14 min read
TL;DR Tap to expand the short version
- Blocklists fail on context, not on judgement. A film shoot story gets blocked for the word shot. Coverage of the band Arcade Fire gets blocked for the word fire.
- Lists also decay. Brands that added Paris after the 2019 Notre Dame fire were still blocking Paris during the 2024 Olympics, because nobody went back and removed it.
- Safety and suitability are different standards. The industry framework separates a floor of universally harmful content from graded risk levels that are a question of brand preference. Most blocklists collapse the two.
- The audit is four steps and starts with exporting your exclusion list, which most buyers have never read.
Almost every programmatic campaign runs against a brand safety blocklist, sometimes called a keyword blocklist or ad exclusion list. Almost nobody has read theirs.
They tend to arrive as a vendor default or get inherited from whoever set the account up, then quietly grow as terms get added after an incident and never removed. Nobody is accountable for the list itself, which means nobody audits it, which means nobody knows what it is currently costing.
There is now a public number for what that costs, and it is larger than most buyers would guess.
How Keyword Blocklists Work
A keyword blocklist is a list of terms. If a page or URL contains one, your ad does not serve there. The term match determines whether the ad can appear.
It has no model of meaning. It does not know whether the word appears in a report on a shooting or a report on a film shoot. Industry lists typically run to thousands of terms and include words like kill, death, shoot and fire, Journalism Institute analysis.
The appeal is obvious. It is cheap, it is instant, and it produces a defensible answer if anyone asks whether you took precautions. The cost is invisible, because a blocked impression does not show up anywhere as a loss. You simply never see the inventory you did not buy.
Reuters Test Results
In 2026 Integral Ad Science and Reuters ran a test together and published the results, which is unusual. Verification vendors and publishers do not normally co-author a critique of a tool the vendor sells.
The study used a specific method. The study looked only at Reuters content that IAS had already classified as low or medium risk under its brand suitability measurement, meaning inventory its contextual targeting product would have cleared and that would have qualified for brand suitable private marketplace deals. Then it asked a simple question, how much of that already cleared inventory would a typical keyword blocklist have blocked anyway?
AdExchanger report, the answers were
- News section, tested in June, 54 percent of URLs cleared by contextual targeting would have triggered a typical keyword blocklist.
- Lifestyle section, tested in September, 4.27 percent of cleared URLs would have triggered blocking.
- On that lifestyle section, those pages carried 13.5 percent of the section’s ad impressions during the test period.
The third number deserves a closer look. A small share of URLs can carry a large share of impressions, because traffic is never evenly distributed. Blocking four percent of pages does not cost you four percent of reach if the blocked pages happen to be the popular ones.
The researchers state two caveats. The study did not run test campaigns, so it measured what would have been blocked rather than observing live ad serving. And on the news side, the 54 percent figure describes URLs rather than impressions, so it is not a claim that half of Reuters news impressions would have vanished.
Two Failure Modes
The examples from the study show this because they show the failure is structural rather than occasional.
Failure one is context. A Reuters lifestyle article about a film shoot for The Devil Wears Prada 2 would have been blocked because the URL contained the word shot. Coverage of the band Arcade Fire gets caught by the word fire. Neither page is unsafe by any definition anyone would defend out loud.
Failure two is decay. This one is more expensive and gets discussed less. Reuters Digital general manager Phil Andraos described how blocking a term tends to pull associated terms in with it. After the 2019 Notre Dame fire, brands blocking fire also added Paris. Blocklists are not reviewed on any schedule, so those same brands were still blocking Paris five years later, during the 2024 Olympics, which is the single largest advertising event in the calendar.
Nobody decided to block the Olympics. The decision was made in 2019 about a cathedral, and then nobody revisited it.
| Measure | Reported figure | How to read it |
|---|---|---|
| News URLs, June test | 54% | Share of contextually cleared URLs that a typical keyword blocklist would have blocked. This is not a share of news impressions. |
| Lifestyle URLs, September test | 4.27% | Share of contextually cleared URLs that would have triggered blocking. |
| Impressions on those lifestyle pages | 13.5% | Share of the section’s ad impressions carried by those pages during the test. It is not the same measure as the URL share. |
How Common Is Overblocking?
The Reuters test is recent. The scale estimate is older and comes from a 2019 study by the cybersecurity firm CHEQ, conducted with University of Baltimore economist Roberto Cavazos, titled The Economic Cost of Keyword Blacklists for Online Publishers.
Its findings, MediaPost report
- US publishers lost an estimated 2.8 billion dollars in 2019 to incorrect blocking, roughly one in every four dollars of programmatic news ad revenue, against an estimated 12 billion dollars spent annually on US online news advertising.
- Only 9 of the 100 most read New York Times articles that year were considered brand safe by keyword blocklists.
- 80 percent of ads served to premium publishers were subject to keyword blocklists.
- The distribution was uneven. Digiday reported that on the LGBTQ publisher PinkNews, 73 percent of safe articles were blocked, driven by terms like lesbian and same sex sitting on standard industry lists.
The industry disagrees on how much of this is still happening, which limits the conclusions we can draw.
IAS chief product officer Srishti Gupta told AdExchanger that the share of ad impressions running through the IAS platform blocked as a result of the exclusive use of keyword blocking is now just 0.25 percent, and that education has made the practice less popular over time. Andraos, on the publisher side, said advertisers appear to use extensive keyword blocklists in the majority of cases, and that the impact on revenue is significant for news and non news publishers alike.
Both statements can be true, and the gap between them is instructive. The IAS figure measures campaigns relying on keyword blocking alone. Gupta noted separately that many advertisers still use some form of keyword blocking even when they have also adopted more sophisticated tools. So a campaign running semantic contextual targeting plus a legacy keyword list would not appear in the 0.25 percent, while the legacy list keeps blocking things underneath.
