Investigation finds Google uses a secret blocklist to stop advertisers from making YouTube ad campaigns for hate terms; it worked less than a third of the time
Many well-known White supremacist and White nationalist terms and slogans were not blocked — First of two parts.
Context & Ripple Effects
This is the fourth act in a four-year arc of failed promises on YouTube ad safety. It began when the UK government, Guardian and Sainsbury's pulled ads after the Times found them running against extremist videos, prompting Google to promise fixes; subsequent reporting showed those fixes didn't hold — a [[a:922335|BuzzFeed investigation found Google was still letting advertisers target racist keywords and even suggesting more of them]], and a year later a [[a:928746|CNNMoney probe caught 300+ organizations' ads on extremist channels despite their use of the sensitive-subject exclusion filter]].
What changed today is that Google's response turns out to have been a secret blocklist — undisclosed to advertisers, who believed the visible filters were protecting them — and it caught fewer than one-third of tested white supremacist and nationalist terms. A companion piece shows the failure is also asymmetric: YouTube blocks 'Black Lives Matter' as an ad-targeting term while leaving phrases like 'White power' open.
First-order effects
- Advertisers relying on YouTube's sensitive-subject exclusion filter are learning it never guaranteed protection — the real enforcement ran through a hidden list that missed most tested hate terms, so brands like the ones that boycotted in 2017 are exposed again without knowing it.
- Google now faces documented evidence that its moderation tooling both under-blocks hate terms and over-blocks civil-rights language, since its own system treats 'Black Lives Matter' as untargetable while 'White power' passes.
Second-order effects
- The 2017-style advertiser exodus becomes the obvious pressure valve: the same coalition of governments and big-spending brands that forced Google's original promises has fresh grounds to demand auditable brand-safety guarantees rather than trust in opaque filters.
- Third-party brand-verification vendors gain a selling point — if Google's first-party filter demonstrably fails two-thirds of tests, buyers have reason to pay for independent verification instead.
Third-order effects
- If the pattern across 2017–2021 holds, keyword-based moderation at ad-platform scale is structurally inadequate: each investigative cycle produces a patch, then a new failure mode, pointing toward external auditing or regulatory standards for ad-placement claims rather than self-certification.
- The asymmetry — blocking racial-justice targeting while missing supremacist terms — suggests these systems encode default assumptions that will keep drawing scrutiny until the blocklists themselves become accountable objects.
The trend: Four years after the 2017 boycott forced Google to promise cleaner ad placement, each investigation shows platform self-policing lagging behind its public commitments, pushing brand safety toward independent verification and regulation.