News & analysis
OpenAI Tests Image Ads, Adds Measurement Tools to ChatGPT
OpenAI plans to test image ads in ChatGPT later this month and is expanding measurement partnerships, but advertisers still face a hard inventory ceiling.
By Roomer Editorial · · 4 min read
What is OpenAI changing in ChatGPT ads?
OpenAI is planning to start testing a new ad format in ChatGPT later this month in the U.S.: image ads shown during image generation. The test has not started yet. Ads will be clearly labeled and kept separate from the images ChatGPT generates for users.
Since OpenAI started selling ads in ChatGPT in February, advertisers have worked with one format. That format includes advertiser name, logo, headline, description, landing page, and a single image. Seven months later, that was still the only option on the table. The planned image ad test is the first expansion of what advertisers can run, giving them a way to show what a product could look like rather than relying only on text paired with a static image.
Advertisers have had three consistent complaints since February, according to Digiday's reporting: measurement, limited formats, and limited inventory. OpenAI's latest round of updates addresses the first two directly. The third, inventory, is left largely untouched.
Why does measurement matter here?
Advertisers have told OpenAI they can't properly measure how ChatGPT ads perform. Digiday describes measurement as one of the key pieces OpenAI needs in order to grow ad budgets on the platform. Without a clear line between ChatGPT campaigns and outcomes, advertisers are likely to keep budgets small and contained rather than scale spend.
OpenAI is responding by expanding its in-house tools and partner integrations, including Kochava along with ten other partners, so advertisers can apply measurement approaches they already use elsewhere. That includes evaluating results beyond last-click attribution, a standard ask from any advertiser used to running campaigns across other channels.
Geo-based incrementality testing is at an earlier stage. It's running in early experimentation with partners Haus, Measured, and WorkMagic, which suggests this piece is further from a general release than the Kochava-style integrations.
OpenAI is also piloting brand suitability evaluation with DoubleVerify and Integral Ad Science, inside controlled testing environments. OpenAI says those pilots will not give either company access to real user conversations, framing the limit as a deliberate privacy safeguard rather than a technical limitation.
Taken together, Digiday frames these updates as a sign of how early OpenAI's ad business still is. The company is building out the formats, measurement tools, and safeguards that advertisers typically expect from a more established ad platform, and it's doing so seven months after the ad business opened rather than before launch.
What's the inventory problem, and can OpenAI fix it?
Inventory is the harder issue, and none of this month's updates touch it directly. ChatGPT handles roughly 2.5 billion prompts a day globally. Google handles roughly 15 billion searches a day. ChatGPT also currently has only one ad placement, while Google has multiple ad placements available to advertisers.
Zeno Group evp of paid social Shamsul Chowdhury laid out the comparison plainly: smaller prompt volume set against Google's search volume, combined with a single placement versus several, creates a gap between supply and demand. New formats and better measurement don't close that gap on their own — they make the existing, narrow inventory easier to justify spending on, not larger.
The timing is notable. These updates land at the start of Advertising Week New York, where OpenAI executives plan to use the event to gather direct feedback from marketers on how the ads product has performed so far. That suggests OpenAI knows measurement and format complaints are the ones it can move on quickly, while inventory is a structural constraint tied to how people actually use the product.
Roomer Editorial's take
None of this changes the math for a travel creator or small operator weighing whether to spend money on ChatGPT ads right now. The measurement tools OpenAI is rolling out are built for advertisers who already have tracking infrastructure elsewhere to connect them to. If you're a solo creator or a small hotel brand without that setup, the Kochava integration and incrementality testing don't do much for you directly. They're aimed at agencies and larger advertisers who can plug ChatGPT data into systems they already run.
The inventory comparison is the part worth sitting with. One ad placement spread across 2.5 billion daily prompts, against Google's far larger search volume and multiple placement options, is a narrow funnel by design, not a temporary bottleneck.
New formats and better measurement don't close that gap on their own — they make the existing, narrow inventory easier to justify spending on, not larger.
A boutique tour operator or independent hotel chasing visibility on ChatGPT is competing for a sliver of attention that search advertising doesn't ration the same way. That's worth remembering before anyone commits real budget to a still-untested format.
There's a broader point here about control. Advertising inside someone else's chat interface means you don't own the placement, you don't set the format, and you're dependent on their measurement partnerships to tell you whether any of it worked. That's a different position than owning a site where a bookable page converts a visitor on your own terms, with your own commission structure on the booking.
Roomer Editorial's view: once image ads roll out, treat any test on ChatGPT as exactly that — a test, not a channel to plan around. Set a fixed budget you're comfortable not seeing again. Track conversions on your own side as closely as the tools allow. Don't build a campaign plan around a format and an inventory pool this early and this constrained. The format is unproven. The placement pool per prompt is small. Spend accordingly, and put the bulk of your effort into the site and bookable pages you actually control.
Source: Digiday
