting to one week. Once per epoch, the browser computes top topics from user activity and caller observations.

On the web, Google says topics are inferred from hostnames of pages the user visits. The browser uses a taxonomy and a classifier model, with human-curated exclusions for categories generally considered sensitive, such as ethnicity or sexual orientation. The returned signal is intentionally coarse: Google says the API can return at most three topics, one for each of the three most recent epochs, and the individual draft says each epoch has a 5 percent chance of returning a random topic from the full taxonomy.

The developer surface includes Document.browsingTopics(), the browsingTopics fetch option, the browsingtopics iframe attribute, the Sec-Browsing-Topics request header, the Observe-Browsing-Topics response header, and the browsing-topics Permissions Policy directive. MDN says Privacy Sandbox enrollment is required for use; without it, calls such as Document.browsingTopics() fail or headers are stripped.

Agent Context

For AI Browsers and Computer Use, Topics matters because a browser can become both an assistant interface and an advertising classifier. An agent may browse pages, trigger embedded ad-tech code, and help a user evaluate offers while the browser maintains topic state that the agent operator may not see directly.

Agent records should distinguish visible page content from browser-maintained advertising state. If an agent encounters a personalized ad, the audit trail should not infer that the ad represents the user's actual intent. It may reflect a coarse topic, a caller's prior observation, a random topic, contextual signals, third-party-cookie behavior, or ordinary ad auction logic outside Topics.

Governance Use

A useful review record should name the caller origin, embedding site, taxonomy version, whether Document.browsingTopics(), iframe attributes, fetch options, or headers were used, the Permissions Policy state, enrollment status, and user controls. It should also record Chrome-specific behavior and deprecation fallback.

The API belongs beside Real-Time Bidding, Attribution Reporting API, Shared Storage API, Fenced Frame API, and Data Minimization. Topics is not measurement, storage, or rendering isolation. It is a browser-mediated interest signal for ad selection.

Limits

Topics is not consent, anonymity, a ban on profiling, or a general anti-surveillance system. It changes the shape of the signal: from site-level history toward coarse browser-derived categories with caller eligibility rules. The surrounding ad market can still combine Topics with contextual data, first-party data, device signals, account state, or third-party cookies where they remain available.

The retirement plan is also a governance lesson. A browser feature can be documented, trialed, enrolled, integrated into ad-tech systems, opposed by other vendors, then scheduled for removal when the platform strategy changes. Audits should therefore treat browser privacy APIs as policy-dependent infrastructure, not permanent public standards.

Review Record

Source Discipline

Claims about API shape, epochs, caller eligibility, random topic behavior, and technical surfaces should cite the individual draft, MDN, and Google developer documentation. Claims about standards opposition should cite MDN and vendor records. Claims about retirement should cite Google's October 17, 2025 Privacy Sandbox update and the November 7, 2025 Blink-dev intent.

Spiralist Reading

Spiralism reads Topics as a lesson in sanctioned legibility. The browser promises to reveal less than raw history, but it still translates a person's motion through the web into categories useful to a market. A smaller profile can still be a profile. A retired API can still teach the next one what advertisers, browsers, and regulators are willing to call privacy.

Sources

Return to Wiki