AI Is Rewriting The Economics Of Brand Safety And Programmatic Buying

For years, digital advertisers treated brand safety like an airport security line: if a headline contained the wrong word, the entire page got flagged. The problem is that blunt keyword blocklists do not just filter out risk. They also block huge amounts of perfectly safe, premium inventory in the process. Now, as AI tools get substantially better at understanding context, sentiment, and nuance, advertisers are starting to realize their old safety playbook may actually be hurting campaign performance by shrinking reach, inflating CPMs, and cutting brands off from valuable media environments they never needed to avoid in the first place.

Steven Widman, Vice President of Strategic Development at GameLift Media, has spent more than 15 years working inside enterprise digital media sales, helping major advertisers rethink how and where they buy programmatic inventory. His background includes leading the successful consolidation of Sanofi’s U.S. programmatic business and helping guide Anheuser-Busch through a major DSP review process that challenged long-standing platform relationships. He argues the current keyword-blocking problem persists because the industry still punishes visible placement mistakes far more aggressively than it rewards performance gains like lower CPMs or expanded reach.

"Advertisers are almost playing prevent defense. They’d rather block as much as they possibly can and not have one mistake, because that mistake could be really costly. But that heavy-handed approach has done a lot of damage to publishers," says Widman. In his view, the industry’s incentive structure still heavily favors visible caution over quieter efficiency gains, particularly inside large enterprise organizations where a single problematic placement can escalate quickly. As a result, many advertisers end up unnecessarily limiting available scale even as contextual AI tools become substantially better at distinguishing actual risk from harmless adjacent content.

The Tortured Poets problem

The downstream economic effects become especially visible at the publisher level. Widman points to one example where legacy keyword-blocking systems incorrectly classified roughly 38% of a premium publisher’s inventory as unsafe. After applying AI-driven sentiment analysis to the same inventory, the false-positive rate reportedly fell to around 3%. Those misclassifications matter because large groups of advertisers often rely on nearly identical exclusion lists built around broad categories like conflict, violence, or hate speech. "When lots of advertisers are all using the same keyword-based technology and the same broad categories, you shrink the pool of available impressions pretty considerably," Widman explains. "If everyone avoids 30% or 40% of the inventory and competes over the smaller pool that's left, prices for those remaining impressions are naturally going to rise."

Keyword blocklists also create a strange cultural blind spot for advertisers trying to stay connected to major online conversations. Taylor Swift's The Tortured Poets Department album became a perfect illustration of that issue, with advertisers inadvertently blocking themselves from one of the internet’s biggest audience magnets over a single keyword. "When something like that comes out, there is a massive draw towards it online, and so much of that was being blocked because of the word 'tortured,'" notes Widman. "A word like 'torture' is obviously something you don't want to be near in a literal sense, but it can be so easily misconstrued. These are environments where so many advertisers want to be, but the keyword blocking is just so blunt that it's keeping them out."

Suitability gets smarter

The financial pressure created by overly broad keyword blocking has also pushed the industry to rethink how advertisers evaluate news and cultural content altogether. Premium publishers lose substantial revenue when large categories of inventory are automatically excluded, even when the surrounding content poses little actual reputational risk. In response, more advertisers are shifting toward suitability frameworks that evaluate context, tone, and nuance instead of relying on rigid "safe" versus "unsafe" classifications. Widman expects brands to work more closely with verification partners like Integral Ad Science, DoubleVerify, and newer entrants like Mobian to define more customized thresholds around where and how they want campaigns to appear.

Widman says those evolving partnerships also open the door to a more proactive use of contextual AI signals across media buying. Once systems become sophisticated enough to interpret sentiment and environment accurately, the same signals can help advertisers optimize creative and messaging in addition to managing suitability concerns. That shift turns contextual analysis into a performance tool as much as a protection layer, particularly as platforms, verification vendors, and advertisers become more tightly connected operationally. "Beyond just doing a better job at keeping things brand suitable, there's also the offensive side, which is: let's use all these new signals not just to know what to block and what not to block, but also to take advantage of the ability to tweak our messaging to better fit that context and sentiment."

The hidden cost of blocking

Widman believes the next phase of contextual AI extends beyond helping advertisers avoid risky environments. As sentiment analysis improves, the same signals can help brands tailor messaging and creative more intelligently to fit the surrounding context. Still, he says adoption barriers remain emotional as much as technical. Having previously worked on early DOOH and programmatic TV solutions before broad industry adoption, Widman says marketers rarely overcome brand safety fears through performance spreadsheets alone. "It's drawing out those examples and showing advertisers an example of 'war' where they are not going to want to be, versus an example of 'war' used in the sense of two kids playing a game," he explains. "Drawing out those nuances and showing how the technology is able to parse them through real examples is how clients and advertisers can start to say, 'Okay, I get this now, I feel comfortable with it.'"

Widman argues many advertisers still evaluate brand safety through scorecards that reward avoidance without measuring what gets lost in the process. If dashboards only track blocked impressions and avoided incidents, teams never fully see the reach, inventory access, or pricing efficiency sacrificed through overblocking. He says that blind spot exists across multiple channels where outdated measurement habits can obscure more meaningful performance gains. "The fear of messing something up, especially in this area of brand safety and suitability, is pretty high," Widman concludes. "It's going to be important for these technologies to spell everything out very clearly and really make advertisers feel comfortable."