AdMarketplace is a 25-year-old search ad network that runs search ads for browsers and other partners. It’s also been a vocal critic of Google and was cited during the recent search antitrust trial as an example of an independent search ad competitor harmed by Google’s dominance.
Well, now one of adMarketplace’s key executives is a former top Googler.
Sam Cox joined adMarketplace as chief product officer in December after stints at Google, Amazon and Integral Ad Science. He was the group product manager in charge of Google’s programmatic exchange business between 2016 and early 2021, and then moved over to Amazon for a two-year stretch as director of technical product management for its DSP.
Between Amazon and IAS, where he spent two years as SVP of product management, Cox took a little time off to work on his Hudson Valley farm, tend to animals – he and his family raise sheep, goats and pheasants – and think about what AI is doing to search.
He landed on two main paths forward: using AI to squeeze more performance out of traditional programmatic media or focusing on how to monetize AI-powered search and new chatbot experiences.
Most of the industry is running down the first path. Cox picked the second.
At adMarketplace, he’s tasked with helping build “the next generation of search advertising,” to quote from the press release about his hiring. That may sound a little grandiose, but in practice just means upgrading adMarketplace’s search stack, including using AI to match queries with more relevant ads and designing ad formats specifically for AI search.
“I think this is a time for betting, and betting hard,” Cox said.
AdExchanger spoke with Cox about his take on ads in ChatGPT, zero-click search behavior and why he’s skeptical of anyone who claims to have cracked generative AI optimization.
AdExchanger: What’s on adMarketplace’s AI road map?
It comes down to three things: relevance, what I call “objects of decoration” and ad selection.
On relevance, we’re using AI to better interpret and normalize queries from long, messy natural language to more structured intent.
What I mean by objects of decoration is that we’re borrowing from OpenRTB. An RTB call can contain hundreds of fields, and third-party search has a lot to learn from that. We’re figuring out which fields matter, which are populated and how to layer causal signals on top of the query term without drifting into user-specific targeting, especially with privacy-centric partners like Firefox and Opera.
The third piece, ad selection, is about using AI to decide which ads, formats and copy to show in a given context.
What do you think about ads in ChatGPT?
The principles they outlined are pretty close to how we think about it. AI chat experiences need ads, but those ads should be privacy-centric and clearly marked. They also shouldn’t influence a model’s answers. The ad results have to be separate from the advice being given by a bot.
Based on what OpenAI has shared so far, that’s the direction they seem to be going in, which I think is the right instinct.
Back in June 2024, adMarketplace published a blog post with the headline “Search No More? How PMax Has Effectively Nullified Your Search Advertising Efforts.” I know that’s from well before you joined, but do you agree?
What I’ll say is that it reflects a tension we’ve seen for a long time between control and performance. When we worked on Performance+ at Amazon, the pattern was the same. Advertisers say they want granular control, but what they really want is better outcomes, and those two things often conflict.
My view is you need to draw hard lines around what can’t happen – brand safety, for example, is nonnegotiable – and be more flexible on targeting and optimization. You also need multiple strategies at once, like a “champion” approach and a “contender” that tries to beat it. If the contender wins, you shift spend and explain to the customer what changed and why.
That’s how you preserve transparency and a sense of ownership while still letting the system automate as much as possible, whether you call it PMax or something else.
I keep getting pitched by vendors that claim they’re able to help brands control how they show up in generative AI search results. How real is that?
I don’t think anyone truly knows. There’s a lot of demand for certainly so, of course, there are people willing to claim they’ve figured it out. But we’re dealing with systems and emergent behaviors that are changing very quickly.
Even if you reverse-engineer something today, there’s no guarantee it will hold true tomorrow as the models get updated. So, I’d treat any very definitive advice with skepticism. We’re in a period of genuine experimentation, and a lot of the “rules” people are selling right now are probably false certainty.
You were deposed for the Google ad tech antitrust case. What’s your reaction to how the trial played out?
I have to be careful answering that. I was directly involved on the Google side, some of my emails are in the record and I was named as a witness, so I’m still kind of not allowed to talk about the substance or the outcome.
What I can say is that it was a surreal experience. I was deposed multiple times during COVID in a tent, which was also my office on the farm. I had a wood stove going, I was surrounded by multiple computers and I was being questioned by the US government.
It’s definitely one of the stories I’ll be telling when I’m an old man.
This interview has been lightly edited and condensed.
For more articles featuring Sam Cox, click here.
