ChatGPT Ads: how to judge a channel that has no benchmarks yet
Advertising on ChatGPT has arrived in Italy. We opened the account and started planning. Here is how it actually works, and why the first numbers mean nothing without downstream measurement.
Luca Lobascio
5 min read
Updated September 2026. ChatGPT Ads is in beta and OpenAI states that delivery systems, inventory, ad formats and buying options will change during this phase: what follows describes the platform as it stands today.
Ads Manager is now available self-serve to Italian advertisers. We opened the advertiser account, set up the first campaigns and configured measurement. We have no results to report yet, and anyone reporting results after a few weeks is reporting noise.
What we can describe is how the platform is built, where it differs from the ones we know, and the only serious way to tell whether it works.
You do not buy an audience, you buy a context
On Meta and TikTok you also work on audiences and behavioural signals; Google Search picks up a demand already expressed through queries. ChatGPT Ads introduces a different logic again: the context and intent of the conversation.
At ad group level you provide context hints: you describe the conversations, topics and situations where your product is relevant. OpenAI is explicit on a point worth reading twice: they are not exact-match keywords and do not guarantee delivery in specific conversations. They steer the match, they do not determine it.
On top of that, personalised ads are currently not available in the European Economic Area or Switzerland. In Italy selection relies on the context of the current conversation, the language and the general location. Anyone selling you «advanced targeting» on this channel is describing something that does not exist here.
What to know before planning
- The audience is the people who do not pay Ads appear on the Free and Go plans. Plus, Pro, Business, Enterprise and Edu users never see them, and neither do under-18s. That is a planning decision, not a footnote.
- The ad sits below the answer Advertiser name, logo, headline, body, image and landing page, clearly labelled as sponsored and separate from the answer. Advertisers cannot influence what ChatGPT replies.
- Contexts excluded by policy No ads near personal health, mental health or politics. Political advertising is not allowed. Regulated sectors are eligible only under strict criteria.
- One account per advertiser There is no Business Manager-style structure: every advertiser has its own account, with its own email, billing and verification. The client completes onboarding, verification and billing; the CleverAd team can then be invited in to manage it. Country, currency and time zone cannot be changed after creation.
- Verification takes its own time Registration, identity and business verification, then manual review by OpenAI. It cannot be expedited: you promise a launch date after approval, not before.
The point that matters: there are no benchmarks
OpenAI says it plainly: ChatGPT Ads does not yet have performance benchmarks for advertisers, industries or campaign types.
Which means nobody can tell you whether your CPC is good. There is no industry average, no history, and the first data from a new platform is the least representative you will ever see. The only public figure is OpenAI's recommendation to start from a maximum bid of 3–5 dollars per click on Clicks campaigns with manual bidding: a suggested starting point, not an average CPC, not a market benchmark and not an expected cost.
With no benchmarks there are two routes. Wait for the market to produce some, and learn nothing meanwhile. Or build your own yardstick: measure that traffic through to commercial outcomes and compare it with the channels where you already have one.
Measurement is possible, but you have to build it
The good news is that the channel is trackable. Static tracking parameters — ordinary UTMs — persist across ad clicks and reach the landing page, where existing analytics tools read them with no special integration. Ads Manager reports impressions, clicks, spend, CTR, average CPC and CPM, and conversions once measurement is configured.
ChatGPT Ads does not stop at reach and clicks: Ads Manager supports campaigns optimised towards conversion events, measured through the Pixel, the Conversions API or both. At the time of writing, conversion-optimized models bill on clicks and, in beta, on impressions.
The bad news is that a conversion objective is worth exactly as much as the signal it receives. If tracking is not right, a campaign optimising on leads optimises on nothing: it delivers, it spends, it learns nothing. The principle holds on every channel, but it weighs more here, because there is no history to compensate.
So the first thing to do is not to raise the budget: it is to verify that the conversion event actually fires, with a manual test, before spending.
Conversions can be sent through the Pixel, the Conversions API or both. In the campaigns we are setting up we use the conversions objective and the Conversions API as well, to add a server-to-server signal and reduce the reliance on the browser alone. On a channel with no history that is the difference between readable data and an anecdote.
How we approach it
We treat ChatGPT Ads the way we treat any new source: we bring it into the same measurement system as the others. CleverAd Server™ attributes source, click and lead; CleverAd Engine™ puts this channel's CPL, redemption and real CPA alongside Meta, Google, TikTok and DEM.
Not because it is a special channel, but because it is the only way to answer the one question that matters — did those clicks become customers? — when the platform itself has no answer to give you yet.
ChatGPT Ads joins our Paid Media Management like any other channel: accounts and data stay with the client, and optimisation looks at conversions.
If you are considering ChatGPT Ads for your company and want it set up so the first data is readable rather than anecdotal, let's talk.