Commerce analytics can miss customers excluded by AI

A healthy conversion dashboard can hide customers who never arrived

A commerce team can see strong onsite conversion while losing prospective customers before its analytics have anything to measure. A shopper asks an AI assistant which product best fits their needs, the assistant recommends a competitor, and the shopper follows that recommendation. The brand sees no visit, abandoned cart, attribution record, or conversion event because the shopper never enters an owned property.

That missing visit changes what a healthy dashboard can reveal about a business’s competitive position. Conventional analytics work best once someone reaches a brand website or another brand-owned platform, because sessions and later actions generate observable events. Consumers removed from consideration by an AI recommendation create none of those events. Commerce AI therefore creates a measurement problem when prospective customers disappear before becoming incoming customers.

Commerce discovery is moving upstream of brand websites

The blind spot grows as product discovery moves away from brand-owned sites. Salesforce research shows a large shift in where digital commerce starts: 82% began on a brand website in 2014, compared with 38% in 2024. When most journeys started on those websites, analytics could observe much of the process from initial engagement onward. As that share falls, more consideration happens before systems controlled by brands start collecting events.

That earlier consideration increasingly includes consumers asking AI platforms where to shop, what to buy, or which product is right for them. An AI assistant can answer those questions before the consumer opens a retailer or brand site, so the assistant helps decide which products deserve further consideration. The people who arrive on a brand site therefore represent only the part of the category audience that made it through this earlier discovery process.

Zero-click behavior shows how discovery can finish before a site visit. Bain research finds that four in five consumers rely on zero-click results at least 40% of the time. A zero-click result provides enough information on the search or answer surface for someone to continue without following a link. The destination site then receives no visit for conventional analytics to capture.

Semrush’s 2025 zero-click study provides another measure of this broader behavior: 60% of searches end without a click. That figure concerns searches generally, so it does not directly measure AI behavior, but it shows an environment where receiving an answer no longer guarantees a visit to the companies involved in that answer. The old assumption that consideration will eventually become observable traffic is becoming less reliable.

Alongside journeys that end without a click, some AI interactions generate substantial traffic. Adobe Analytics recorded more than 800% year-over-year growth in AI-driven traffic to retail sites, indicating that AI platforms are becoming a meaningful intermediary between consumers and retailers. That growth makes AI visible in acquisition reports because referred visitors create sessions. Those reports cover the people who ultimately click through.

AI-mediated discovery can therefore produce two different measurement outcomes. An AI platform can recommend a product and send the shopper to a retail or brand site, producing an observable arrival. It can also answer the shopper’s question well enough for that person to narrow or settle their options without visiting every brand the AI system considered. The second path creates the blind spot because traffic and conversion systems never see the excluded options.

AI can shape the shortlist without generating a measurable visit

That second path matters more when AI determines which brands enter the shortlist in the first place. Treating AI as another acquisition source starts measurement only when the platform sends traffic. An AI answer engine can instead create or narrow the products somebody considers before any engagement with a brand, and a zero-click interaction can make the answer itself the consumer’s destination.

Rezolve Ai commissioned research among 1,500 US consumers in January 2025 that illustrates how far this behavior can go. The research found that a majority of shoppers who use AI for product research make purchase decisions directly from AI-generated recommendations without returning to a search engine or brand site for verification. Rezolve Ai has a commercial interest in AI-powered commerce, so its commissioned research comes from a company that can benefit as businesses invest more heavily in this area. Its survey nevertheless records the specific behavior of shoppers making decisions from AI-generated recommendations.

That direct decision-making gives an AI-generated shortlist more influence over which brands receive a later visit. By the time one shopper reaches an owned property, the purchase decision may already have been made. Another shopper may have considered and rejected the brand through an AI interaction and consequently never reach the property at all. The difference between those shoppers determines whether the brand gets an event it can measure.

AI-mediated discovery can intensify the same zero-click mechanism seen in search because its zero-click behavior is described as “structurally higher.” When the answer completes the discovery task, an omitted brand receives no downstream opportunity to expose the shopper to its products. The effect occurs before site analytics begin, making omission itself difficult to detect from session data.

That omission introduces selection bias into the brand’s observed audience. If an AI answer excludes the brand, the shopper may never consider it and never create a session on its site. Existing instrumentation consequently has no corresponding “AI excluded you” event to record, even though the answer may have influenced which company receives the eventual purchase.

Selection bias then explains how strong conversion performance can coexist with competitive loss before the visit. Conversion analytics describe what happens among people who reached the measurable environment, while the earlier AI interaction helped determine who entered that environment. A rising conversion rate can remain a valid measure of onsite performance while revealing little about how often the brand appeared in AI-created consideration sets.

