← Back to blog
Measurement

How to measure ChatGPT ads when the only read was OpenAI's own

By zAI

How to measure ChatGPT ads when the only read was OpenAI's own

Ads in ChatGPT are no longer an experiment. In our note on what early testers reported we followed the Criteo conversion data, the Ahrefs CTR figures and the tensions of a channel that is scaling before its measurement infrastructure is mature. The annualized revenue of the ChatGPT Ads pilot was reported at over $100 million in its first weeks, with projections near $2.5 billion for 2026, though both figures come from the platform or its partners and should be read as interested-party signals, not independently verified evidence.

The gap that persisted in that picture was independent measurement. From launch, advertisers depended on what OpenAI returned in its own Ads Manager: impressions, clicks, spend, CTR, average CPC and conversions, all aggregated with no way to cross-reference against an external source. It was the same problem programmatic had around 2015, before the first verification vendors appeared: the channel was running, the numbers were whatever the platform reported, and nobody could say how much of it was real.

What Profound launched on July 7

Profound Ads Studio is not a standard performance tracker: it is a platform that combines measurement and campaign generation for AI search engines. The distinction matters because most ad intelligence tools were built to track links in traditional search results, not conversations in natural language environments.

The platform introduces two proprietary metrics. Paid Share of Voice measures how often a brand's ad appears, relative to competitors, in conversations that return sponsored results. It is the conversational version of the classic media SOV, applied to an environment where ranking is not a list of links but a generated response that may or may not include advertising. Relevance Score consolidates three dimensions into a single number: how well the ad matches the audience making the query, whether the message is specific enough to motivate action, and whether the brand and the conversational context are compatible.

The dataset the platform runs on is 1.9 billion real user prompts, which allows evaluating query distribution before launching a campaign and scoring hundreds of candidate ads against that corpus. The underlying logic is that meaningful optimization in AI search is not about keyword bids but about semantic fit within the specific context of each prompt: an ad can have a high Relevance Score in one query cluster and a low one in another that superficially seemed equivalent.

Guideline, same date, another piece

On the same July 7, Guideline launched verified ad intelligence for AI platforms, with reported coverage of approximately $200 billion in media investment data across 65 countries. Where Profound attacks the problem from presence measurement and ad generation, Guideline attacks from investment intelligence: how much is being spent in the channel, by category and market, from a source that is not the platform itself. They are two different pieces of the same puzzle, and the fact that both appeared on the same day does not seem coincidental: the independent measurement market for AI ads is organizing itself as an industry.

Both Profound and Guideline are companies with a commercial interest in their platform launches, and their own claims about data quality or coverage should be read with the same standard applied to any vendor presenting a new product: the only way to verify is to test it.

The pattern we already know

The parallel to programmatic circa 2015 is not decorative. At that stage the channel was already distributing billions in investment, the reports came from the buying and selling platforms themselves, and the question of fraud, brand safety and real viewability had no external answer. The first independent verification vendors, Integral Ad Science, DoubleVerify and those that followed, did not create the problem: they made it visible. And those who adopted them early had better-grounded budget allocation decisions during the years it took the ecosystem to standardize those metrics.

The difference in AI search is that the problem is more structural: it is not just about verifying whether the ad was served in the right context, but about measuring presence in a conversational environment where the unit of analysis is not an impression on a page but a generated response that can change with every query. Profound solves exactly that layer: it converts "did my ad appear?" into a relative frequency metric against competitors, and adds the dimension of whether that ad was relevant to the conversation where it appeared.

The tensions that remain

There are at least three tensions worth naming before projecting how this gets used. The first is access: Profound did not publish pricing or a commercial model at launch. For medium or small teams, a measurement tool without a clear cost is difficult to evaluate as an investment, and the history of verification vendors in programmatic suggests that costs for this layer are not trivial. The second is geographic coverage: the 1.9 billion prompt dataset has no public breakdown by market, and it is not clear how well Argentina, Chile, Colombia or Mexico are represented in that data. The third is interoperability: how Paid Share of Voice and Relevance Score interact with OpenAI's pixel and Conversions API, already configured by whoever entered the channel, is not yet documented.

None of those tensions invalidates the relevance of the launch, but they do define what type of operation can benefit from this today versus twelve months from now.

Who this is for (and who it is not)

For a large team with active ChatGPT Ads campaigns, adding an independent measurement layer is the natural move: any efficiency analysis of the channel that relies solely on OpenAI's dashboard inherits the same problem Profound is designed to solve. The relevant question is evaluating how representative the prompt dataset is for the specific category before making scaling decisions based on the Relevance Score.

For a mid-sized team evaluating channel entry, the value of Profound at this stage is primarily as a pre-launch diagnostic tool: assessing the competition's Share of Voice and the distribution of relevant queries before spending, not only after. It is worth exploring access.

For operations in LATAM markets without a confirmed ChatGPT Ads launch date, the move right now is not subscribing to Profound but understanding the measurement framework that will be used when the channel arrives, so there is no starting from scratch when the launch is imminent. The preparation window is now, not the day after the announcement.

For small teams without dedicated analytics resources, this measurement layer is likely premature at this stage: the channel itself is still maturing in most markets, and the priority is having server-side conversion infrastructure running before adding a third-party measurement tool on top of it.

What we do not know

Profound did not publish Ads Studio pricing. There is no public information on which markets have significant representation in the 1.9 billion prompt dataset, which is relevant for evaluating the usefulness of the metrics outside the United States. Interoperability with OpenAI's Conversions API and other ad servers is not documented. And it is not clear whether the Relevance Score has industry benchmarks or whether each advertiser starts from zero when interpreting what number is good. Guideline has also not published what fraction of its $200B dataset corresponds to AI platforms versus traditional media.

Sources

This content was developed with AI assistance and reviewed by the Zenda team. The bad ideas are 100% ours.