OpenAI Is Building an Ad Business That Routes Around Programmatic

OpenAI is building an advertising business, and the notable part isn’t that it’s selling ads. Everyone with attention eventually sells ads. The notable part is that it’s being built to bypass the legacy programmatic networks entirely.

Twenty years of ad tech infrastructure exists to solve one problem: matching a buyer to an audience across millions of sites that the buyer has never heard of. Exchanges, supply-side platforms, demand-side platforms, data management, identity resolution, attribution. Every layer takes a cut, and every layer exists because the audience is scattered.

An answer engine doesn’t have a scattered audience. It has a session, an intent stated in plain language, and a conversation history. There’s nothing to resolve. The user just told the system what they want, in words, unprompted.

Why the middle layer becomes optional

Programmatic infers intent. Someone reads three articles about hiking boots, gets a segment applied, and the segment is sold to a footwear brand at auction. The whole apparatus is a guessing machine with a settlement system attached.

When a user types out what they’re trying to do and then asks follow-up questions, the guessing is finished before the auction starts. That collapses most of the stack into two parties, the platform and the advertiser, and there’s no obvious argument for keeping the rest.

It also means the inventory isn’t a page. It’s a moment in a conversation, which nobody has a standard for pricing yet. Cost per thousand impressions was built for a banner sitting on a page whether or not anyone looked at it. That unit doesn’t survive contact with an interface where a recommendation appears inside the answer the user asked for.

The transfer nobody is pricing

Follow the money backwards. Publishers lost search traffic to AI answers, which cost them the ad revenue that the traffic carried. Now the AI layer starts selling advertising against the same intent that used to arrive on publisher pages.

That’s the same demand, the same advertisers, the same buying budgets, landing one step earlier in the chain. The revenue didn’t evaporate. It moved upstream and skipped the people who produced the material the answers were built from.

Whether publishers get compensated for that depends entirely on licensing negotiations happening right now, mostly in private, mostly between large AI companies and large publishers. The trade press will find out the terms afterward, if at all.

What it does to ad tech

The independent ad tech sector has spent two years being squeezed from both ends. Retail media took the bottom-funnel budgets. The walled gardens took the reach. If conversational surfaces take the intent-heavy queries as well, what remains for the open programmatic market is the residual: display inventory on sites with falling traffic, sold against audiences that are harder to identify every year.

Consolidation follows automatically. There are too many companies doing overlapping middleware work for a market that just got smaller, and the exits are already visible in the deal flow.

What to watch

The formats will tell you most. If the ads sit clearly outside the answer, this is a conventional media business with a new surface. If they sit inside it, blended into a recommendation the user asked for and trusts, then the product being sold isn’t attention. It’s the answer itself, and disclosure becomes the entire regulatory story of the next two years.

Watch what an advertiser can buy, and how obviously the user can tell.