Skip to main content
A cohort is a privacy-safe representation of an audience.Traditionally, user IDs like third-party cookies have been used to communicate information about audiences and activate (target) them. A publisher might communicate behavioral data for a user to an ad server by attaching the data to a cookie ID. The ad server uses the cookie ID to record information about the user—both information from the publisher and other data parties they work with—and the ad server can later target an ad for a request from that user based on the information it has attached to their cookie ID.With cohorts, a code represents the group of users that fall into an audience. In the example above, when the publisher requests an ad from the ad server for the user, it includes the cohort code in the request, but not a user ID. The ad server can use the cohort code(s) to decision on the ad to respond with, but no other publisher or user data is leaked and a profile of the user cannot be built.By removing user identifiers from the equation, users' privacy is protected, the publisher's data cannot be targeted off of their site/app, and publishers can provide reachability (scale) to the ecosystem since cohorts can be built & activated in environments like Apple Safari, Mozilla Firefox, and iOS where third-party cookies and other deterministic identifiers are prohibited.

​
Cohort types

Permutive supports several types of cohorts to address different targeting needs:
  • Custom cohorts: Audiences defined by publishers using their first-party behavioral data, such as users who have read articles in a specific category or engaged with particular content.
  • Contextual & affinity cohorts: Audiences based on page-level signals rather than user-level data. Contextual cohorts target content that matches criteria such as categories, keywords, or sentiment derived from NLP classification. Affinity cohorts extend this by leveraging insights from consented users—Permutive calculates an affinity score for each URL based on how likely users in a specific custom cohort are to engage with that content compared to the site average, allowing publishers to apply behavioral insights to contextual targeting without requiring user consent.
  • Modeled cohorts: Audiences expanded through lookalike modeling or classification models, allowing publishers to reach users who share characteristics with a seed audience.
  • Standard cohorts: Pre-built audiences based on common taxonomies (such as IAB categories) that provide consistent targeting options across publishers.