OpenAI Adds Ad Measurement and Plans Visual Placements
OpenAI expanded ChatGPT advertising measurement on 5 October and announced a US test of visual advertisements during image generation later in the month. The package combines new placement inventory with tools intended to show advertisers whether spending produces useful outcomes. OpenAI cited WeightWatchers’ attributed acquisition cost at 15.3% below a blended paid-search benchmark, alongside results from other advertisers using different measurement methods. These figures are specific campaign claims, not a platform-wide return. For OpenAI, the commercial step is to give marketers a basis for allocating recurring budgets to ChatGPT while expanding the circumstances in which an advertisement can appear within the service’s free and lower-priced tiers.
DoubleVerify’s Rockerbox measurement includes ChatGPT as a touchpoint in a wider customer journey. Specialist publication PPC Land highlighted missing detail around the WeightWatchers result, including spending, campaign dates and conversion counts, and noted that DoubleVerify called the comparator an established benchmark while OpenAI described blended paid search. That wording difference does not prove the numbers conflict, but it limits replication. WeightWatchers growth executive Jake Dmochowski described “an encouraging early signal” as the company evaluated scaling. The evidence therefore supports an advertiser considering greater investment; it does not establish that every additional dollar will retain the same acquisition cost as the initial measured campaign.
OpenAI’s separate measurement account distinguishes attribution from experiments intended to estimate additional sales. It reported that Dose’s WorkMagic analysis found 2.3 times as many incremental orders as last-click attribution counted, while the broader announcement said 67% of incremental purchases were net new. Portland Leather’s measurement, by contrast, concerned new visitors. These indicators answer different commercial questions and cannot be combined into a common conversion rate. A visitor is not necessarily a purchaser, and an attributed purchase is not automatically one caused by advertising. The useful advance is the availability of several ways to examine performance, provided advertisers preserve those distinctions when comparing channels and deciding whether to increase spending.
The proposed visual placement test is narrower than a general change to all ChatGPT answers. OpenAI said it would begin with selected US advertisers during image generation for Free and Go users, with advertisements labelled and separate from the generated image. It also described brand-suitability pilots with DoubleVerify and Integral Ad Science using controlled testing rather than access to private conversations. Those pilots address a different purchasing obstacle from conversion measurement. A campaign can be profitable on average yet unacceptable to a brand if it appears beside unsuitable material, making the placement rules and the ability to examine them commercially relevant in their own right.
There is a tension between the detail advertisers want and the privacy of a conversational service. Traditional placement review often examines a page or video; a conversation can change context as it develops. Controlled evaluation can help test policies without exposing personal exchanges, but it should not be mistaken for independent observation of every live placement. OpenAI has announced a mechanism for assessment, not a universal assurance that every advertiser’s preferences will be met. Similarly, flexible attribution windows improve reporting choice but do not make unlike campaigns automatically comparable. Buyers still need consistent definitions of outcomes, sufficient sample sizes and a clear distinction between measurement settings and the system’s optimisation behaviour.
Analysis
Better measurement can unlock advertising budgets more effectively than additional inventory alone, because marketers need defensible evidence before shifting recurring spend. A 15.3% lower acquisition cost would imply about 18.1% more attributed acquisitions for the same budget if costs remained constant; that arithmetic is conditional, not a forecast of performance at scale. Incrementality tests are more valuable for deciding whether revenue would disappear without the campaign. OpenAI captures value if these tools turn exploratory spending into durable demand, but must preserve user trust and avoid allowing favourable attribution rules to masquerade as additional economic output.