Because the missing event occurs upstream, ordinary attribution cannot close the gap. A company can add AI as a traffic-source category and learn more about visitors referred by AI assistants or answer engines, improving its understanding of customers who arrive. A consumer who receives a competitor recommendation and never visits creates no attribution record for the brand to inspect.

The resulting distinction is between attribution and observability. Attribution assigns credit for a visitor or outcome that has already been observed, while observability means being able to detect the relevant event or state in the first place. When an AI omission produces no visit, the conventional session record contains nothing to attribute. More detailed downstream attribution therefore cannot reconstruct that missing population.

AI discoverability is harder to audit than search visibility

Once the problem is framed as observability, search engine optimization provides a useful contrast. Search presence has traditionally produced an inspectable signal: a brand can check its ranking for a query, identify weak presence or absence, audit the gap, and decide what to change. That process gives teams evidence about competitive discovery before they evaluate conversions from resulting traffic.

AI answer engines make the equivalent audit harder because their answers do not inherently give a brand a record of responses from which it was omitted. A consumer can receive recommendations that exclude a particular company without generating a signal inside that company’s systems. The company then has no internal event identifying either that consumer or the answer that shaped the consumer’s consideration set.

That gap sits earlier than the part of the journey where commerce instrumentation is already sophisticated. Landing pages, sessions, carts, purchases, and related events can describe the observable path after arrival, while AI assistants can participate in product selection before that path starts. Companies that judge competitive performance principally through onsite and session-based analytics consequently leave this pre-visit population outside their field of measurement.

Measurement has to begin before the session

Because the missing information originates before arrival, measurement has to examine how AI systems represent the brand before traffic attribution begins. The practical questions are: “How does my brand appear when consumers ask AI for recommendations in my category? What language does AI use to describe my products? Where am I present, where am I absent, and where am I being described in ways that don’t reflect my positioning?” These questions turn pre-visit discovery into an area a commerce organization can deliberately inspect.

That inspection needs to capture several dimensions of AI representation together: whether the brand appears, where it is absent, the descriptive language attached to its products, and whether that description aligns with its intended positioning. Presence alone cannot show whether a recommendation characterizes the product in a way that changes who considers it suitable. Representation matters because AI participates in the consumer’s judgment about fit and can shape whether a later visit occurs.

Once representation becomes measurable, the infrastructure objective also changes. Analytics and commerce teams need visibility into how AI answer engines and assistants represent the brand across consumer discovery questions, followed by a way to track those representations over time and ultimately influence them. The goal is to observe the discovery layer early enough to see how consideration is forming before a measurable session exists.

That upstream view complements better identification of AI-referred sessions. Referral measurement remains useful because it explains traffic that reaches the brand, while AI-discoverability measurement examines the answers that determine whether a visit happens. Together they separate two different questions: what happened after AI sent a shopper to the brand, and how AI shaped the shopper’s options before any referral existed.

AI-discoverability measurement is still emerging

Measuring upstream representation establishes presence, absence, and positioning, but commercial impact is a separate measurement problem. The tools for examining AI discoverability are still emerging, and measurement frameworks have yet to become standardized. A business can track how frequently AI systems surface the brand and how those systems describe its products, then use those observations as evidence about the discovery process rather than automatically treating each omission as a measured revenue loss.

That boundary gives teams a practical way to build measurement without assigning unsupported financial meaning to every observation. Repeated audits can show how AI representation changes over time and where a brand enters or disappears from category answers. Connecting those observations to commercial outcomes requires a model that links upstream representation with later consumer behavior. Until that link is measured, AI discoverability and downstream revenue remain distinct quantities.

Key highlights

• Conversion can hide lost demand: Strong onsite conversion measures shoppers who arrived. AI recommendations can exclude a brand before a visit occurs, leaving lost prospects outside conventional analytics.
• Discovery is moving upstream: More product discovery now happens beyond brand-owned websites, including AI assistants and zero-click experiences. Commerce teams need visibility into these earlier interactions to understand how consideration forms.
• AI shapes the shortlist before attribution begins: AI recommendations can determine which brands consumers consider without generating a session. Adding AI referral categories captures arriving traffic but leaves AI-driven exclusion unobserved.
• AI discoverability requires its own audit: Search rankings provide an inspectable signal of visibility, while AI omissions rarely create records in brand systems. Commerce teams need systematic monitoring of when and how AI systems surface their products.
• Measurement needs to start before the session: Analytics teams can track AI presence, absence, product descriptions, and positioning across relevant consumer questions, then monitor those signals over time alongside AI-referred traffic.
• Commercial impact requires separate validation: AI-discoverability measurement can establish how brands appear across AI systems, but it does not by itself quantify revenue impact. Organizations need evidence linking changes in upstream representation to downstream consumer behavior before assigning financial value.