The Problem of Measuring Ads Shown to AI Agents Takes Center Stage

Publishers and advertisers looking for ways to place ads in front of a new kind of audience — not humans, but AI agents — are running into a familiar problem: how do you know whether an ad actually worked? Answering that question is now fueling a fierce race among startups building ways to serve ads to AI agents and measure the results that follow.

It's still early days, which means most startups claiming to have already "cracked the code" probably haven't fully solved it yet. The measurement process remains murky, and the industry has no single widely accepted standard — many projects are still stuck at the proof-of-concept stage.

OpenAds, for instance, is taking a relatively traditional route — adapting unique referral codes for AI. Say a user asks ChatGPT for the best sneakers for a half-marathon. As ChatGPT scans the web for an answer, OpenAds inserts an ad for a brand's sneakers along with a 10% discount code. If ChatGPT judges the code relevant enough to the user's query, it "survives the culling process," says Steven Liss, co-founder of OpenAds, an AI-native advertising platform. When the user clicks the link that shows up in the AI's response, brands can track that the click came specifically from ChatGPT.

Another startup, Oasy, is an advertising and analytics platform for web publishers built to monetize AI bot traffic; it's currently working on partnerships with major existing native-advertising platforms.

The bigger significance of this race is that as AI agents increasingly act online in place of humans, the advertising industry is being forced to scramble and build an entirely new kind of measurement infrastructure for this audience — and for now, the process involves far more competing guesses than a clear roadmap.