Intent-Based CPM Pricing vs Click-Based Buying in AI Channels

A user types "running shoes" into a search bar. Another user asks an AI assistant what shoe to buy for a first half marathon on pavement. Both are shopping. Only one has told you almost everything a marketer could want to know: the distance, the surface, the experience level, the moment of decision. Traditional advertising has never had that second sentence to work with. It has only ever had proxies, a page load, a keyword match, a scroll past a banner, each one a guess dressed up as a signal. In a conversational exchange, the guessing stops, because the user has already said what they want, in their own words, in real time.

That changes the unit being sold. An impression next to a search result is a bet on inferred interest. An impression inside a conversation carries declared intent, with detail no cookie or lookalike audience could reconstruct. Whether a click follows or not says less about whether value was created and more about whether the user happened to act in that particular session. So the debate over whether to buy this inventory by the thousand impressions (CPM) or by the click (CPC) is really a debate about where value gets created. It is a debate about where, in the arc of a conversation, value actually gets created: at the moment intent is expressed, or at the moment someone acts on it. Everything that follows in the AI advertising market right now, the pricing swings, the auction design, the measurement gaps, traces back to that unresolved question.

The AI advertising market and the still-small slice of chatbot inventory.

The numbers attached to "AI advertising" are large enough to obscure what they actually contain. The overall AI-in-advertising market was valued in the tens of billions of dollars in 2024, and forecasts point to dramatic growth through 2030, with compound annual growth rates that outpace most of digital advertising. Those figures get repeated often, and they are measuring something broader than most people assume. They are just measuring something broader than most people assume.

The bulk of that number comes from AI-generated creative, AI bidding optimization, and AI-augmented search, with most of it not conversational ad inventory. Almost none of it is ad inventory sold inside a chatbot conversation itself.

The chatbot slice, specifically, is still small. EMARKETER projects US chatbot advertising revenue will stay below $1 billion in 2026, a fraction of the broader AI ad figure and a reminder of how early this inventory actually is. A market that small has not settled on pricing norms yet. Auction dynamics, floor prices, and buyer expectations are all still being negotiated in real time, which is exactly why the CPM versus CPC choice carries more weight right now than it would in a mature channel. Whatever conventions take hold in the next year or two will shape how this inventory gets bought for a long time afterward.

How ChatGPT's ad launch moved from CPM to CPC in ten weeks

OpenAI officially confirmed advertising tests on January 16, 2026, and ChatGPT ads went live on February 9, 2026. The format was deliberately restrained: a 50-character headline, a 100-character description, a link, an image no smaller than 256 by 256 pixels, no video, positioned below the AI's response rather than folded into it. Access was tightly controlled too. Only invite-only, logged-in adult users on the Free and Go tiers in the US saw ads at all; Plus, Pro, Business, Enterprise, and Education users saw none, a constraint OpenAI's own documentation spells out clearly.

The launch price reflected that scarcity. Roughly $60 CPM, with a $200,000 minimum commitment, brought in launch partners like Target, Adobe, Williams-Sonoma, and Albertsons, a price built to ration a small amount of inventory among a handful of large brands willing to pay for early access to something unproven. That is a price built to ration supply, not to reach a wide market. It is a price built to ration a small amount of inventory among a handful of large brands willing to pay for early access to something unproven.

Then the arc bent. Read in sequence, this is not a random walk. CPM launched high to throttle demand during a pilot phase, softened as more inventory came online and fewer scarcity constraints applied, and CPC arrived as the mechanism that gave performance marketers, who think in cost-per-action, a way in.

It would be easy to read that sequence as a verdict against CPM. The more accurate reading is narrower. The market signaled that the initial CPM, however rich the intent behind each impression, was not a price performance buyers would hold onto without a click-based alternative sitting next to it. But that same sequence also suggests the original CPM was partly an artificial ceiling set by scarcity and pilot-phase caution, not a considered rejection of impression-based buying as a concept. An annualized run-rate near $100 million within six weeks of launch says advertisers showed up with real appetite. It does not say the market has found its equilibrium price. Ten weeks is too short a time to settle something this new. It is enough time to observe the first correction.

