AI is shaping the shortlist.
Checkout is catching up.
AI referrals are still a small channel, but their conversion signals deserve attention. The practical work starts with product information that a buyer—and an AI assistant—can trust.
The numbers worth reading together
- 0.2% of visits came from AI referrals in Contentsquare’s cross-industry benchmark, drawing on 99 billion sessions across 6,500 sites. Contentsquare’s benchmark.
- Nearly 50% higher conversion for AI-referred visitors than organic-search visitors in Shopify’s Q1 2026 comparison of sessions beginning on product detail pages. Shopify’s analysis.
- Nearly 13× more AI-referred orders year over year on Shopify in Q1 2026. Organic search still sent more sessions than all tracked AI platforms combined. Shopify’s channel data.
These are different datasets, populations and measures. The 0.2% figure is not a Shopify market-share estimate; the conversion comparison is not a promise for every store. None of these figures, on its own, measures fully autonomous purchases.
Growth without share can exaggerate a channel. Share without growth can make you overlook it. For a commerce team, the useful question is whether AI-referred visits are becoming commercially meaningful in its own category.
Checkout changed direction. It did not disappear.
In March, OpenAI said the initial version of Instant Checkout lacked the flexibility it wanted. It shifted its focus toward product discovery while allowing merchants to use their own checkout experiences. Its announcement also described a Walmart app supporting account linking, loyalty and payments. OpenAI’s update.
By June, Target said shoppers could buy directly in Google Search and Gemini through Universal Commerce Protocol (UCP). It also described checkout in Copilot and a Target app in ChatGPT. These are real purchasing experiences, with the customer controlling the decision. Target’s overview.
That makes “agents aren’t buying yet” too broad. Discovery and checkout are developing unevenly across platforms. Loyalty, delivery, payment and order handling still have to work together; a persuasive recommendation is only one part of that job.
A product can win a place on the shortlist before the checkout experience is fully settled.
Why it matters for commerce: the shortlist needs usable facts
Imagine a shopper asking for a cordless drill under two pounds, delivered within two days, with free returns. A polished description helps, but weight, delivery conditions and return terms make the comparison possible. This is an illustrative buying task, not a claim about one universal ranking algorithm.
Product feeds matter, and so do the pages behind them. Profound’s June study found that about 88% of product-offer instances in its month-long ChatGPT sample came from web product pages. Its separate eight-month dataset covered roughly 548 million offers. Those two samples should not be treated as the same denominator. Profound’s study and methodology.
The study is observational: it does not prove that adding a field causes a higher ranking. It does show why maintaining product pages remains relevant alongside feed integrations.
Missing product information makes a fit harder to verify. It is not evidence of an automatic exclusion rule.
What to check this week
- Audit one important product range. Compare the product page, catalog and channel feed: identifiers, variants, price, stock, weight, dimensions, delivery promises and return conditions. Fix contradictions before expanding the catalog.
- Use the integration your platform actually supports. OpenAI says Shopify Catalog already supplies merchant product data to ChatGPT. Other merchants should check the available feed or partner route and its requirements. Check OpenAI’s merchant update.
- Treat a protocol as an integration project. UCP’s
/.well-known/ucpprofile describes supported services and capabilities. Publishing that file alone does not create a working catalog, checkout or fulfillment process. Confirm what your provider implements. Read the UCP specification. - Track outcomes in your own analytics. Compare AI referrals with other channels using sessions, orders, conversion and revenue per session. Keep the sample size visible. Referral reports capture attributable clicks, not every AI-assisted decision.
WIZIUM perspective
Our reading: product data is operating work
The useful preparation starts before a marketing team writes another description. When a specification changes, who updates the listing? When stock moves, who checks the delivery promise? When two sources disagree, who decides which one is current?
Those are responsibilities inside the business. A feed can distribute a mistake as efficiently as a fact.
At WIZIUM, we are building toward an operating environment that keeps tasks, evidence, decisions and their owners together. Our view is that AI readiness depends on the quality of that shared work—not just on how quickly a team can generate content.
Your side of this story
How is AI changing the work behind your storefront?
WIZIUM Research Lab is speaking with agency leaders, founders and commerce operators about one real work route: where AI helps, where information gets lost and what still needs human judgment.
The conversation takes about 50 minutes. Leave your work contact details and professional profile; we’ll get in touch to arrange a convenient time. We plan to share the findings on the Lab. Any quote published with your name will be agreed with you first.
Take part in an interview