As of November 7, 2026, the holiday readiness question for online stores has a number attached: 65 out of 100. Blend Commerce sent a self-identifying shopping agent to 200 retailers in the UK and US between September 8 and 15, scored each one on whether an agent could reach the site, find products, read the product data and push a basket to checkout, and published the results on October 6, three weeks before the peak week begins. The average says less about technology than about housekeeping: most stores are reachable, fewer are readable, and very few are transactable.
What the audit measured and what it found
The four layers explain the spread better than any single score. Access is mostly a security configuration question, and it is the layer where the gap between policy and practice is widest: a marketing team can approve AI crawlers in a document while a bot-management rule blocks a declared agent at the edge. Discovery is cheaper to fix than most teams assume, since a product sitemap either exists or does not. Comprehension is where most catalogs lose points, and transaction is where platforms, not merchants, have already decided the outcome.
Doors shut against a declared agent
The refusal pattern is the most quoted finding, and also the most misread. Twenty-nine of 200 stores turned the agent away twice, a rate of one in seven, and large retailers refused at roughly 20% against 2.5% for emerging ones. The agent identified itself and declared its purpose, so this is not a case of stores defending against anonymous scraping. For the 24 refusers whose crawler rules could be read, every one of them publicly welcomes the major AI crawlers. The block sits between the published policy and the request.
The data behind the door
Where the agent did get in, product data was the binding constraint. Structured product schema appeared on 82% of the 171 readable stores and a price on 76%, which sounds healthy until the fields that decide a comparison are counted: shipping details on 22%, a returns policy on 20%, and a GTIN on 18%. Without an identifier, an agent comparing two listings has to guess whether they describe the same item, and guessing is not a behaviour that survives a price comparison. Delivery and returns terms are what a shopper asks about in the last minute before paying, and they are the fields least likely to be machine-readable.
Where structured data breaks
The platform split makes the mechanism visible. Thirty-eight of the 200 stores published a live agent checkout endpoint, and 37 of those run on Shopify, where the capability arrives as a default rather than a project. Those 37 stores averaged 91 against 59 for everyone else, and the gap barely narrows when the endpoint checks are removed from the scoring. Shopify stores were not complete on product data either: 89% included price, 92% availability, but only 32% a GTIN and 27% a returns policy. Defaults cover name, brand, price and availability; everything a shopper needs in the final mile is added by a theme or an app, or not at all.
What the gaps cost before the peak
The commercial context has moved faster than the plumbing. Google began rolling out direct checkout through its Universal Commerce Protocol to Canadian shoppers on November 4, with the merchant remaining seller of record, and 59% of Canadian holiday shoppers say they plan to use an AI tool this season. Salesforce projects that AI agents will generate 20% of holiday ecommerce traffic in 2026. Consumer demand is shifting toward surfaces that read structured product data, and the stores that publish it are the ones that can be compared at all.
| Sector | Average score | What the number reflects |
|---|---|---|
| Baby and nursery | 88 | Dense, uniform product data and clear delivery terms |
| Furniture and beds | 81 | Large-ticket items force shipping and returns into the open |
| Beauty | 69 | Rich imagery, thin structured fields |
| Electronics | 55 | Specifications exist but rarely in comparable form |
| Fashion | 49 | Size and variant logic defeats flat product schema |
Sector averages describe ten retailers each, not a market, and the report says so. Still, the ordering is instructive: the categories that score highest are the ones where a single wrong answer is expensive for the buyer, so delivery and returns were forced into structured form years ago. Fashion, which lives on variants and seasonal rotation, scores lowest precisely because its data model is the hardest to flatten.
Readiness also splits inside sectors. An emerging UK fashion retailer scored 90 while a large US toy retailer scored 94, which weakens any argument that scale alone buys readiness. What separates them is whether someone owns the fields: in the companion survey of 59 ecommerce leaders, 63% said an agent cannot currently complete their checkout, 61% had no 2026 budget for agent optimisation and 57% had nobody internally who owned AI readiness.
A checklist for the next twenty days
Three actions fit before the peak week without a platform migration. First, test the door: send a declared request to your own store from a clean client and confirm that bot management does not answer for you. Second, audit five product pages per category against the fields an agent needs, starting with GTIN, shipping and returns, and compare the schema with what the page renders. Third, confirm the guest checkout path and express payment on the product page, because those are the two steps where a basket stops being a sale.
None of this requires a new budget line, which matters given how few teams have one. It requires an owner and a list of fields. The audit's most useful lesson is that the difference between a 49 and an 88 is rarely ambition; it is whether the boring fields were ever filled in.
FAQ
Does a high score guarantee sales from AI assistants?
No. The score measures whether an agent can reach, read and transact. Whether it recommends you still depends on price, availability and how your product data compares with rivals on the same shelf.
Why does bot protection block agents that robots.txt allows?
Because the two controls are owned by different teams. Crawler rules are a marketing and SEO decision, while bot management is a security rule set, and a declared agent can trip the second one without anyone noticing.
Is a GTIN really necessary for a small catalog?
It is what lets an agent decide that two listings are the same product. Without it, price comparison becomes a guess, and a guessing agent will skip your listing rather than risk a mismatch.
What should a store without a checkout endpoint do first?
Fix the product data. Access and discovery are binary and cheap; comprehension is the layer where most catalogs lose points and where the fix stays under your control.
