OpenAI today presented a picture-led ad unit for ChatGPT that will first run while people create images in the app, widened its roster of measurement vendors to 18 named companies, and said DoubleVerify and Integral Ad Science are developing brand suitability evaluation pilots. Trials of the unit begin later this month in the United States with an initial group of advertisers, according to OpenAI.

In Short

OpenAI is adding a new kind of ChatGPT ad built around pictures, and it will first appear while people make images in the app, starting with a small group of US advertisers later this month. It matters to brands and agencies that buy ChatGPT ads because the same announcement adds more tools for tracking results and two outside companies, DoubleVerify and Integral Ad Science, that will check how well brand protection works. In practice, more third-party tools can now report on ChatGPT ad results, while the picture ads and the checks are still early-stage with few public details.

What OpenAI described

According to OpenAI, the new unit uses images to show product inspiration, product usage or the experiences a product makes possible, so that people can "imagine how products and services could fit into their lives." The mock-ups in the post show a grocery brand with a dinner-themed headline, a photograph and a Learn more button.

The first trials will sit inside the app's image-creation flow. Ads will be "clearly labeled, and remain separate from the image being created," according to OpenAI, and as with its other formats, advertising does not influence the answers ChatGPT provides. Testing begins later this month in the US. The post gives no price, no list of participating advertisers, no market beyond the United States and no word on whether the unit will appear in the self-serve Ads Manager.

Audience size is the stated rationale. ChatGPT reaches 1.2 billion people each week, according to OpenAI, which also argues that advertising lets it serve "a much larger market than subscriptions alone." The figure compares with 900 million weekly users disclosed on February 27 and a number above 1 billion that David Dugan cited on August 31. Against February, the new number is about a third higher, reached in a little over six months.

How the format fits the product so far

Image creation was not always an ad context. When PPC Land reviewed OpenAI's help documentation for the Ads Manager in April, the text stated that ads did not appear after a user generated an image, and that Free-plan users who chose an ad-free configuration lost access to tools including image creation. The post does not say whether that rule has changed, or how the ad-free configuration treats the new unit.

Unit shapes have shifted several times. The original card carried an advertiser name, a favicon, a headline, a short description and an image, and sat below a response. Digiday reported a larger image unit with an optional call-to-action button on May 21. A multi-product carousel for product feed campaigns was reported live in August, and on September 16 OpenAI began testing Sponsored Agents, which open a separate conversation with an advertiser's agent after a click.

The commercial base has grown quickly. The pilot began on February 9 at a $60 CPM, passed $100 million in annualised revenue within six weeks with more than 600 advertisers, and reached a $1 billion run rate on August 31, roughly a tenfold rise in just over five months. Prices went the other way. CPMs were reported as low as $25 by April, a fall that PPC Land linked in part to the limited means buyers had of showing what the impressions produced. The ad platform's list now covers 63 countries.

Measurement: a longer roster and partner-reported results

Data connections

According to OpenAI, integrations with Hightouch, Tealium and LiveRamp make it easier for advertisers to send conversion data from their existing systems to ChatGPT Ads. The receiving end has existed since May 5, when OpenAI opened the self-serve Ads Manager with a pixel and a Conversions API. LiveRamp connected its Conversions API Hub on June 10, and Hightouch, a composable customer data platform provider, surfaced with Triple Whale in an August advertiser email. In late September LiveRamp said it had become a technology partner for customer-list audiences. The post describes no event schema, match rate or latency for any of the three connections.

Attribution partners

The attribution list has ten names: AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge and Tenjin. Seven of them (Adjust, Airbridge, AppsFlyer, Branch, Kochava, Singular and Tenjin) are the mobile measurement partners that carry Meta's certification badge. The other three are Triple Whale, DV Rockerbox and Northbeam. According to OpenAI, the partners enable Conversions API, reporting and click attribution.

Two of the names are not new to the platform. AppsFlyer and Adjust went live for app installs and in-app events across seven markets in July, and a one-day view-through window arrived in August, excluded from bidding and billing. Whether the wider roster inherits those market limits or that window is not stated.

Incrementality

Incrementality measurement "plays an important role in building a fuller understanding of advertising's business impact," according to OpenAI, though the company describes its own work on the subject as "still in its early stages." It is partnering with Haus, Measured and WorkMagic on geo-based experiments. Such an experiment divides a market into regions, changes advertising in some of them, leaves others untouched as a control, and compares outcomes between the groups.

Measured also appears among the full-funnel and advanced measurement partners, alongside Fospha and INCRMNTAL. Counting each name once across the post's four lists - data connections, attribution, full-funnel and incrementality - gives 18 vendors.

Early results

Three findings are offered as early evidence. According to DV Rockerbox, WeightWatchers' attributed cost per acquisition on ChatGPT Ads was 15.3% lower than its blended paid-search benchmark. WorkMagic reported statistically significant lift for the wellness brand Dose, with 67% of incremental purchases coming from net-new customers. And according to Triple Whale, 93% of Portland Leather's visitors from ChatGPT Ads were new.

The three numbers measure different things by different methods. The first compares an attributed cost against a benchmark the post does not define, and attributed figures depend on the credit rules behind them. The second describes the make-up of a lift, but no size for the lift is given. The third is a share of visitors, and a high share of new visitors says nothing about whether those visitors would have arrived anyway. None comes with spend, dates, sample sizes or confidence intervals. All reach the reader through a chain of partner, then OpenAI, then the post, which makes them vendor-supplied figures rather than independently verified ones.

