CPM, cost per thousand impressions, is the oldest pricing unit in digital advertising. It originated as a way to put a number on something inherently fuzzy: how many people had a chance to see an ad. A page loaded is a potential impression. Enough potential impressions, priced at a rate reflecting the audience quality, and publishers could run businesses and advertisers could plan budgets.
The unit has always been imperfect. The page view does not tell you if someone read the article or bounced in two seconds. Viewability standards were an attempt to address this, requiring a minimum portion of the ad unit to be visible for a minimum duration before counting as an impression. Even so, a page view is a coarse proxy for the thing advertisers actually want: a moment of attention from a person who might care about what they are selling.
A conversation turn is something different. Not better in every dimension, but genuinely different in ways that matter for how we think about ad impressions and what we should measure.
What a Conversation Turn Actually Represents
When a user sends a message to an AI assistant and receives a response, that exchange is bounded in a way a page view never was. The user's attention is sequential and active. They wrote something, they are waiting for a response, they are reading that response. The cognitive mode is query-response, not browse-and-scroll.
This has practical implications for what an impression means. A user reading an assistant's response to a specific question about home loan refinancing is not in the same attentional state as a user who landed on a home finance article page and started scrolling. The question they asked is direct evidence of what they are thinking about right now. The response they are reading is directly engaging that thought. An ad inserted into that moment, if it matches the topic, is entering a focused information-seeking context.
That is the fundamental argument for why conversation turn should become a first-class advertising unit. Not because it replaces page views, but because it captures something page views never could: the moment of explicit, articulated intent.
How Publishers Should Think About Monetizable Turns
Not all conversation turns are monetizable. A user sending a one-word message, a turn containing flagged content, a turn in a session where the advertiser demand for that topic is thin. Publishers should think of their monetizable turn volume as a subset of total turns, shaped by topic mix, session depth, and fill rate.
A useful way to model this for planning purposes: take your monthly conversation volume, apply a conservative fill rate estimate (for reference, our early-access publisher program has been averaging fill rates in the 55-70% range for assistants with good topic-advertiser category overlap, based on internal data from the program), then price that at your expected RPM per thousand monetizable turns. Our published benchmarks from the early-access cohort give concrete figures to anchor that estimate. The result is a revenue estimate more grounded in what the inventory actually represents than a raw turn count would be.
Fill rate is where publishers often underestimate variability. An assistant that handles a broad range of topics may have excellent intent quality on the turns it does fill, but large swaths of its conversation volume fall into topics with no advertiser demand. A specialized assistant covering a narrow vertical with strong advertiser interest tends to have higher fill rates against lower total volume. Both can produce comparable revenue per active session; they just get there differently.
The RPM Question: What Is a Turn Worth?
Publishers coming from display advertising instinctively try to map conversation turn revenue to familiar CPM benchmarks. This usually leads to confusion because the comparison breaks in multiple directions at once.
A display CPM on a content site might be $2-$8 for general-audience inventory, higher for premium finance or technology categories. That CPM covers a thousand page views. A page view may last 30 seconds or 10 minutes; the CPM does not distinguish. A conversation turn, by contrast, is specifically the user-assistant exchange. You might have 5 turns in a session that takes 15 minutes. The turn is a denser unit of attention than the page view by construction.
When we express publisher earnings as RPM at Velocity, we are pricing per thousand conversation turns delivered to the scoring system, not per thousand that produce a filled ad. The RPM metric in your dashboard represents revenue from successful matches spread over all eligible turns, which means fill rate directly affects the RPM number. A higher fill rate produces a higher RPM against the same underlying CPM bids.
This is an important distinction. When comparing Velocity's RPM to display RPM, you are not comparing like to like unless you also account for the different density of the underlying unit.
Session Depth and Revenue Concentration
One pattern we see consistently across publisher data is that revenue concentrates in longer sessions. A user who has a five-turn conversation generates roughly twice the monetizable opportunity of a user who has a two-turn conversation, and the intent signals tend to be more consistent across a longer session, which improves match quality.
Publishers who understand this optimize their assistant products to encourage depth, not just volume. A quick one-turn FAQ lookup is useful for the user but thin from a monetization standpoint. An assistant experience that naturally guides users toward follow-up questions, exploration of adjacent topics, and comparison-style conversations generates both better user outcomes and denser monetizable inventory.
We are not suggesting that session depth should be engineered artificially to capture more ad impressions. That would damage user trust faster than any revenue benefit would compensate. The point is that publishers who build genuinely useful, exploratory assistants tend to generate better monetization outcomes as a natural byproduct of building a better product. The incentives are aligned, not in tension.
What the Transition Means for Advertisers
For advertisers, the shift from page-level impression to conversation turn impression requires rethinking targeting and measurement. You are no longer targeting against audience segments defined by browsing history or demographic inference. You are targeting against live, declared topic intent expressed in the current conversation. This is a different kind of signal with different properties.
The upside: the intent signal is often clearer and more specific than anything you can infer from behavior. Someone actively asking an AI assistant how to compare credit card cashback programs is demonstrating financial product intent more explicitly than someone who visited a financial news site three days ago.
The question advertisers will reasonably ask is how this scales and how it compares to search intent. We think honest answer is that it is currently a smaller channel with different inventory characteristics and emerging measurement standards. The per-unit intent quality is high; the total volume is constrained by how widely AI assistants are deployed with monetization enabled. Both of those facts will change as more publishers integrate conversation monetization into their products.
The page view had about 25 years as the dominant advertising unit before programmatic fragmented it into viewability, engagement, and attention metrics. The conversation turn is earlier in that arc. Publishers and advertisers who build fluency with it now are building knowledge that will compound as the format grows.
