AI discovery affects more than whether an ecommerce brand gets mentioned. It can change which product enters consideration, which merchant receives the purchase, how much margin the brand keeps, and whether the customer enters a relationship the brand can reach again without paying for another acquisition.

That is why Raoul’s original argument still holds: AI discovery eventually becomes a P&L question.

The important update for ecommerce is that an AI “win” can produce very different economics depending on what happens after the product is recommended.

Consider a $100 order.

The product wins. The P&L question starts after that.

A shopper asks an AI assistant for a lightweight carry-on under $120.

Your product appears.

At first glance, the result is unambiguously positive. The brand entered consideration and the SKU beat competing products for a commercially relevant question.

But the shopper still needs somewhere to buy it.

ChatGPT, for example, can present multiple merchants for the same selected product. OpenAI says those merchants may be ordered using factors including price, availability, merchant quality and whether the seller is the maker or primary seller.

The AI system has therefore made two economically different decisions.

It has helped choose what to buy.

It can also influence where to buy it.

That second decision is where a visibility report can start diverging from the P&L.

Suppose the shopper buys the $100 carry-on directly from the brand.

For a simplified illustration, assume:

Now take the same product recommendation, but the shopper chooses a marketplace where the brand is the seller.

Assume the same $100 selling price, $35 product cost, $10 fulfillment cost and $15 of marketplace selling costs.

These numbers are deliberately illustrative. They are not the fee schedule of any particular marketplace.

The point is the $14 difference.

The AI visibility result has not changed. The same SKU won.

The commercial result has.

Raoul’s P&L argument begins there: recommendation share is not revenue quality.

Marketplace leakage does not mean marketplace revenue is bad

Calling every marketplace transaction “lost revenue” would be wrong.

For many brands, Amazon, specialist retailers and other marketplaces are deliberate distribution channels. They can provide reach, convenience, fulfillment and customer trust that the brand would struggle to recreate economically on its own.

A product being recommended and purchased through one of those channels may be an excellent outcome.

The useful concept is marketplace leakage relative to the channel outcome the business wanted.

If the company deliberately wants marketplace growth, there may be no leakage at all.

If its strategy depends on expanding DTC contribution margin and first-party customer relationships, then a growing proportion of AI recommendations terminating elsewhere deserves attention.

The measurement question therefore becomes more precise:

When our product wins an AI-shopping decision, what proportion of the resulting merchant opportunities goes to Brand.com, marketplaces and retailers?

That is merchant routing.

It belongs beside product recommendation because the two measurements can move in opposite directions.

A brand could increase recommendation share for three consecutive months while seeing fewer of those product wins route toward its highest-contribution channel.

A visibility dashboard would celebrate.

A finance team might ask a different question.

The customer relationship changes the economics again

There is another difference between those two $100 orders that does not appear in the first transaction.

The DTC customer may give the brand a direct customer relationship: an account, an email address, purchase history, consented marketing access or another first-party connection depending on the checkout and applicable rules.

A marketplace transaction often gives the manufacturer much less control over that relationship.

Now imagine the shopper needs another travel product six months later.

If the brand can bring an existing DTC customer back through email, an app, loyalty or direct navigation, the incremental acquisition economics can be very different from having to win that customer again through paid media.

This is where AI discovery and CAC become connected—but the relationship needs to be stated carefully.

AI visibility does not automatically lower CAC.

CAC is still an acquisition equation:

customer acquisition cost = acquisition spend ÷ new customers acquired

What can change is the path by which customers enter the business.

If AI increasingly sends qualified new shoppers directly to the brand and those shoppers convert without equivalent paid-media spend, paid dependency can fall.

If AI recommends the product but routes the initial transaction elsewhere, the brand may gain the sale without gaining the same direct customer relationship.

And if the brand later pays Google, Meta or another channel to acquire that person into DTC, part of the economic opportunity created upstream has been captured by someone else first.

That is not something recommendation share alone can show.

Think about the second order, not only the first one

Return to our illustrative customer.

The first $100 purchase happens through a marketplace.

Several months later the customer sees a paid ad from the brand and makes a $100 purchase directly.

