ChatGPT ads: the measurement gap capping budgets

0:05Host:Welcome back to Leaders Insights. OpenAI's measurement gap is capping ChatGPT ad budgets, and why it matters this week. An agency runs a lead generation campaign inside ChatGPT, the forms are filling up on their side, and the platform reports zero conversions. Two weeks later, still zero. Is that a bug or is that the product?

0:24Expert:That is the Accuracast case Digiday published on October 1, and I would call it a product that is younger than its ambitions. The agency ran a UK lead generation campaign on ChatGPT alone, watched the form fills come in, and two weeks later the platform still showed zero conversions. Lead generation here means a campaign whose success is someone filling in a form, so the conversions are the entire point. The same agency found ChatGPT reporting around 100 clicks where its own tracking picked up 20, and OpenAI's support team at the time could not say why.

1:01Host:A hundred against twenty. That is a five to one gap on the most basic number in digital media.

1:06Expert:It is, and the direction matters. Accuracast's group CEO Farhad Divecha told Digiday the problem on Google and Meta tends to be the opposite, with the platforms claiming more conversions than they drove. So the usual industry suspicion, that a platform flatters itself, is not quite the story here. One agency saw more clicks reported than it could find, and conversions it could see going unreported. That is noise in both directions, which is harder to plan around than a known bias.

1:38Host:Before we build a thesis on one agency, who disagrees?

1:41Expert:Plenty. Digiday says execs from AdRoll, Jellyfish and another independent US agency had not seen click discrepancies. A different firm, Web Guide Partner, framed it as a timing problem: conversions that do register arrive 24 to 36 hours late against a few hours on Google, and the reporting carries no demographic or prompt data. My opinion, not reported fact: late and thin is a workable problem, missing is not. If you cannot see the outcome at all, you cannot defend the line item.

2:16Host:Let me put the uncomfortable version to you. Marketers are complaining about a platform they themselves refuse to feed. Is this not partly self-inflicted?

2:25Expert:Largely, yes. Those tracking problems only apply to advertisers who have switched conversion tracking on, and many have not, because doing so means signing OpenAI's tracking terms, which legal teams at a lot of advertisers will not sign, citing liability. Digiday describes it as a catch-22: advertisers need to give OpenAI better data for its conversion tracking to improve, but the advertisers with the most data are the least willing to hand it over, and without reliable conversion data OpenAI cannot offer the optimisation that makes Google and Meta useful to performance marketers.

3:03Host:So what are people actually buying on, in the meantime?

3:05Expert:Shallow signals. Many advertisers are still optimising toward clicks or impressions, though that is starting to change as conversion-optimised bidding becomes more common, giving some ability to bid toward cost per acquisition, the price you are willing to pay for one customer action. Jellyfish's Victor Batista told Digiday the bigger shift comes when OpenAI builds optimisation toward metrics like cost per acquisition or return on ad spend, which is revenue divided by media cost, and that requires OpenAI to have confidence in its own numbers first. Until then, advertisers are bidding with little to go on.

3:46Host:Delivery itself sounds unstable too.

3:48Expert:Divecha's words to Digiday were that the biggest issue is inventory, that it is not predictable, that there is room to serve more, and that there is no obvious explanation for the change. Inventory here means the number of chat responses available to carry an ad. If that supply moves without warning, your spend pacing moves with it, and a weekly report stops being a report and becomes a guess.

4:13Host:Why is anybody still in, then?

4:15Expert:Two reasons, one commercial and one political. The commercial one: this is now a real buying channel. Cost per click has been introduced since launch and now accounts for most of the ad spend, according to Digiday. In Europe, advertisers in the new markets can buy through Publicis, Omnicom, WPP, Havas, Dentsu and MediaPlus. The political reason is that no CMO wants to explain to a board why a competitor was in the room first. Batista put the honest version on the record: the platform is still very much unproven, very new, built at a fast pace, and the platform in use now is not the one released in the pilot, it has grown massively.

4:59Host:Independent verification. Where is it, and does it fix any of this?

5:03Expert:It is arriving slowly and it fixes less than people hope. OpenAI's ads and monetisation lead Asad Awan told a press briefing that third-party measurement was in the works, without partners or an exact timeline. Since then LiveRamp, the data matching firm, became the first independent ad tech company to pipe conversion data into OpenAI's conversions API, which is a server to server feed that sends purchase events back to the platform, and that deal is tightly scoped: US only for now, limited to select mutual clients. OpenAI has also signed mobile measurement partners including AppsFlyer and Adjust, which matters because so much ChatGPT use happens in the app. And the caveat that should go on a slide: Divecha said a verification partner would confirm ads were seen without showing whether they worked.

5:52Host:Give me the strongest argument against your own caution.

5:56Expert:It is a good one. Every large ad platform looked like this in year one, with thin reporting, odd delivery and no outside auditor, and the advertisers who learned the auction early paid less for years afterwards. Digiday also notes OpenAI is carrying infrastructure costs that far outweigh revenue, and ads are now core to lightening that load, which is a strong incentive to fix measurement fast rather than slowly. If you wait for a mature measurement stack, you will arrive when prices reflect certainty. The counter to the counter is that a two week blind spot on conversions is not a rounding error, it is the difference between a test and a budget transfer.

6:37Host:You have one minute. What should a marketing leader do this week?

6:40Expert:One thing, and it is not a bigger budget. Decide your measurement before you spend, in writing. Mile Marker's vp of biddable centre of excellence Monica Shukla told Digiday it was important to establish an independent measurement framework before testing the platform rather than taking its own reporting at face value, and her point was to define the KPIs first and go back to them, instead of accepting platform-first metrics and data at face value. In practice that means a geographic holdout, where you switch the channel off in matched regions and compare, so the answer does not depend on OpenAI's own counting. Then do the boring task: put the tracking terms in front of your legal team now, with a decision date, because you are currently arguing about numbers you have refused to collect.

7:29Host:And what to watch.

7:30Expert:Two markers. Whether OpenAI names third-party measurement partners with a timeline rather than an intention, and whether conversion-optimised bidding starts delivering a stable cost per acquisition in your own category. Until both move, keep this funded from test money and keep your reconciliation sheet open next to the dashboard.

7:50Host:Sources for today's episode: OpenAI's measurement gaps are keeping ChatGPT ads budgets at test level, Delivery Apps And Retailers Unite; How Do You Measure, Measure An Ad, OpenAI takes ChatGPT ads into Europe, OpenAI opens up ChatGPT ads manager to the U.S. while promising third-party measurement, CPA bidding, OpenAI’s ads get their first measurement partner in LiveRamp. That's it. The CMO decision tools are live at mba-training.com.