‘Ethnic Affinity’ Targeting Isn’t Personalization. It’s Profiling With Better Branding.

Somewhere inside every social media platform’s ad-targeting dashboard sits a quiet admission: the platform believes it knows your race, even though you never told it. It doesn’t call this “race,” of course. It calls it “affinity,” or “cultural interest,” or “multicultural audience segments,” softer language for a harder mechanism, sorting hundreds of millions of people into racial and ethnic buckets based on inferred behavior, then selling advertisers the ability to target them, or just as easily, to exclude them entirely from ever seeing a given ad.

The clearest documented example of this practice traces back to a feature Facebook introduced years ago called “ethnic affinity”, which classified US users into categories including African American, Hispanic and Asian American. Facebook never collected race directly when people signed up for the platform. Instead, the categories were inferred from users’ behavior on the platform, pages liked, content engaged with, connections made, and packaged into a targeting tool available to any advertiser willing to pay for it. On paper, the tool was framed as a way for advertisers to reach audiences more likely to respond positively to culturally relevant content. In practice, it functioned as a fully operational proxy for race, built without ever asking a single user how they identified.

What made the tool dangerous wasn’t just that it existed, but how easily it could be weaponized. In 2016, ProPublica journalists purchased a housing-related ad and successfully excluded users categorized as African American, Asian American or Hispanic from ever seeing it, a move that violates the Fair Housing Act of 1968, which explicitly bans racial discrimination in housing advertising. The ad was approved by Facebook in roughly 15 minutes, with no apparent review flagging the exclusion as legally problematic. A year later, after Facebook publicly promised reforms in response to the initial backlash, ProPublica tried again, this time purchasing rental housing ads excluding not just racial affinity categories but also mothers of high schoolers, people interested in wheelchair ramps, Jewish users, Spanish speakers and expats from Argentina, categories that map closely onto legally protected classes under US anti-discrimination law. Every ad was approved again, this time in around three minutes, suggesting the platform’s promised safeguards had done little to change how the underlying system actually functioned.

The fallout eventually reached federal regulators. The Department of Housing and Urban Development, under then-Secretary Ben Carson, opened an investigation into whether Facebook’s ad platform violated fair housing law, a probe that placed the company under sustained federal scrutiny for the first time over its advertising infrastructure specifically. Facebook responded by retiring the specific “ethnic affinity” label in 2020 and rebranding the underlying categories as “multicultural affinity,” while adding certification requirements for advertisers running housing, employment or credit ads, categories considered especially sensitive under civil rights law. But researchers who tested the platform again the following year found the rebrand changed less than it appeared to on the surface. A 2023 Brookings study found that certain renamed “cultural interest” targeting categories were actually more precise at identifying users by race than the original tool had been, in one case reaching a nearly identical number of African American users under the new label as the old one had, even as the tool’s overlap with white users dropped sharply, evidence that the underlying racial inference had, if anything, gotten sharper rather than dismantled.

Also Read: Bad News From Home Never Really Stops for the Diaspora. Here’s What That Actually Does to You

The rebranding pattern didn’t stop with fair housing categories either. Political advertisers found their own uses for the same infrastructure, well beyond the retail and housing contexts the tool was originally marketed around. Investigators examining ads submitted to Congress as part of the inquiry into Russian interference in the 2016 election found that a significant share of the ads specifically targeted users interested in Black identity issues, including Black Lives Matter, the civil rights movement and figures like Martin Luther King Jr., using the exact same affinity infrastructure originally built for retailers selling sneakers and skincare. The overlap between commercial ad-targeting tools and politically motivated disinformation campaigns illustrated just how flexible, and how easily repurposed, this kind of racial inference technology could be once it existed at scale.

There is, to be fair, a version of this technology’s defenders would point to as legitimate and even beneficial. Small businesses genuinely built around serving a specific cultural community, hair care lines, ethnic grocery chains, faith-based organizations, can reach their actual customer base more efficiently through affinity targeting than through blunt, undifferentiated advertising that wastes budget reaching audiences unlikely to convert. The tool isn’t inherently exploitative in every application. It becomes a civil rights problem specifically at the moment it’s flipped from inclusion to exclusion, from “show this to people who might want it” to “hide this from people who happen to belong to a group,” a distinction that sounds simple in theory but has proven remarkably difficult for platforms to enforce consistently in practice.

That flip is exactly what makes ethnic affinity targeting structurally different from ordinary market segmentation, and why researchers continue to study how the people being categorized actually experience it. A 2023 study surveying over 1,000 Black social media users found that people’s comfort with being targeted by inferred ethnicity data depended heavily on how transparent the platform was about collecting and using that data in the first place, and that a lack of transparency pushed users toward defensive coping behaviors, avoidance, distrust, disengagement, rather than the increased relevance the technology promises advertisers in pitch decks and case studies. In other words, the industry’s own research suggests the people being sorted into these categories generally didn’t ask to be, don’t fully know how it’s happening, and respond by pulling back rather than leaning in, undermining the very premise that this kind of targeting improves the user experience it claims to personalize.

None of this requires believing platforms set out to build discriminatory infrastructure on purpose. What the record shows instead is a consistent pattern: a tool gets built, gets caught enabling discrimination, gets renamed rather than dismantled, and often becomes more precise at doing the exact thing it was renamed to stop doing. Calling it “affinity” instead of “race” doesn’t change what the system is actually sorting people by underneath the surface. It just makes the sorting easier to defend in a press statement, and considerably harder for the people being sorted to ever fully see or contest.