A shopper asks ChatGPT for a weekend bag under $150. The model names three options, and the shopper picks one and completes the purchase without leaving the chat, using an instant checkout integration wired straight into the merchant’s cart. The merchant’s analytics show nothing: no referrer, no session, no page view. That gap is the tracking problem AI shopping agents create, and it is a different problem than tracking whether a brand got recommended in the first place.
Key Takeaways
- A recommendation citation and a completed in-chat purchase are separate events that happen in separate systems, and treating them as one funnel step loses the data that matters most.
- Instant checkout integrations do not pass a referrer or session, so last-click attribution models cannot see the order without a new tagging layer from the merchant side.
- A usable tracking setup needs three distinct conversion definitions: being named as a source, being clicked through to, and completing a transaction inside the agent.
When Checkout Happens Inside the Chat, What Are We Actually Measuring?
Before any purchase happens, there is a prior event: the model deciding to name a brand at all. That is what Answerburst tracks as Share of Citation, how often a model pulls a brand into an answer and cites it as a source. It is distinct from Share of Voice, which just counts mentions across the web. A brand can rank well in general conversation about a product category and still get zero citations when someone actually asks which one to buy.
Instant checkout changes what happens after that citation. The shopper no longer clicks through to a product page, browses, and converts on the merchant’s own site where GA4 or a similar tool can see the session. The transaction closes inside the chat, through an API call between the model provider and the merchant’s commerce platform. Two different systems, two different data trails, and until recently, no shared record between them.
Two Different Jobs: Getting Cited vs Getting Paid
Getting cited is a retrieval and trust problem. The model has to find a brand’s product data, structured correctly, current enough to trust, and corroborated by other sources it already treats as authoritative. That is the work behind structured data, entity optimization, and citation building: convincing the model your brand is a safe answer to name.
Getting paid, once the model has named you, is a completely separate problem. It depends on whether your checkout integration exists at all, whether the price and inventory data the agent sees match what is actually in stock, and whether the payment flow completes without the shopper bouncing to verify something the agent could not confirm. A brand can win the citation and still lose the sale at checkout for reasons that have nothing to do with AEO.
A citation is proof the model trusts you enough to name you. A completed checkout is proof the shopper trusted the answer enough to pay.
What Counts as a Conversion at Each Stage
Collapsing this into a single “AI conversion” metric hides where a brand is actually losing ground. There are three distinct stages, and each needs its own definition:
- Stage one, citation. The model names the brand and attributes the recommendation to it, with or without a visible source link. This is measured by tracking answer share and citation frequency across the engines a brand’s buyers actually use, not by web analytics.
- Stage two, click-through. Applies only when the agent sends the shopper out to a product page rather than completing the transaction natively. This is the one stage that still resembles traditional attribution, since it produces a session and a referrer.
- Stage three, completed transaction. The order closes inside the agent, through an API handshake between the model provider and the merchant platform. No referrer, no click, no session. This has to be captured at the order level, tagged by the checkout integration itself, or it does not exist in the data at all.
Reporting only on stage one tells a brand it is winning while revenue tells a different story. Reporting only on stage three tells a brand it is losing sales without explaining whether the problem is visibility, trust, or the checkout flow itself.
total AI citations tracked for Moburst’s own brand, with an average position of 1.77 when cited across engines.See the case study
Why the Old Attribution Chain Breaks Down
Last-click attribution assumes a browser session: a referrer header, a UTM parameter, a landing page a script can fire from. Instant checkout removes all three. The order arrives at the merchant’s system as an API call from the model provider, and unless that call carries a source flag, the order looks identical to any other transaction that came from nowhere in particular.
This is where revenue attribution has to move upstream, into the checkout integration itself rather than the analytics layer bolted on afterward. A merchant needs its checkout partner to pass a parameter identifying the order as agent-originated, ideally naming which model surfaced it. Without that, a brand can watch citation volume climb for months with no way to connect it to a single dollar of revenue, and no way to tell finance which pillar of the work is paying for itself.
What to Do Next
Start by asking your checkout or commerce platform provider one direct question: does the instant checkout integration pass any source or agent metadata with the order, and can that field be added to your order database as a custom parameter. If the answer is no, that is the first gap to close, before any amount of citation optimization can be connected to revenue.
Separately, set up citation tracking as its own dashboard, not a subsection of existing SEO reporting. It answers a different question: whether the model names the brand at all when a buyer asks. Keep that number next to, not folded into, whatever order-level attribution data you build. The two numbers moving in opposite directions is itself the finding, and it tells you whether the problem sits in trust with the model or in the mechanics of the checkout flow.
FAQs
Is a Product Recommendation Citation the Same as a Conversion?
No. A citation means the model named the brand as a source when answering a question. A conversion means a transaction closed, either after a click-through to the brand’s site or inside the agent through instant checkout. A brand can be cited frequently with no purchases following, and the two need separate metrics.
Why Does Instant Checkout Break Standard Analytics Tools?
Standard analytics tools rely on a browser session with a referrer and a landing page to attribute a sale. Instant checkout completes the transaction through an API call between the model provider and the merchant’s commerce platform, with no session or referrer generated, so tools built around last-click tracking never see the order.
What Data Should a Merchant Ask a Checkout Integration Partner For?
An order-level parameter identifying the transaction as agent-originated, and ideally which model or platform surfaced the recommendation. Without that field, agent-driven orders are indistinguishable from any other transaction in the order database.
Should Citation Tracking Replace Traditional Conversion Tracking?
No. They measure different stages of the same funnel. Citation tracking shows whether a model trusts a brand enough to name it. Conversion tracking, once it can see agent-originated orders, shows whether that trust turns into revenue. A tracking setup needs both, reported separately.
