ChatGPT will show ads while it draws your picture. OpenAI's three proof points all come from OpenAI.
A new visual ad format starts testing in the US later this month. The numbers meant to sell it to advertisers are worth a closer look.
OpenAI announced on October 5 that it will test a new kind of ad inside ChatGPT: pictures. The first place they’ll appear is the screen where you’re waiting for ChatGPT to finish generating an image.
That’s the news. The more interesting part is the page of results OpenAI attached to it.
What’s actually changing
According to OpenAI’s post, the new format uses images to show “product inspiration, product usage, or the experiences they make possible.” Mashable described the mockups: a banner with a headline and a “Learn more” button, or a row of scrollable photos, both marked “Ad” and sitting under the progress indicator of an image still being made. The mockup brand is a fictional grocery chain.
OpenAI says the ads will be clearly labeled and “remain separate from the image being created.” It repeats the line it has used for every ad format so far, that advertising “does not influence the answers ChatGPT provides.” Testing begins later this month, in the US only, with an initial group of advertisers. OpenAI didn’t say how many, didn’t name them and gave no prices.
Image generation is a new spot for ads. PPC Land, an ad-industry outlet, went back to OpenAI’s Ads Manager help text from April and found it said ads didn’t appear after a user generated an image. The October post doesn’t say whether that rule has been rewritten or just overtaken.
The three numbers
To make the case to advertisers, OpenAI quoted three early results from its measurement partners:
| Advertiser | Reported by | Result |
|---|---|---|
| WeightWatchers | DV Rockerbox | Attributed cost per acquisition 15.3% lower than its blended paid-search benchmark |
| Dose | WorkMagic | Statistically significant lift, with 67% of incremental purchases from net-new customers |
| Portland Leather | Triple Whale | 93% of visitors from ChatGPT Ads were new |
I’d keep three things in mind. Each one is a single advertiser. OpenAI picked them and published them in its own announcement. And they measure different things, so you can’t line them up: one is a cost comparison, one is a lift test, one is a share of visitors.
The 93% figure is the weakest of the three. “Visitors were new” says nothing about whether they’d have bought anyway. OpenAI itself calls its incrementality work “still in its early stages”, which is the part of the post I’d quote back at anyone treating these as proof.
None of this means the ads don’t work. It means the evidence so far is a sales deck.
Why now
ChatGPT reaches 1.2 billion people a week, according to OpenAI, up from the 900 million it disclosed in February, PPC Land noted. TechCrunch’s Sarah Perez points out that ads started this year to support the free and low-cost tiers, and that a more annoying free tier could nudge casual users toward a paid plan. OpenAI’s own framing is that advertising lets it serve “a much larger market than subscriptions alone.”
The same post also lists a long run of measurement vendors, among them AppsFlyer, Adjust, Kochava, Hightouch, Tealium and LiveRamp, plus brand-safety pilots with DoubleVerify and Integral Ad Science. OpenAI says those outside firms will check how its safeguards work without seeing private conversations. That’s a promise I can’t verify from here.
My read
I don’t think the labeled card is the problem. A clearly marked ad that sits apart from the image is about as honest as this gets. What I’d watch is the commercial pull: once ad revenue depends on how many people are generating images, the product team has a reason to want more of them doing it, and the company has told us the answers won’t be affected but hasn’t said much about everything around the answers.
The first real data point comes when the test starts and someone outside OpenAI sees what a free-tier user actually gets. For how other companies are fighting over who controls the checkout button for AI agents, see Amazon locked AI shopping agents out. Eight days later, Shopify handed them the checkout button.