OpenAI Is Testing Visual Ads in ChatGPT. The First Test Needs More Than Clicks

Eric Seufert, a quantitative marketer and Mobile Dev Memo editor, says the new placement likely appears where there’s “little to no commercial intent.” He made the point in a LinkedIn post about the announcement. OpenAI will test visual ads during image generation later this month. For a small marketing team, the useful move is to decide what result would earn this format a test budget.

OpenAI announced the format on October 5, along with new conversion-data integrations, attribution partners, and early work on brand-suitability checks. TechCrunch reported that the first test is limited to the United States and an initial group of advertisers. It isn’t broadly available yet.

The placement changes the creative question. A text ad asks a marketer to explain a product in a short line. A visual ad beside an AI-generated image can show a product in use or in a setting that helps someone picture it. OpenAI says the format is meant to help users explore products and services while they generate images. The company says its ads remain separate from the image being created.

That separation matters. The ad is a sponsored placement beside the creative task. OpenAI says the ads will be “clearly labeled” and that “advertising does not influence the answers ChatGPT provides.” Teams evaluating the test will still need to understand how the ad appears, which image requests trigger it, and what happens when a user follows through to a product page. OpenAI hasn’t described the full targeting, placement, or pricing rules for this format.

A visual slot needs a measurement plan

OpenAI is building the measurement plumbing alongside the ad. Its announcement adds connections through Hightouch, Tealium, and LiveRamp, plus more partners for web and app attribution. The company is also testing geo-based experiments with Haus, Measured, and WorkMagic to estimate incremental impact. OpenAI says that work is in its early stages.

Those additions address a practical question for marketers: can the campaign signal be connected to an outcome they already track? Digiday reported that advertisers have asked for stronger measurement and more formats. Its coverage also notes that limited inventory remains a constraint. The ad business is still being built out, even as OpenAI adds ways to buy and evaluate it.

The format adds a new kind of creative, though it doesn’t create unlimited room to show it. A marketer should judge the visual unit separately from ordinary ChatGPT text ads and from image ads on social platforms. The user is asking ChatGPT to make an image, and the sponsored product appears alongside that task. If that moment carries little commercial intent, creative fit alone may not be enough to produce a qualified action. That context deserves its own test rather than being lumped into an existing channel report.

OpenAI is also working on brand-suitability pilots with DoubleVerify and Integral Ad Science. Those partners will evaluate safeguards in a controlled test environment without accessing private user conversations. In OpenAI’s announcement, Integral Ad Science CEO Lidiane Jones said marketers need brand-safety protections that are “durable, consistent, and built to operate at scale.” The pilot may help answer how OpenAI applies its rules, though it remains an evaluation project rather than a mature independent rating system.

Keep the first test narrow

This is a new surface for visual discovery, and its first test is too limited to support a broad budget shift. OpenAI hasn’t released results for the format, and its incrementality work is still early. Seufert’s read on the placement is an operator’s assessment, not campaign-performance data. That leaves scale and availability open questions for buyers.

For a small B2B team, the near-term task is simple: write down the job an image-generation ad would do, then name the one conversion that would justify repeating it. A product with an obvious visual use case may be easier to demonstrate in this format than a service whose value depends on a long explanation. That is a hypothesis to test, not evidence that one category will outperform another.

The team should also decide how it would compare the placement with its current paid channels. Set a baseline for qualified visits or a completed action, agree on how the ad click and later conversion will be joined, and keep brand discovery distinct from demand capture. If those records cannot be connected, a high click count will say little about whether the ad created a customer.

That is the useful read on OpenAI’s announcement. The image ad is the visible product change. The test will matter to marketers when they can see who it reached, what the visual context did, and whether the resulting business outcome was incremental.

Related coverage: OpenAI is testing ads that send the click into a business-agent conversation and ChatGPT Ads moved into HubSpot and Shopify.