ChatGPT Ads: $22 CPCs, zero benchmarks, no auction data

Six months after OpenAI began rolling out ads inside ChatGPT, the platform has produced one data point with any shape to it: CPCs ranging from single digits to $22. Search Engine Journal reports that advertisers working inside the system still cannot answer the basic question every media buyer asks on day one: is this result good?

That is not a teething problem. It is a structural one.

Google Ads took years to build auction transparency, but it launched with a coherent theory of relevance. ChatGPT Ads has no auction insights, no industry-level benchmarks, and no reliable audience data. Advertisers are running campaigns against a model that will not tell them who it is serving, at what clearing price, or how their creative is performing relative to category peers. A $22 CPC could be efficient for a niche B2B software buyer. It could be catastrophic for a consumer goods brand. Without context, the number is noise.

The benchmark problem is not cosmetic

Performance marketing is a discipline built on comparison. Every metric in paid search derives its meaning from a reference class: average CPC in your vertical, your own historical CPL, auction impression share versus competitors. Strip those reference points away and a campaign manager cannot optimise; they can only guess. What ChatGPT Ads has produced so far is a collection of isolated data points held by individual advertisers, none of whom can share enough to form an industry picture.

This is particularly acute for the sectors most likely to test the channel early. Financial services advertisers, where compliance teams require documented justification for media spend and where CPCs on Google routinely exceed $50, need benchmark data before they can build a business case internally. Multilateral institutions and policy-adjacent organisations, cautious about brand placement near AI-generated content, need audience verification before they will commit budget at all. Industrial groups running long B2B sales cycles need evidence that ChatGPT's conversational context produces qualified intent, not casual curiosity. None of that evidence exists yet.

OpenAI's incentive to fix this is obvious but not urgent. The platform is growing its ad inventory from a position of strength on the demand side: advertisers want access to a high-intent audience, and OpenAI knows it. Transparency costs the seller leverage. Google learned the same lesson and spent a decade being pushed, partly by regulatory pressure and partly by advertiser coalitions, toward more disclosure. OpenAI may follow the same slow arc.

What the opacity reveals about AI search as an ad channel

The deeper issue is that ChatGPT Ads sits inside a generative answer, not beside a search result. The ad is not competing with ten blue links; it is adjacent to a response the model has constructed. That changes the economics of attention in ways nobody has measured yet. A sponsored placement in a conversational answer could carry more weight than a sidebar ad, or less, depending on how users process the hybrid output. The honest answer is that no one knows, including OpenAI.

For brand visibility in AI-generated answers, this uncertainty cuts two ways. Brands that pay for placement are buying adjacency to model outputs they cannot audit. Brands that do not pay are being cited, or not cited, on the basis of training data and retrieval logic that is equally opaque. The paid channel and the organic channel share the same fundamental problem: neither offers the attribution clarity that B2B marketing teams need to justify allocation to their CFOs.

Six months of data producing a CPC range of $1 to $22 is not progress. It is the starting point. The benchmark question will eventually get answered, most likely when a third-party measurement firm aggregates enough advertiser data to publish vertical norms, the same role that WordStream and later Wordstream-era Google played in normalising paid search economics. Until then, ChatGPT Ads is an experiment that cannot be graded.

Brands willing to run ungraded experiments will accumulate the reference data that eventually becomes the benchmark. Everyone else will wait, and pay higher CPCs when the auction matures and competition arrives properly.