The Unit Problem
Every advertising pricing model is a theory about what constitutes a unit of value. CPM says that 1,000 exposures to a message is the unit. CPC says that a click is the unit. CPA says that an action is the unit. Each model reflects a different assumption about where in the funnel advertising value actually gets created.
AI chat introduces a new unit: the conversation turn. A turn is a single exchange where the user sends a message and the assistant responds. Turns are not the same as page views. They carry more information (the content of the exchange), they imply an active state of mind, and they occur within a session structure that can be traced through a decision process. How pricing maps onto this unit is genuinely unsettled, and the way it gets settled will shape publisher economics for years.
We have been working through this at Velocity since we started building the platform in late 2023. Here is where the economics sit today, and where we think they are heading.
CPM on Conversation Turns: What the Number Actually Means
RPM in conversational advertising, which is CPM applied to conversation turns rather than page views, measures revenue per 1,000 turns. The metric is familiar enough that publishers who come from display advertising understand it immediately, and it makes reporting comparisons relatively straightforward.
What is different is the interpretation. A $2.00 RPM on conversation turns is not the same as a $2.00 CPM on display impressions, even though the numbers look the same. The attention quality of a conversation turn is higher: the user is actively engaged, has specific intent, and is reading the response rather than glancing at a banner. The advertiser is getting a different kind of exposure than a display banner sitting below the fold.
The practical consequence is that RPM benchmarks from display advertising are not directly comparable to conversational RPM. Publishers should not anchor their revenue expectations to display CPM figures and apply them to conversational turns. The relationship is more complex, and the direction of the difference likely favors conversational over the medium term as advertiser demand for the channel grows.
The current range we observe in our early-access data is $1.50 to $3.50 RPM on matched turns, with the variation driven primarily by topic category and publisher audience characteristics. Our cohort benchmark report breaks down how RPM and fill rate interact across different publisher types. Those numbers will shift as the advertiser market matures.
CPC in Conversational Contexts
CPC in a conversational ad requires a click on a sponsored card or inline link. The mechanics are similar to native advertising CPC on content platforms: a formatted unit appears, the user interacts or does not, and the advertiser pays per click.
Observed CTR on well-matched conversational ads in our early cohort runs between 1.5 and 4 percent, which is above typical display CTR but below what you would see on high-intent search ads. The spread reflects variation in topic category, ad creative quality, and how well the ad fits the moment in the conversation. A recommendation that appears directly responsive to what the user just asked will outperform a generic promotional unit that could have been placed anywhere.
CPC has a natural fit in conversational advertising because the click-through intent signal is cleaner than in display. When a user in the middle of a focused conversation clicks on a sponsored suggestion, that click represents a more deliberate choice than a click on a banner loaded alongside an article. Advertisers pay for a more qualified click, and the pricing should reflect that over time.
Where Existing Models Fit Poorly
Both CPM and CPC were designed for surfaces where the ad is visible simultaneously with content the user is reading. The user sees the article and the ad at the same time. In conversational advertising, the ad appears as part of a sequential exchange. The user reads it as part of a response, not as a parallel element.
This sequential structure creates measurement complications. Viewability standards, as defined by the MRC for display advertising, do not map cleanly onto a text response in a chat interface. There is no "50 percent of pixels visible for 1 second" criterion that makes sense when the unit is a line of text that the user reads. The industry will need to develop new viewability or engagement definitions, and until that happens, conversational CPM is being measured against a proxy that may not accurately capture delivered value.
Similarly, frequency capping in display advertising operates at the impression level across a session or a day. In conversational advertising, the natural unit for frequency management is the session or the conversation thread, not the turn. Showing an ad for the same product on turns 3 and 7 of the same conversation is a different experience than showing the same banner twice on different pages. Publishers need capping controls at the conversation level, not just the impression level, and that distinction affects how CPM is calculated when reporting.
What a Conversational-Native Pricing Model Might Look Like
There are two candidate models that make more conceptual sense for the conversational context than CPM or CPC.
The first is cost per conversation turn with intent threshold (CPT-IT): advertisers pay only for turns where the relevance score meets a minimum threshold they set. This is what we effectively run at Velocity under the CPM label, but it deserves its own designation because the key variable is the threshold, not the volume. An advertiser paying $4.00 CPT-IT at a 0.7 relevance threshold is buying something qualitatively different from an advertiser paying $2.00 CPT-IT at 0.5.
The second is cost per conversation engagement (CPCE): an action-based metric where the advertiser pays when the user continues the conversation in a way that indicates engagement with the sponsored suggestion. "Ask the AI a follow-up question about the sponsored product" is a traceable event in a conversational interface. It is not a click, but it is an intent signal that might actually be more predictive of purchase intent than a click.
Neither of these exists as a standard unit in the market today. They require platform-level tracking infrastructure that most conversational ad systems do not yet have. But the economic logic is coherent, and as conversational advertising scales, the pressure to move toward more granular intent-based pricing will grow from both sides: advertisers want to pay for demonstrated intent, and publishers want to demonstrate that their inventory is worth a premium over display.
The Current Practical Picture
For publishers making decisions today, the practical landscape is: CPM and CPC as the reporting currency, with RPM on matched turns as the primary publisher-side metric. Revenue per session (total earnings divided by unique sessions) is a useful secondary metric for understanding the economic value of your engaged AI assistant audience versus your display ad inventory.
The most important thing for publishers to understand right now is that conversational ad economics are still being established. Early participants in the market are shaping the pricing norms by what they accept, what they optimize for, and how they report. Publishers who engage seriously with measurement and reporting are contributing to a market infrastructure that will benefit the entire publisher side as the channel matures.