One further detail bears on that chain. DoubleVerify bought Rockerbox in 2025 to add multi-touch attribution, and DoubleVerify is also one of the two companies named for the brand suitability pilots. One corporate group therefore appears in two roles in the same post. The post does not treat this as an issue, and nothing in it suggests the roles are connected.

Brand suitability: guardrails, Negative Phrases and two pilots

The third strand concerns brand suitability, the advertiser-specific tolerance for sensitive but lawful content near its ads. According to OpenAI, placement guardrails assess whether a conversation is appropriate for advertising, "helping keep ads out of emotionally vulnerable, sensitive, or otherwise unsuitable contexts." Automated review, human oversight and ongoing monitoring identify policy violations, and a new control called Negative Phrases is available to qualifying advertisers for narrow requirements tied to a brand's own policies.

The post gives no qualification criteria, phrase limits or matching rules. PPC Land has documented that positive targeting runs on advertiser-supplied context hints at the ad group level rather than on keywords, and how Negative Phrases relate to that mechanism is not explained. Earlier documentation also listed dating, health, financial services and politics advertisers as excluded from the US test, and the post does not say which category rules apply to the new unit.

The DoubleVerify and IAS pilots

OpenAI is developing brand suitability evaluation pilots with DoubleVerify and Integral Ad Science. According to OpenAI, the approach lets independent partners assess how its brand safety standards are applied "in a controlled testing environment, without accessing private user conversations." That sets it apart from the measurement both vendors already run over live social inventory, such as DoubleVerify's post-bid coverage of Threads from October 2025 and IAS measurement on X.

Both chief executives are quoted. According to Lidiane Jones of IAS, "marketers need confidence that brand safety protections are durable, consistent, and built to operate at scale." Mark Zagorski of DoubleVerify describes ChatGPT as an environment where "context is defined by the conversation itself rather than a page or video." No pilot start date, scope or reporting format is given, and the post does not say whether results will be published. It also does not mention Media Rating Council accreditation, a label that carries weight because the MRC limited use of the term brand safety in October 2025 to vendors that analyse content rather than only properties.

The ownership of both named verifiers has also changed, or is due to. Novacap took IAS private for $1.9 billion in a deal that closed in December 2025, and Nielsen agreed on August 6 to acquire DoubleVerify for about $2.15 billion. The post does not discuss ownership.

The pressure behind the pilots is visible in two data points. Sensor Tower found that ads shown per user per hour on the US mobile app rose 163% between April and August. And a DoubleVerify study of 22,000 consumers and 1,970 marketers found that 65% of advertisers express brand suitability concerns in walled gardens, a vendor-run survey, and DoubleVerify sells brand suitability measurement.

Why this matters for the marketing community

The announcement lines up three things buyers on other channels already expect: a creative unit suited to visual categories, a measurement stack with attribution and experiments, and third-party verification. Five months separate it from the May 5 arrival of the pixel and Conversions API, which PPC Land flagged at the time as the first way to close the loop between a click and a downstream action. Each strand is also still early. The unit is not yet in market, incrementality work is described as early, and the pilots are "developing."

The placement raises its own question. A text answer carries a topic; an image request carries a creative brief. What does a suitability tier mean when the "page" is a one-line request for a picture? The post says guardrails assess whether a conversation is appropriate, and it does not say how that assessment applies to image prompts, or whether the generated image is assessed at all.

Buying routes are multiplying in parallel. Amazon DSP buyers gained a US pilot for ChatGPT placements in September, and HubSpot and Shopify became the first CRM and ecommerce routes on September 16. Whether the visual unit will be available through those routes, through the Ads Solutions team or only through the self-serve interface remains unstated, as do pricing, conversion windows and the sample behind each early result.

OpenAI's post closes by saying it will keep evolving the platform with new formats and capabilities and that businesses can sign up at ads.openai.com. The next verifiable information will come from the trials themselves: who enters them, at what price, and whether any of the three measurement claims is repeated with disclosed methods.

Timeline

Summary

Who: OpenAI, with measurement vendors including Hightouch, Tealium, LiveRamp, AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge, Tenjin, Fospha, Measured, INCRMNTAL, Haus and WorkMagic, and verification companies DoubleVerify and Integral Ad Science. The announcement affects advertisers, agencies and measurement teams buying ChatGPT ads.

What: A new visual ad format to be trialled during image creation in ChatGPT, an enlarged measurement partner roster, three partner-reported early results (15.3% lower attributed CPA for WeightWatchers, 67% net-new share of incremental purchases for Dose, 93% new visitors for Portland Leather), Negative Phrases for qualifying advertisers, and brand suitability evaluation pilots with DoubleVerify and IAS.

When: The announcement was made today. Trials begin later this month. No dates are given for the pilots.

Where: The visual format trial is in the United States. The post gives no other market for the format. OpenAI's availability list for the ad platform covers 63 countries.

Why: According to OpenAI, ChatGPT reaches 1.2 billion people each week and advertising lets it serve a larger market than subscriptions alone. The expanded measurement and brand suitability work is meant, in OpenAI's account, to help businesses understand campaign impact and build confidence in the contexts where their ads appear.