Why CPM is not the wrong model for conversational inventory, it is the right model applied to the wrong question

The initial $50–60 CPM was compared unfavorably to Meta's $20–25 CPM. That comparison is tidy, but it compares two different things as if they were the same thing. A feed impression on a social platform is an inference: the platform guesses, from browsing history or demographic modeling, that a person might care about a product. A conversational impression is fundamentally different. The user has already described, in plain language, what they want. Nobody guessed. They asked.

So the question a CPM on this kind of inventory actually needs to answer is not "how many thousand impressions did this buy." It is closer to "what is an impression built on stated intent worth, relative to a click from a user with no declared context at all". Those are not comparable units, and pricing them as if they were flattens the distinction that makes conversational advertising interesting.

Early conversion data gives that question some shape. Criteo's figures show LLM referrals converting at approximately 1.5x the rate of other referral channels, approaching 2x in key retail categories. If that lift holds up across categories and over time, the arithmetic on CPM shifts in a way that headline price comparisons miss entirely. A $50 CPM that converts at a meaningfully higher rate can produce a lower effective cost per acquisition than a $20 CPM that does not. The intent quality baked into the impression belongs in that calculation, not just the sticker price sitting on top of it. CPM was never the wrong tool here. It was priced, at launch, as though the impression underneath it was ordinary, when the entire premise of the channel is that it isn't.

Why CPC bidding appeals to performance marketers in a conversational context.

CPC has a straightforward appeal that shouldn't be dismissed. Paying only when someone clicks ties spend directly to observable action, and that is a familiar, defensible accountability model for any performance team that has to justify a budget line to a finance department. ChatGPT's $3–5 CPC is between Meta and Google Search by cost, which puts the real question where it belongs: not on the headline rate, but on conversion quality.

There are objectives where this fits well. It is a legible model, and legibility has real value when you're reporting results upward.

Where the logic strains is at the seam between conversation and action. A user can read a sponsored suggestion inside a chat, absorb it, quietly narrow their consideration set to include a brand they hadn't thought of, and never click in that session. Something changed in their thinking. CPC records none of it, and assigns that influence a value of exactly zero. A second distortion accompanies the first: click-through rates on a surface built around fewer ads, shown to fewer people, at higher-stakes moments, produce modest-looking numbers next to a social feed by design, not by failure. Judging the model by a CTR built for a different kind of surface mistakes restraint for weakness.

The sharpest version of the problem is coming. As AI interfaces get better at resolving a query fully inside the conversation itself, without sending anyone anywhere, CPC becomes a thinner and thinner measurement of what actually happened. It counts the click the platform facilitated. It says nothing about the influence the conversation exercised on everything that came after.

The auction mechanics inside ChatGPT, where relevance beats budget in a way search bidding does not.

Nothing about ChatGPT's targeting resembles a keyword auction. The targeting relies on no keyword layer. The primary targeting input is a 280-character natural-language context hint, set at the ad group level, matched against whatever the user is actually saying in the live conversation. That alone marks a break from two decades of search advertising built around bidding on phrases.

The auction itself weighs relevance alongside the underlying context hint, the ad's headline and copy, and the landing page it points to, and all three of those signal how relevant the ad actually is to what's happening in the conversation. That means creative precision and targeting discipline carry more structural weight here than they do in an auction where the size of the bid tends to dominate. A well-written context hint, matched to a real conversational pattern, can beat a bigger budget attached to something vague. That is not how search bidding has traditionally worked, and it is most exploitable right now, while most advertisers on the platform are still writing context hints the way they'd write keyword lists.

This mechanical detail reshapes the CPM versus CPC decision rather than sitting apart from it. Buying on CPM means paying for the chance to show up in a relevant moment. Buying on CPC means paying only for the confirmed action that followed. But because the auction rewards relevance directly, a CPM buyer who writes a precise, well-matched context hint competes favorably on cost per relevant impression, not merely on cost per click. Precision is a core requirement of good creative practice on this surface. It is the mechanism that determines what you actually pay.