Assume the second DTC order has the same $54 contribution before acquisition cost, and the brand spends $24 in attributable paid acquisition to generate that order.

Contribution after that acquisition cost becomes $30.

Had the first AI-mediated purchase established a healthy direct relationship, the path to the second purchase might have been different.

It might still have involved paid media. It might have come through email, branded search, loyalty, direct navigation or not happened at all.

We cannot simply claim that the brand “lost $24” because AI routed the original purchase to a marketplace. That would require a counterfactual we do not know.

What the business can say is that channel routing changes the economic conditions under which retention and reacquisition happen.

That is measurable.

Compare cohorts.

How often do AI-referred DTC customers return?

What does it cost to acquire or reacquire them?

How does their contribution margin compare with customers whose first observed order arrives through other channels?

Then the P&L argument moves from theory into company-specific evidence.

This is why AI discovery should not sit inside one marketing metric

Google’s own commerce reporting is starting to acknowledge that shopping decisions happen before the final click. Merchant Center’s AI performance insights separates conversational shopping activity into Discovery, Evaluation and Ready-to-buy stages and reports metrics including share of voice and products showing. The current report covers organic AI activity rather than paid ads.

That creates useful upstream visibility.

But a Head of Ecommerce still needs to connect it to downstream economics.

Suppose recommendation share improves.

That answers: Are we earning more consideration?

Then merchant routing changes.

That answers: Who gets the commercial path after our product wins?

Then analytics records direct sessions, orders and revenue.

Finance can finally ask: What contribution did those transactions produce, and what did customer acquisition cost us?

Those numbers should not be collapsed into one synthetic score because they describe different parts of the business.

The value comes from following the same economic journey far enough.

Paid media looks different once AI becomes a discovery layer

For years, ecommerce brands have been able to buy access to demand through paid search, Shopping ads, social platforms and marketplaces.

AI discovery introduces another route into product consideration.

Some of that activity is currently organic. Google explicitly states that its Merchant Center AI performance report is limited to organic AI traffic and excludes paid-ad traffic. OpenAI likewise says its current ChatGPT product results are selected independently rather than being influenced by commercial partnerships, with ads treated separately.

For a growth leader, that creates an interesting economic question.

If a category begins generating meaningful product discovery through AI, does the brand eventually acquire some customers who would otherwise have required paid media?

Possibly.

But that should be demonstrated in the numbers rather than assumed.

A brand should look for movement in its actual channel mix: AI-referred customers, paid spend, new-customer volume, CAC, conversion, merchant routing and contribution.

If organic AI visibility rises while paid acquisition remains identical and DTC customer volume does not move, claiming CAC improvement would be premature.

If AI-originated direct demand grows while equivalent new-customer volume is maintained with less incremental paid spend, the economics become much more interesting.

What should Raoul’s P&L view actually contain?

Not another marketing dashboard.

The useful view connects enough of the journey to explain where economic value moved.

For an important product category, leadership should be able to see whether products are entering AI consideration, where those products are being routed, what observable traffic reaches the owned store, what those sessions convert into, and how the resulting customers compare economically with other acquisition cohorts.

The critical connection is:

AI consideration → product → merchant → customer → contribution

Marketing owns pieces of that chain.

So do ecommerce, marketplace, merchandising, analytics and finance.

That is precisely why the problem becomes bigger than marketing.

A product recommendation is valuable only in context

The mistake is not measuring AI visibility.

The mistake is stopping there.

An AI recommendation may generate a high-margin DTC customer.

It may generate a profitable marketplace sale.

It may generate awareness that later converts through paid search.

Or it may create a product win from which the brand captures surprisingly little economic value.

Those outcomes should not be valued equally merely because the product appeared in the answer.

Raoul’s original P&L argument becomes much sharper in ecommerce once the transaction is included.

The relevant question is no longer only:

“Are we visible in AI?”

It is:

“When AI helps create demand for our products, how much of that value does the business ultimately capture?”

Gravton’s Measurement layer connects the upstream AI-shopping decision with recommendation, merchant routing and downstream outcomes so ecommerce teams can investigate that question with commerce data rather than visibility alone.

Follow the recommendation far enough to see the economics.