Check your account instead of assuming the industry figure applies to it.
Safety and Suitability
Most buying practice has yet to catch up with a distinction the industry established years ago.
In 2019 the Global Alliance for Responsible Media published its Brand-safety framework. IAB explanation, the framework maps eleven content categories against four risk levels, low, medium, high and floor, and is activated through the Tech Lab’s content taxonomy.
The distinction it draws is the useful part
- The floor is content no advertiser should appear beside. Placement beside such content is a safety question and it is not a question of taste.
- Everything above the floor is graded risk, and where a given brand draws its line is a preference. A funeral insurer and a theme park will reasonably answer differently about the same page.
A keyword blocklist cannot make that distinction, because it operates on strings rather than categories. Every term on it is treated as floor level regardless of what the brand thinks.
There is also a historical detail. GARM was discontinued on 8 August 2024, following a US House Judiciary Committee investigation and antitrust allegations raised by X. The organisation is gone. The framework it produced remains in use, embedded in verification platforms and in the IAB Tech Lab taxonomy, and the Brand Safety Floor is MRC guidance on content level context and brand safety verification. So the standard did not die with the body that wrote it.
Who Bears the Cost?
It rarely lands on the person who set the list up. By the time the decay has compounded for two or three years, that person has usually moved teams or moved on, and the cost has quietly shifted to three other people who never chose it.
The publisher on the other side loses reach they never knew they lost. A newsroom covering the Olympics or a court case has no visibility into which advertisers are blocking them over a term from an unrelated 2019 incident. The demonetisation is invisible to the very people it demonetises, which is exactly why the Reuters test above needed a verification vendor’s cooperation to surface it at all.
The media buyer inherits a number they cannot explain. When a client asks why cost per impression crept up or why reach underperforms the plan, a bloated blocklist is rarely the first place anyone looks, because it does not show up as a line item. It shows up as vague underperformance that gets blamed on the platform, the creative, or the season.
The brand itself pays twice. Once in the reach it is quietly forfeiting, and again in the false confidence that a long exclusion list equals a strong safety posture. A list nobody has read is not a control. It is an unexamined assumption with a compliance-sounding name.
None of these three people did anything wrong individually. The absence of an individual mistake makes decay a structural failure rather than an error you can point to and fix once. It is also why step 4 below, naming an owner, is the step that prevents this from recurring rather than just cleaning it up once.
Audit Your Exclusion List
Four steps. None of them require new budget, and the first one is the one most teams have never done.
1. Export and Read
Why check this you cannot evaluate a policy you have never seen. Lists inherited from a vendor default or a previous agency represent decisions nobody at your company made.
How it is done ask your DSP or verification partner for the current active exclusion list as a file. Read it top to bottom. Expect it to be longer than you think and to contain terms nobody can explain.
2. Separate Risk Terms
Why check this this is the safety versus suitability line applied to your list. Some terms genuinely map to floor level content. Others are ordinary English that appears in sport, entertainment and current affairs.
How it is done sort every term into keep, review or remove. Shoot, fire, shot, attack, death and dead almost always belong in review, because their presence in a URL says nothing about the content. Place names added after a specific event are the clearest removal candidates, per the Paris example.
3. Measure Blocking Costs
Why check this the industry average tells you nothing about your account. What you want is your blocked impression volume, and how much of it was blocked by keyword rules rather than by category classification.
How it is done ask for a breakdown of blocked impressions by cause. Vendors have built tooling for exactly this, including products that give publishers visibility into how buyers are blocking them so overblocking can be diagnosed. If your partner cannot produce the split, that is itself a finding.
4. Assign an Owner
Why check this decay is the failure mode that costs the most, and it happens entirely through neglect. Terms get added under pressure during an incident. Nothing ever triggers their removal.
How it is done name a person, set a quarterly review, and record the reason next to any term added reactively. A term added because of a specific event should carry the date of that event, so the next reviewer can see it is five years old.
None of this is an argument against brand safety. A floor exists for good reason, and no serious advertiser wants to fund content that sits below it.
The argument is narrower. A list that nobody has opened in two years, inherited from a vendor default and grown by reflex after incidents nobody remembers, is not a brand safety policy. It is a leftover, and it is quietly deciding where a meaningful share of your budget cannot go.
Export yours this week and read it. Most people find something in the first five minutes that nobody can justify.
FAQ
Brand safety concerns content no advertiser should appear beside, described in the industry framework as the floor. Brand suitability is graded above that floor and is a question of individual brand preference. A keyword blocklist cannot tell the two apart, because it matches strings rather than assessing content.
In a 2026 test by Integral Ad Science and Reuters, 54 percent of Reuters news URLs already cleared as brand suitable would have triggered a typical keyword blocklist. On the lifestyle section the figure was 4.27 percent of URLs, but those pages carried 13.5 percent of the section’s ad impressions during the test period.
Because terms are added reactively during incidents and almost never removed. Reuters described brands that added the word Paris after the 2019 Notre Dame fire and were still blocking it during the 2024 Olympics, since the lists are not reviewed on any schedule.
The industry disagrees on this. IAS says only 0.25 percent of impressions on its platform are blocked through the exclusive use of keyword blocking, while also acknowledging that many advertisers still run some keyword blocking alongside more sophisticated tools. Reuters says advertisers appear to use extensive blocklists in the majority of cases. Checking your account settles it.
The organisation was discontinued on 8 August 2024 following a US House Judiciary Committee investigation and antitrust allegations from X. The Brand Safety Floor and Suitability Framework it produced remains in use across verification platforms and the IAB Tech Lab taxonomy, and the Floor is incorporated by reference into Media Rating Council guidance.
Spotted an error? Contact the desk. Our corrections policy explains how we handle corrections and updates.