The promise: relevance without an ID
Cohort-based targeting means buying an ad against a group of browsers or readers who share an inferred trait—such as car intenders or high-intent shoppers—rather than a cookie ID tied to a stored individual profile. It emerged because third-party cookies and mobile ad IDs were under pressure from browsers, operating systems and regulators. A cohort was meant to be a middle path: enough signal to price an impression, but not enough to identify one person.
That explains why marketers should care. Safari and Firefox traffic, consent-refusing users, connected TV and much app inventory are effectively ID-less. Cohorts were supposed to keep those impressions addressable.
How the pieces fit together
Most systems follow a familiar flow. First, behaviour is collected. Then a rulebook or algorithm sorts that behaviour into groups. A privacy control—usually a minimum group size, or k-anonymity—limits re-identification. Finally, only the group label travels to the buyer.
The key split is who does the grouping:
- Browser-side: computed on-device from local history, like Google’s FLoC and later Topics.
- Seller-side: a publisher or data partner classifies first-party audiences and passes taxonomy IDs in the bid request.
- Platform-side: walled gardens and clean rooms share a segment name and aggregate reporting, but not member lists.
If an identifier travels with a cohort label, the label simply enriches a profile. The privacy promise only holds when the two stay separate. In OpenRTB, seller-side cohort signals can ride in user.data using IAB Tech Lab’s segtax codes, with Audience Taxonomy 1.1 carrying more than 1,600 attribute nodes.
From FLoC to Topics to retirement
Chrome’s FLoC origin trial ran through versions 89 to 91 in 2021 and produced 33,872 cohorts from browsing history. Google claimed at the time that simulations showed at least 95% of cookie-based conversions per dollar. Independent testing later painted a different picture: CMA-supervised tests found publisher revenue per impression around 30% lower without third-party cookies even with the APIs available, while Criteo estimated declines up to 78%.
Google replaced FLoC with Topics in January 2022, but Topics also faced doubts over bluntness and signal quality. Finally, after keeping third-party cookies in April 2025, Google announced the retirement of Topics and most Privacy Sandbox APIs in October 2025. The browser-cohort era largely ended before it scaled.
What that means for your audience strategy
For performance marketers, the useful frame is simple: treat cohorts as a reach layer, not a replacement for first-party data. The mechanics have moved to seller-defined audiences—now being positioned as Curated Audiences—plus connected TV, on-device modelling and clean rooms.
Before you spend, ask three questions: Where is the cohort built? What taxonomy or rule set defines it? Can you verify the label against your own conversion data? A games enthusiasts segment should not be repackaged as high net worth individuals, and self-attested labels carry that exact risk.
The most practical next step is to run small, measurable experiments with seller-defined audience IDs from trusted publishers or supply partners, while keeping your own first-party data as the control group. Watch incremental revenue, not just CPM.
Keep one eye on Australia, where a draft bill treating group-level targeting as direct marketing closes for submissions on September 18, 2026. If it passes in a strong form, cohort strategies built from personal information may need stricter consent and governance.
Source: PPC Land
