ChatGPT Is Coming for Your Ad Budget

A while back, a friend of mine who does e-commerce media buying was venting to me.

He told me he's pouring hundreds of thousands every month into Google and Meta, and the traffic he buys keeps getting more expensive while conversions keep getting worse. Then he asked me: "Don't you feel like user attention is migrating?"

I said, "Migrating to where?"

"To ChatGPT," he said.

That comment got me thinking for a long time. And honestly, he's right. Think about it — how many people, when they run into a problem today, don't even think to open a search engine? They just ask an AI. Once user attention shifts, ad dollars follow.

The question is: can you actually run ads on ChatGPT?

OpenAI recently made two moves that tell me they're not just "testing the waters." They're seriously building something.

First Move: Products Can Now Line Up on Their Own

About three months ago, OpenAI launched a feature called the "product feed."

What's a product feed?

Simply put: you feed your product catalog to OpenAI, and it generates ads for you automatically.

At first, those ads could only feature a single product. Whatever you were promoting, that's what it showed. But recently they upgraded it — now a single ad can showcase several products at once, displayed as a carousel at the bottom of the conversation.

Think about what that means.

Before, running an ad on ChatGPT was like putting up a sign by the side of the road — people walking by would glance at it and keep moving. Now? It's a mini storefront window, displaying several items at once.

But here's where it gets interesting: advertisers can't choose whether to show a single product or a carousel.

OpenAI makes that call for you.

If it decides a single product is right for the moment, it shows one. If it thinks a carousel fits better, it shows a row. And each carousel currently features products from only one retailer — no mixing.

It's a clever move, but also a heavy-handed one. Clever because OpenAI has the data, and it knows better than you which format suits which context. Heavy-handed because, as the one footing the bill, you've lost a chunk of control.

Second Move: You Can Finally See Returns on Ad Spend

What's the biggest pain in running ads?

It's that you have no idea what your money is actually delivering.

Before, if you ran a product ad on ChatGPT and someone saw it, clicked it, even bought something — you almost certainly had no idea. OpenAI didn't give you an attribution tool (a way to link "saw the ad" to "made a purchase"). You had no way to draw a line between the two.

Without that line, would you scale up your budget? No way.

So OpenAI brought in AppsFlyer.

What does AppsFlyer do? They specialize in mobile attribution tracking — figuring out which user downloaded your app because they saw which ad, what they bought inside the app, which services they subscribed to. They help you connect all those dots.

Once that line is drawn, everything changes.

People running app promotions can finally treat ChatGPT as a real advertising channel. It can sit right alongside Google, Meta, and other paid channels in the same report — and the cost-efficiency comparison is immediately visible.

Word is that about 40 brands are currently testing this attribution setup, with Grubhub among them.

That said, this setup isn't a silver bullet.

It currently can't measure incrementality (the causal lift that an ad actually drives) — in other words, it tells you how many people converted after seeing the ad, but it can't tell you whether those people would have found you on their own anyway. That distinction is exactly what determines whether an ad is worth the money.

It also can't tell you how to optimize. Which creative performs better, which time of day sees more action, which types of users are more responsive — all of that remains a black box.

But make no mistake: going from "flying completely blind" to "seeing the basic numbers" is a real, tangible step forward.

What Is OpenAI Building?

Put these two moves together and it becomes clear what OpenAI is doing.

They're building the infrastructure for performance advertising.

Three months ago, they launched the product feed to let retailers' data flow in. Now they've upgraded to carousel displays, making ad formats richer. Then they brought in AppsFlyer for attribution, so advertisers can actually see results.

Data comes in, ads go out, results come back — the loop is closed.

It reminds me of something.

When Google built AdWords, their core advantage wasn't "how accurate the search results were." It was that they were the first to connect "what the user searched for" with "what the user ultimately bought." That single connection built a business worth hundreds of billions of dollars a year.

What OpenAI is doing now follows the exact same logic.

Word is they're rushing to give advertisers more guidance on product feed campaigns before Q4. Why the rush? Because Q4 is the holiday shopping season — the most intense advertising period of the entire year.

Whether ChatGPT can prove itself during the holiday season may directly determine the scale of its ads business going forward.

But Two Problems Remain Unsolved

Problem one: How much control do advertisers actually get?

As I mentioned, OpenAI decides whether to show a single product or a carousel. Advertisers don't know how that decision gets made. They don't know how wide the performance gap is between formats. And they don't know whether they'll get more control down the line.

For a small brand, that's fine. But for a major brand spending hundreds of thousands of dollars a month, handing all display decisions over to the platform is genuinely risky. They need transparency.

Problem two: How big can this scale?

With carousels and attribution in place, running ads on ChatGPT is definitely easier than before, and the cost of experimenting is lower. But "easy to test" and "worth committing to long-term" are two very different things.

What advertisers really want to know is: can they get enough conversions at a competitive cost on the ChatGPT channel? If the volume is too small and the cost is too high, then no matter how impressive the attribution numbers look, the budgets won't stick around.

Neither of these questions has an answer yet.

OpenAI has essentially built the performance-advertising machine. Now it has to prove that this machine can actually run — and deliver.

And that is precisely the hardest part.