OpenAI builds tool to track whether ChatGPT ads convert

It is the latest sign of how quickly OpenAi is building out its ad infrastructure: last week, code reported by AdWeek pointed to upcoming support for conversion-based campaigns. Now there is evidence of the measurement layer needed to make them meaningful. When a user clicks on ads inside ChatGPT and then completes an action on an advertiser’s site — say, signing up to something, making a purchase or even booking a trip — the pixel fires, sends data back to ChatGPT and closes the loop. OpenAI knows the ad worked. The advertiser can prove it. 

The mechanics will be familiar to anyone who has run campaigns on Meta or Google. A small piece of JavaScript loads when a user lands on an advertiser’s page, and reports back when they complete a defined action. In the code Digiday reviewed, that action was a completed registration — a default example provided to developers to show how the pixel works. But it’s not the only action that will be available. OpenAI’s ads manager also lists other event types, including lead created, order created, page viewed, subscription created and trial started, as other actions that the pixel will track. 

The pixel itself is already live in parts of the platform, according to the exec who preferred to remain anonymous, but it’s not available to all advertisers in the pilot just yet. Instead, OpenAI appears to be selectively enabling the pixel for certain advertisers — effectively gating access as it tests and iterates the capability.

The result is the closed initialize-identify-measure loop employed by the alternatives from Meta and Google. But unlike those two, OpenAI is starting without the surround ecosystem advertisers have built up over the years to interrogate what the numbers actually mean — the media mix models, incrementality tests and third-party attribution tools that sit across multiple platforms (well sort of) and sense-check what any single pixel reports. That ecosystem takes time to develop on any new platform. Meta and Google earned advertiser trust through years of accumulated data, established attribution logic and iterative measurement infrastructure. OpenAI is at the beginning of that process. How quickly the market builds the tools and track record needed to benchmark ChatGPT’s numbers will go a long way to determining how fast performance budgets flow.

That is, after all, one of the biggest question marks hanging over OpenAI’s ad ambitions. Whether ChatGPT becomes a meaningful performance platform depends entirely on how people actually use it. Someone asking ChatGPT to compare SUVs, plan a trip or find the best protein powder is a performance advertiser’s dream. A user asking it to debug code or summarize a spreadsheet is worth considerably less. Right now nobody, including OpenAI, has a settled answer on which of those dominates.

“It’s a topic our clients in the pilot are asking about,” said Ashley Fletcher, CMO at Adthena, the ad agency, whose clients are among the first to be testing ads on the platform. 

A conversion tracking pixel, he continued, would address what is currently the most significant limitation of ChatGPT ads: the reporting right now is essentially impressions and clicks only, with no conversion attribution. So directionally, a pixel is a “welcome development” for any advertiser trying to justify spend or optimize campaigns, he added. 

For now, most of what is flowing into ChatGPT is test and innovation budget — marketers allocating smaller sums to determine the platform rather than committing meaningful media spend. The harder question, and the one that will define whether OpenAI’s ad ambitions amount to anything structural, is where that money eventually comes from. Net platforms rarely create new ad budgets, they redistribute existing ones. If ChatGPT’s usage does coalesce around high intent categories — think travel, auto, consumer electronic and shopping comparisons — it will start asking questions of the platforms that currently own those same signals. That is a significant if. ChatGPT’s 900 million user base doesn’t yet come close to Google or Meta’s advertising scale, nobody has confirmed the use cases break the right way, and the incumbents are not standing still. 

But it is the question the industry is beginning to ask — quietly for now, and with considerably more urgency than the test budgets currently flowing in would suggest.

As Collective Measures’ group director of performance media, Lauren Beerling explained, not every advertiser is willing to pull budget from their highest-performing lower-funnel channels to test an emerging platform. 

“What we’re actually seeing is budget shifting from the top of the funnel to fund what’s being positioned as a new direct response channel,” she said.

Getting the pixel is only the start. The harder problem is what happens to the signal inside a conversational AI environment, where the user journey is non-linear in ways that standard attribution models weren’t built for. Someone may engage with an ad inside ChatGPT but convert days later through organic search or direct traffic. Depending on the attribution model OpenAI ultimately deploys, that kind of journey could see ChatGPT’s contribution undercounted — a problem the company will need to address if it wants advertisers to trust the numbers. 

There is also a more fundamental technical headwind that has nothing to do with ChatGPT specifically. JavaScript pixels leak signal — privacy-first browser behavior and ad blockers have degraded client-side pixel reliability across the industry. Server-side implementation can mitigate much of that, and it is not yet clear which approach OpenAI is taking. But the broader point stands: OpenAI is building its measurement infrastructure at a moment when the underlying tool is under structural pressure industry-wide — a challenge it shares with every other platform running the same playbook but one worth watching as the system scales.

“If I ask ChatGPT to book a holiday and it surfaces Expedia because they paid, fine, it’s an engine,” said Robert Webster, founder of AI marketing consultancy TAU. “But if I ask ‘what’s a great laptop’ and the answer is whoever bid highest, the product breaks. The line between answer and ad is the whole thing. Google has spent 20 years defending that line and still gets it wrong half the time.”

OpenAI did not respond to Digiday’s request for comment.