What Microsoft Copilot's performance data and Google's AI Max results suggest about intent alignment and channel economics

The clearest outside evidence for the stated-intent thesis comes from adjacent platforms, and it needs to be handled carefully. Microsoft reports that ROAS climbs substantially when a user engages with Copilot before running a search. That's a vendor claim, not an independent study, and it should be read as one. Even with that caveat, the direction of the claim lines up with everything the ChatGPT sequence suggests: contexts where a user has already articulated intent to an AI system tend to convert better than contexts where they haven't.

Google offers a similar shape of evidence. One cited example showed a large revenue increase after a brand turned on AI Max. Again, that's a single reported case rather than a controlled study, and it shouldn't be treated as proof of a general rule. But taken together with the Copilot figure and the Criteo conversion data, a pattern starts to firm up: when an ad meets a moment of expressed need rather than a demographic guess, the economics downstream tend to move in the advertiser's favor.

That pattern is exactly why the question of which pricing model captures and rewards intent alignment matters as much as it does. If intent alignment is what's driving the lift, then the real question is which pricing model actually captures and rewards that alignment, rather than which one is more familiar. CPM on a high-intent surface prices access to the intent moment itself. CPC prices only the behavioral confirmation that follows it, and in cases where the conversion happens somewhere else, or later, CPC may record nothing at all, even though the intent moment already did the work.

The measurement gap that makes the CPM vs. CPC choice harder than it looks

None of this would matter much if measurement inside these platforms were solved. Measurement inside these platforms is not solved. Advertisers get aggregate numbers back, impressions, clicks, conversions, and nothing at the level of the individual conversation.

That gap is not cosmetic. Conversational influence often happens well before the action that eventually gets measured, and often on a different surface. A user asks ChatGPT what shoe to buy for a first half marathon, reads a sponsored suggestion inside that exchange, and buys the shoe three days later through a completely different channel.

Under CPC buying, that entire sequence registers as zero spend and zero return, which means the model is structurally built to undercount whatever the channel actually contributed. Under CPM buying, the advertiser has paid for the impression up front and has to trust that impressions built on real intent produce lift somewhere downstream, a trust that only holds up with incrementality testing, assisted-conversion modeling, or brand-lift studies standing behind it, none of which last-click attribution can supply.

Neither model solves how to attribute value to the right moment in a conversation. Each one simply assumes a different answer to it. CPC assumes the click is the value event. CPM assumes the moment intent gets expressed is the value event. Neither model solves how to attribute value to the right moment in a conversation; they assume different answers to it (CPC assumes the click is the value event, CPM assumes the intent moment is the value event), and choosing between them is choosing an attribution theory, not just a cost structure.

How to think about which model fits which objective in AI channels today

None of this resolves into a clean verdict, and it shouldn't. The right framing is a narrower and more useful question." It's a narrower and more useful question: what signal actually represents the value a given campaign is trying to capture?

CPM tends to fit better when the goal is brand consideration or category entry, since the intent moment itself is the thing being paid for, not whatever click may or may not follow it. It fits when an advertiser already has the measurement infrastructure, incrementality testing, marketing mix modeling, assisted-conversion analysis, to trace influence across sessions and surfaces rather than relying on last-click data alone. It fits when the conversational context is highly specific, since a user asking a detailed, product-category question is a rare and valuable audience moment that CPM prices directly rather than approximates. And it fits categories with long consideration cycles, financial services, healthcare, travel, high-ticket retail, where the click almost never lines up with the actual purchase decision.

CPC fits a different set of circumstances. It works when the goal is immediate traffic or lead capture and the landing page can close the loop inside a single session. It works when an advertiser doesn't yet have the infrastructure for multi-touch attribution and needs a cost model that is self-evidently accountable to whoever signs off on the budget. And it works in low-consideration categories, where the decision that follows a click can complete itself in the same sitting that triggered the ad.

The channel is still new enough that most advertisers are borrowing decision frameworks built for search and social, because those are the frameworks on hand. Conversational inventory doesn't behave like either one. It carries a signal, stated intent, that neither the CPM conventions of feed advertising nor the CPC conventions of search were built to price. Getting that pricing decision right starts with naming, honestly, what a campaign is actually trying to capture, and then choosing the model built to reward it.