ChatGPT Ads: why budgets are still in the test phase

One client spends $10 million a month on Google and less than $100,000 on ChatGPT — not because they don't want to spend more, but because they can't tell if the money is working.

Seven months of advertising on ChatGPT — and we still can't tell what works

OpenAI started selling ad space inside ChatGPT in early 2026. Seven months later, the main problem isn't the cost of ads or the quality of the audience: it's that no one can measure results reliably. Digiday reports this after speaking with several agencies running live campaigns on the platform.

This doesn't mean ChatGPT Ads doesn't work. It means that, today, you can't know for sure. And without certainty, scaling your budget is a leap in the dark.

What doesn't add up in the numbers the platform shows you

The most telling case: a lead generation campaign on ChatGPT produced completed forms visible in real time, yet after two weeks the platform still showed zero recorded conversions. Same scenario, opposite direction: the platform reported about 100 clicks on a campaign, while the external tracking system had detected 20.

Even when the data arrives, it arrives late. Some agencies see conversions show up 24-36 hours late, compared with the few minutes Google has taught us to expect. And the reports include no demographic information — age, gender, interests — nor the topics of the conversations where the ad appeared.

For anyone running campaigns across multiple channels, this creates a real problem: you can't compare ChatGPT with Google or Meta on equal terms, so you can't decide where to move your budget. If you want to dig deeper into how attribution models can mislead spending decisions, the article on advertising attribution and the models that fool your budget is a good place to start.

Why many companies don't even turn on tracking

To receive conversion data, you have to accept OpenAI's tracking terms. Many legal departments refuse, worried about liability for customer data and uncertain about how OpenAI will use it. The paradox is that these same companies have been sharing similar data with Google and Meta for years — but with those platforms, familiarity has built up over time. With OpenAI it hasn't, and the company hasn't shown any willingness to negotiate custom clauses.

There's also an API (a direct technical link between systems) that could improve data quality, but large companies are slow to implement it and smaller ones often lack the technical resources. The result is a vicious circle: OpenAI needs more data to improve tracking, but those with the most data are the least willing to share it.

At SHM Studio, we believe technology should serve decisions, not create new blind spots. Adopting an advertising tool without an independent measurement system isn't innovation: it's giving up control over your own budget.

Optimizing for clicks and impressions doesn't tell you if you're selling

Without reliable conversion data, most ChatGPT campaigns are optimized for clicks and impressions — that is, for how many people saw or clicked the ad. These are visibility metrics, not results metrics. They don't tell you whether anyone bought, filled out a form or requested a quote.

The platform is introducing the option to optimize toward cost per acquisition (how much you spend to get a customer or a lead), but it only works well if OpenAI has enough verified conversion data to calibrate bids. Today that data is scarce. Meanwhile, some agencies receive automatic alerts asking them to raise bids without saying by how much — alerts that sometimes disappear on their own after a few days, with no explanation.

Ad delivery is just as unpredictable: thousands of impressions one week, fewer than a hundred the next, for no obvious reason. This makes it hard to learn from one campaign and apply the lessons to the next. To understand how we got here, it's worth rereading the article on how OpenAI opened the door to advertising in 2026 and the one on how OpenAI is monetizing ChatGPT.

OpenAI has struck deals with a few measurement partners for mobile apps, but agencies report that the available integrations only confirm the ad was seen — not that it produced a result. The click-to-chat format, which keeps users inside ChatGPT instead of sending them to your site, complicates traditional tracking even further.

If you want a broader picture of how to evaluate AI channels in advertising, the article AI and advertising: why results matter more than hype offers a useful perspective. You can find more insights on the topic in the Marketing and performance ADV section.

What to check before increasing your ChatGPT Ads budget

• Do you have an external tracking system in place (Google Analytics, a CRM, an independent pixel) that measures conversions independently of the platform?
• Does your ChatGPT click and conversion data match your external system's, or are there significant gaps?
• Has your legal team reviewed OpenAI's tracking terms and given the green light?
• Are you optimizing for a results metric (leads, purchases, quotes) or just for clicks and impressions?
• Have you defined your success indicators (KPIs) in advance, rather than relying only on the platform's reports?
• Is your current budget small and sustainable as a test phase, with no pressure to scale before you have solid data?

Frequently Asked Questions

A channel you can't measure isn't an opportunity: it's a cost you can't be sure you're paying for any good reason.