The AI knows why you saw the ad. What about the advertiser?

Yesterday, we asked what an ad impression inside an AI conversation is actually worth.

The answer, broadly, was that conversational AI may be less about monetising the impression than monetising what several industry leaders described as the “decision moment”: the point at which a user’s need, constraints, timing and intent begin to converge towards a choice.

But that immediately raises the harder question.

If the platform knows why that moment is valuable, serves the ad against it and then tells the advertiser whether it worked, how does the advertiser independently know what actually happened?

That question sits at the heart of the next stage of conversational advertising.

Worth of ad impression in AI conversation. Read here 

Search, for all its imperfections, was relatively legible. The buyer could see the query, the ad, the click and the eventual conversion. Social was more opaque, but still offered audience definitions, behavioural signals and familiar measurement frameworks.

Conversational AI potentially flips that relationship.

The platform may understand the user’s intent far more deeply than either search or social, while the advertiser may see considerably less of the underlying context.

Nikhil Kumar, media technologist and former CGMO of Affle, sees that asymmetry as one of the central challenges facing the format.

“Search gives us keywords, social gives us audience signals; conversational AI could potentially understand intent at a much deeper level, while revealing much less of that intelligence to the advertiser,” Kumar said.

That does not necessarily mean the advertiser should see the conversation itself. Much of that context will rightly remain private.

But it does mean that the party with the deepest view of the user’s intent may also be the party deciding when the ad is relevant and reporting whether it worked.

For Kumar, that makes the standard buying currencies inadequate as proof of value.

“I would look beyond CPC and CPM to incremental reach, quality of engagement, consideration and, ultimately, business outcomes versus existing channels,” he said.

“If conversational advertising can demonstrate a materially better outcome because it understands the consumer better, it earns a premium. Otherwise, we are simply creating another media format with a more sophisticated story around it.”

Taranjeet Singh, ad sales veteran and APAC business leader, arrives at the same problem from a slightly different direction.

With search, the advertiser can see the keyword. With social, it can at least understand the audience and engagement signals being used. Conversational AI may understand substantially more about the user’s decision process, even as much of that context remains inaccessible.

“That makes it harder for advertisers to understand exactly what they are buying,” Singh said.

And if the buyer cannot inspect the signal, then the burden shifts to measurement.

Singh said he would not judge conversational advertising simply on CPM, CPC or even CTR, but would instead look for incremental conversion and revenue lift, new customer acquisition, brand-search lift and results from proper control or holdout groups.

The test, in other words, is not whether the ad was present before a conversion.

It is whether the ad created an outcome that would not otherwise have happened.

That distinction between attribution and influence is where the conversation gets uncomfortable.

Digital advertising has always had a tendency to claim credit for demand it merely intercepted. A consumer may already have decided to buy before clicking the ad that sits closest to the transaction.

Conversational AI could make that problem more acute because the assistant itself may have shaped the consumer’s reasoning before the sponsored placement ever appeared.

Subhash Pais, Founder and CEO of SutrAi, describes the resulting problem as an information asymmetry.

“The platform could know why an impression was valuable, while the advertiser only knew that an impression happened and someone clicked,” he said.

“If that happens, advertisers are effectively being asked to trust a black box.”

For Pais, CPC and CPM remain useful buying currencies. They are not evidence that the media created value.

“They are buying currencies, not proof of value,” he said.

The harder distinction is between influence and attribution.

“If someone has already decided to buy and happens to click an AI ad, the platform shouldn't get disproportionate credit for a decision it didn't create.”

That is why holdout groups, conversion-lift studies and downstream revenue begin to matter more than another dashboard populated with impressions and clicks.

Tejas Maha, Associate Director - Media at White Rivers Media, reduces the structural problem to a single sentence: “One party then controls both the intent signal and the measurement.”

That problem is not entirely new. Digital advertising has spent years dealing with walled gardens in which platforms possess more audience information than advertisers or agencies.

Conversational AI potentially deepens the imbalance because the hidden information is not merely an audience profile. It may be the very context used to decide why the advertising opportunity exists.

Maha said he would consequently put less weight on CPC and CPM and more on intent cohorts, exposure quality and downstream consideration, supported by holdout and lift testing.

Vipin Yadav, Vice President and Head of Marketing at DriveX, is even more direct.

“If AI owns the intent and the measurement, advertisers need proof of incremental influence, not just platform-reported performance,” he said.

His test is simple: “Did the AI interaction change the consumer's decision, or merely capture an existing one?”

That question becomes even harder when the assistant itself is already recommending the brand.

Ambika Sharma, Founder and Chief Strategist at Pulp Strategy, argues that brands may eventually have to establish what the model says about them organically before attempting to measure the impact of paid placement.

“If the model already recommends you, the ad is buying demand you owned, not new demand,” she said.

Without that baseline, a conversion could be attributed to the ad even when the unpaid answer was already steering the consumer towards the same brand.

Her conclusion is less diplomatic than most. “I do not trust a platform grading its own paper.”

Prashant Puri, Co-Founder and CEO of AdLift, lands closer to the middle ground. He argues that conversational advertising will need a broader stack of proof: assisted conversions, qualified engagement, downstream conversion rates, cost per incremental customer, brand lift and controlled holdout studies.

Those measures all ask variations of the same question.

Did the media create something?

Or was it simply present when something happened?

That is where the promise of conversational advertising and its measurement problem collide.

The platform may know the user’s need, budget, objections, alternatives and proximity to purchase. It may know why an impression was served at that particular moment.

The advertiser may know only that it paid for one.

Search showed the query. Social showed the audience. Conversational AI may know the reasoning.

And until advertisers can independently establish how much that reasoning changed the outcome, the most valuable signal in the system may remain the one they are least able to see.