AI shoppers are not reading your blog

Most ecommerce AI search advice still assumes the funnel works the way it did in 2019: publish a buying guide, rank it, catch the researcher, walk them to the product page. Shopify's second-quarter data says half the AI-referred traffic never sees that path. It arrives already decided, on the product detail page, having done its comparison inside the assistant.

The growth rates are the part everyone will quote. AI-referred sessions to Shopify merchant storefronts grew 197% year over year in Q2 2026. Organic search sessions grew 12% over the same period, off a far larger base. Both numbers are true and neither is the interesting one. The interesting one is where those sessions landed and what the merchant had done to earn them.

The number that reframes ecommerce AI search

Fifty percent of AI-referred sessions went directly to a product detail page. Not a category page, not a guide, not the homepage. That is a different shape from general AI referral behavior, where the homepage absorbs a large share of arrivals. We have written before about how AI referral traffic concentrates on a small set of page types; in retail the concentration is sharper and it points one level deeper, at the individual SKU.

An assistant that sends a shopper to a product detail page has already performed the comparison step. It has decided this SKU answers the question, and it is handing off for the transaction. That is why the conversion numbers look the way they do: AI-referred shoppers converted roughly 80% better than organic shoppers overall. It is the same pattern we found when seven studies disagreed about the AI conversion multiple and the honest read was that AI traffic arrives later in the journey, so of course it converts better.

Structured catalog data is the conversion lever

The most actionable finding in the Shopify release is the comparison between two ways an engine can learn about your products. Merchants whose product data reached AI surfaces through structured Shopify Catalog feeds saw AI-referred shoppers convert at twice the rate of merchants whose product data was picked up from scraped or third-party sources.

That gap is not mysterious. A scraped product record is whatever a crawler could reconstruct from rendered HTML at some point in the past. It goes stale, it misses variants, it guesses at availability, and it frequently carries a price that is no longer correct. A structured feed carries current price, current stock, variant-level attributes, and canonical product identifiers. When an assistant recommends a product it cannot verify is in stock at the price it quoted, the shopper lands on a page that contradicts the answer and leaves.

This is worth saying plainly because it cuts against a lot of ecommerce GEO advice: the highest-return work here is operations, not editorial. A merchandising team that fixes feed freshness and variant structure will move AI-referred revenue faster than a content team publishing another comparison guide. Both matter. Only one of them is currently underfunded on most ecommerce programs.

Why spec-led categories move first

Shopify broke the conversion advantage out by category, and the spread is the tell. Spec-led categories saw AI-referred shoppers convert at roughly twice the rate of organic. Watches came in around 2.4x, necklaces around 2.3x, apparel at 1.6x. The more a purchase decision reduces to comparable attributes, the more an assistant can do the work, and the further down the funnel the shopper arrives.

Apparel lags because fit, feel, and taste do not reduce to a spec sheet. That does not make apparel exempt, it makes the qualifying attributes different: fabric composition, model measurements, fit notes, care instructions, and return terms are the attributes an assistant can reason over. Merchants who treat those as marketing copy rather than structured fields hand the answer to whoever formatted them properly.

“In a spec-led category, the engine is not choosing your brand. It is choosing the product record that answers the question completely. Formatting is the whole competition.”

What is missing from the Shopify disclosure

Be careful with these figures in a board deck, because Shopify published outcomes without publishing method. There is no merchant count, no transaction count, no statement of how AI-referred sessions were identified, and no confidence interval on any of it. Growth of 197% off a small base is arithmetically easy, and the company selling structured catalog feeds is the same company reporting that structured catalog feeds double conversion.

The selection effect is the honest objection and it does not change the recommendation, because feed hygiene is cheap. Even if only half of that 2x is causal, the payback on a week of catalog work is better than almost anything else on an ecommerce roadmap right now. The direction of the finding also agrees with what we see in retail-heavy AI answer surfaces consolidating toward large merchants, where the winners are the ones whose product data is machine-legible at scale.

Where the citation and the sale split apart

Retail is the first category where the two halves of AI visibility have visibly come apart, and it is worth being precise about what each half does. Getting cited is an editorial problem. The engine is answering a question like which running shoe suits a wide foot, and it pulls from whatever page explains the tradeoff credibly. That page is usually a guide, a comparison, or a review, and it is often not yours. Getting the sale is a data problem. Once the engine has decided which product answers the question, it needs a record it can trust enough to name, price, and link.

Those two jobs have different owners inside most retailers. Content sits with brand or acquisition marketing. The product feed sits with merchandising or ecommerce operations, and in a lot of organisations nobody has told that team their work is now a visibility asset. The result is a familiar failure: a well-optimized buying guide earns the mention, the assistant recommends a competitor's SKU because the competitor's record was complete, and the guide gets credited with nothing.

You can diagnose this in an afternoon. Take twenty product questions a buyer in your category would actually ask, run them across ChatGPT, Gemini, and Google AI Mode, and log two things separately: whether any of your pages were cited, and whether any of your specific products were named with a price. Brands routinely score well on the first and badly on the second, and the gap is almost always a feed problem rather than a content problem. Our generative engine optimization engagements start retail clients with exactly this split, because it tells you which budget to move before you spend anything.

The ecommerce AI search work order

Run this in the order below. It is deliberately front-loaded with operations work, because that is where the measured advantage sits, because it does not require a content hire, and because a clean feed keeps paying long after this quarter's growth rate stops being newsworthy. Four weeks is a realistic timeline for a single category with one engineer and one merchandiser.

Week 1  Audit the feed
        Top 500 SKUs by revenue. Check price, stock, variant coverage,
        GTIN/MPN/brand fill rate, and update frequency.
 
Week 2  Fix identifiers and variants
        Every variant its own record. Every record an identifier.
        Kill collapsed listings and duplicate SKUs.
 
Week 3  Enrich the attributes buyers ask about
        Materials, dimensions, compatibility, fit notes, warranty,
        return window. Structured fields, not paragraph copy.
 
Week 4  Instrument and baseline
        Segment AI referrals, split by landing page type,
        and record conversion for the fixed category vs a control.

Then, and only then, look at content. Comparison pages still earn the citation that puts you in the answer, and that has not changed: comparison content remains the format engines quote most. But in retail the citation and the conversion are increasingly two different assets. The comparison page earns the mention. The product record closes it. Most merchants have been funding the first and neglecting the second, which is exactly backwards for a channel where half the arrivals skip straight to the SKU.

If you want a starting benchmark, pick your highest-margin spec-led category, fix its feed, and measure AI-referred conversion against a category you leave alone. One quarter, two numbers, no vendor required. Source: Shopify Enterprise, Q2 2026 AI search category behavior, published August 11, 2026 (Shopify).

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.