Recommendation Eligibility

Executive summary

For twenty years, the goal was visibility: rank, get found, win the click. In AI-mediated shopping, visibility is necessary but no longer sufficient. The new currency is recommendation eligibility, whether an AI assistant can confidently find, understand, trust, and recommend your product when a shopper asks. Eligibility is earned in your product data, not your media plan, and most catalogs were never built for it.

The stakes are concrete. AI assistants are already sending retailers high-intent, high-converting traffic. AI referrals converted 31% more than other traffic sources over the 2025 holiday season (Adobe, via Digital Commerce 360). But a CatalogSignal CEI Benchmark of 100 brands across 10 verticals and four AI providers (more than 100,000 AI shopping queries) found uneven readiness: mean funnel accuracy was 47.7%, mean hallucination was 9.3%, and mean commercial harm was 10.6%. Some high-familiarity brands still fell into hallucination-risk positions. Eligibility, not familiarity, is what converts.

This guide defines eligibility, breaks it into five plain-English dimensions an assistant evaluates, shows what good and poor look like, and lays out how to measure and close the gap.

1. Visibility vs. eligibility

A human shopper who lands on your site can forgive thin data. They will click, scroll, infer, and figure it out. An AI assistant is less forgiving. It has to decide, in one turn, whether your product satisfies the shopper's request, and it is more likely to stake a recommendation on what it can confirm. If your listing does not clearly establish that the boot is waterproof, available in a wide fit, and under $150, the assistant is unlikely to guess in your favor. It tends to recommend the competitor whose data does confirm it, and when the data is weak it can misread or omit you entirely.

That is the difference between visibility and eligibility:

• Visibility: the assistant can see your product exists.
• Eligibility: the assistant can confirm your product fits, can trust the claims, and can therefore recommend it.

Adobe's own read of the surge is that AI traffic is growing faster than retail sites are becoming readable by machines, with product pages averaging just 66% readability in its AI visibility benchmark (Adobe, April 2026). The demand is here. Eligibility is the bottleneck.

A fair question at this point is whether eligibility is simply the next version of SEO. It is not. SEO is a content and authority discipline aimed at ranking; eligibility is a data discipline aimed at being recommended. They share hygiene, crawlability, structured markup, clean information, but they answer different questions and usually sit with different teams. SEO works the marketing site; eligibility works the catalog, where the assistant actually looks. Done well, one feeds the other, and neither substitutes for it.

2. The five questions behind eligibility

Eligibility is not a black box. It resolves into five questions an assistant implicitly asks of your catalog. Think of them as the five conditions a product must meet to make the shortlist.

1. Foundation: can it read the catalog? Is the product data complete and consistently named? A listing with a clear title, price, materials, sizes, and structured attributes is legible; one with a title and a price is a guess. Inconsistent vocabulary ("galvanized" vs. "galv." vs. "zinc-coated") quietly tells the machine these are different things.

2. Differentiation: can it tell products apart? If a category of products shares near-identical, boilerplate descriptions, an assistant has no basis to explain why one fits a shopper's need better than another. Differentiation is what lets the machine make a recommendation instead of a coin flip.

3. Retrieval: can it find the right product for a real query? Shoppers ask in constraints: "wireless, noise-canceling, under $200, good for travel." Retrieval is whether the right product surfaces and the constraint is honored. A retriever that returns a $260 pair for an "under $200" query has failed the shopper, and the brand.

4. Integrity: can it trust the claims? Assistants increasingly avoid asserting claims they cannot stand behind, because shoppers punish errors and verify what they are told. Yext's 2026 consumer-search research found that more than 90% of high-trust AI users take at least one verification step after a recommendation (Yext, 2026). Claims that are internally consistent and backed by evidence get repeated with confidence; claims that conflict get hedged or dropped.

5. Authority: does the wider internet vouch for you? Beyond your own catalog, assistants weigh reviews, mentions, and authoritative coverage. And shoppers still verify: after an AI recommendation, 62% search Google, 58% visit the business's own site, and 52% click the sources the AI cited (Yext, 2026). The story your catalog tells must survive that second look.

3. What good and poor eligibility look like

The gap between an eligible and an ineligible catalog is rarely about price or product quality. It is about whether the data lets the machine say yes with confidence.

• Poor: a high-share category is missing material, fit, or compatibility attributes across many SKUs. The assistant cannot confirm constraint matches, so the products drop off shortlists they should win.
• Poor: multiple products carry copy-paste descriptions. The assistant cannot differentiate, so it defaults to whichever competitor reads as distinct.
• Good: structured attributes are complete and consistent; descriptions are specific and differentiated; claims reconcile with the rest of the catalog and with reviews. The assistant recommends confidently, and is right.

CatalogSignal's June 2026 CEI Benchmark (directional) suggests these gaps cluster by category and model. Among qualifying verticals, consumer electronics led AI Shopper Readiness at 0.747, fashion and apparel had the highest commercial-harm mean at 0.154, and sporting goods and outdoor had the highest hallucination mean at 0.157, while general merchandise showed the lowest readiness. Provider-level mean accuracy ranged from 12.5% to 53.2%. Eligibility is a data discipline, unevenly distributed even among sophisticated brands. (Directional panel metrics; 100-brand benchmark; harm taxonomy inferred; no individual brands reported.)

4. The business case

Eligibility compounds in two directions. On the upside, AI-sourced shoppers are demonstrably better customers: revenue per visit from AI referrals grew 254% year over year and those shoppers spent 45% more time on site than non-AI traffic over the 2025 holiday (Adobe, via Digital Commerce 360). On the downside, an ineligible catalog loses the sale silently. The shopper never lands on the site to be re-marketed, because the assistant already chose someone else.

And the channel is hardening into infrastructure. OpenAI and Stripe co-developed the Agentic Commerce Protocol and open-sourced it in 2025; Google, Shopify, and partners launched the Universal Commerce Protocol in January 2026 (Fast Company, 2026). As agents move from advising the shopper to acting for the shopper, the catalog that an agent can parse and trust becomes the catalog that gets bought.

5. How to measure and close the gap

Eligibility is measurable, which means it is manageable. A disciplined program looks like this:

1. Baseline it. Measure how AI actually reads your catalog today, across the five dimensions, with product-level evidence rather than opinion. This is CEI Diagnose™.
2. Prioritize the fixes. Not all gaps are equal. Sequence by readiness impact and effort; structured-data and terminology fixes are often the fastest wins.
3. Generate the artifacts. Turn findings into ready-to-apply fixes, from attribute files to schema to description briefs, so teams execute instead of re-investigating. This is CEI Activate™.
4. Protect it. Gate new and updated products against your own quality bar before they publish, so eligibility does not drift backward as the catalog changes. This is CEI Protect™, a pre-publish quality gate.

That closed loop, measure, fix, protect, is the operating model behind the Commerce Eligibility Index™ (CEI): a 0 to 100 score of whether AI systems can find, understand, trust, and recommend your products, with the evidence and fix queue to act on it.

Visibility got you found. Eligibility gets you recommended. In an AI shelf, only one of them ends in a sale.

Map your eligibility, pillar by pillar. A baseline CEI assessment turns eligibility into an executive scorecard, product-level evidence, and a prioritized fix queue. Request a CEI assessment at catalogsignal.com.

Sources