The agency Accuracast saw around a hundred clicks in a ChatGPT report, its own measurement captured twenty, and OpenAI support could not explain why. Digiday described the case on 1 October 2026 in a piece on why brands are keeping their ChatGPT ad budgets at test level. Four days later, OpenAI connected two dozen firms that measure ads to ChatGPT Ads and published results that it picked itself.

The same agency ran a lead-generation campaign in the UK in ChatGPT only. Completed forms kept coming in, but after two weeks the platform still showed zero conversions. Three other firms that Digiday spoke to did not see such a mismatch in clicks. Another agency, Web Guide Partner, sees conversions from ChatGPT with a delay of 24 to 36 hours, against a few hours on Google.

Advertisers have been waiting for measurement since spring. As early as April 2026, Ben Kahan of the agency Brainlabs spoke about clients who have money for ChatGPT but are waiting until they get the measurement they need. Half a year later, one executive described clients who give Google ten million dollars a month and ChatGPT less than $100,000, even though they would like to spend more. The channel stayed in the experimental bucket, even though at the end of August OpenAI announced that ChatGPT ads had reached an annualised revenue of one billion dollars in under 200 days since launch.

OpenAI has now connected firms in which advertisers already count ad results, from AppsFlyer to Kantar. In WorkMagic, for example, ChatGPT thus appears to the advertiser alongside the rest of the media mix. So far, OpenAI is only trying to measure incrementality through geographic experiments, that is, how many sales would not have happened without the ad.

Robert Webster of the consultancy TAU pointed out back in April that “Until someone independent can verify what a ChatGPT impression is actually worth, advertisers are taking OpenAI's word for it.” According to PPC Land, outside measurement was added to revenue only gradually, from the pixel in May to the partner list in October.

Brands’ results were measured by DV Rockerbox, Fospha and Triple Whale

In its announcement of 5 October 2026, OpenAI presented five results that, it says, partners verified. For the brand WeightWatchers, DV Rockerbox measured an attributed cost per conversion in ChatGPT that was 15.3% lower than its average from paid search. According to Fospha, ChatGPT advertising brought the same advertisers 83% higher revenue per thousand impressions than their average from paid social. Both comparisons target the budgets that search engines and social networks get today.

The results were verified by firms that offer ChatGPT Ads measurement to their own clients. DV Rockerbox belongs to DoubleVerify, which announced its acquisition last February. The published numbers came without spend, dates or sample size. An independent audit is missing.

Triple Whale found that 93% of visitors from ChatGPT ads to the brand Portland Leather Goods were new. But a new visitor does not yet mean an additional sale. AdExchanger warned as early as April 2025 that even incrementality reported by the platform itself cannot be taken literally: “Incrementality reporting by a platform may be useful and directionally correct, but it is not true.”

ChatGPT Ads’ advanced optimisation now also learns from conversions without a click

In August 2026, conversions after a mere ad view appeared in ChatGPT reports. At the time, however, they did not feed into bids, billing or optimisation. According to OpenAI’s announcement, advanced conversion optimisation now learns from them, but OpenAI did not state the date or availability. OpenAI added its own analysis of global campaigns, in which 52.7% of eligible conversions in a one-day window came within an hour of the ad being shown.

“A short lag describes when conversions arrive, not whether the ad caused them, because credit and cause are different things,” PPC Land objected. Webster called view-through conversions the harder part back in spring, saying that click-based measurement is easy. According to OpenAI, advanced optimisation is thus meant to learn also from purchases whose cause is harder to prove than with clicks.

In Czechia, ChatGPT ads run without personalisation

Since the launch in Europe on 24 August 2026, ChatGPT ads have been running in 31 European countries, including Czechia. Since the end of August, Czech companies can buy them themselves in Ads Manager. In the European Economic Area and Switzerland, however, they are not personalised. Of the data about the user, only the content of the current conversation, approximate location and language are used to choose the ad.

For none of the published results does OpenAI state in which country the campaign ran, and it bases its analysis of view-through conversions on global campaigns. A comparison for advertising without personalisation, the kind a Czech brand buys, is missing from the announcement. The advertiser also receives from OpenAI only aggregate data, without details of individual users and their conversations.

Shamsul Chowdhury of Zeno Group pointed out one more brake beyond measurement: ChatGPT inherently has a smaller query volume than Google search.

Monica Shukla of the agency Mile Marker advises setting up measurement before a test and not taking the platform’s metrics at face value. A buyer who splits money between media, performance channels and creators thus needs to know, for every number from the new channel, who measured it and against what benchmark. A Czech buyer also has to find out whether the campaign ran with personalisation. ChatGPT ads here do not have it.

The WeightWatchers number was measured by a company from the DoubleVerify family. DoubleVerify’s parent company, together with IAS, is at the same time preparing pilot tests meant to assess brand suitability in ChatGPT, without access to users’ conversations. In August 2026 it agreed to be bought by Nielsen for roughly $2.15 billion. PPC Land asked of that transaction who watches the watchman.

By Tereza Tiller