What 80,000 store audits reveal about the machine-readable layer that AI assistants and AI search actually read, and how little of it most stores have built. For store owners and operators who want to stay findable as AI reshapes how people shop.
When a shopper lands on your product page, they read the design: the photos, the layout, the trust badges. When an AI assistant or an AI-powered search result reads the same page, it ignores almost all of that. It looks for a machine-readable layer underneath, and across the 80,000 stores we audited, most of that layer is missing.
This is not a how-to. It is a measurement of how ready real stores are for the way buyers increasingly find products, and the short version is: not very, for reasons their owners mostly can’t see.
The headline findings
- AI assistants and AI search read a store’s machine-readable layer, not its design, and across 80,000 stores most of that layer is missing.
- The clearest example: structured return-policy data is missing on as few as 2.7% of stores on the newest platform and as many as 66% on the oldest. It is the second most common failure of any kind in the benchmark, absent on 33,148 stores in total.
- 25,683 stores give a language model nothing clean to summarize, and 25,515 hide their shipping terms from machines.
- AI-readiness tracks platform age more than platform effort: modern platforms ship the markup by default, older frameworks leave it to the merchant, and most merchants never add it.
- The gap is mechanical, not a matter of product or design quality. Stores are missing from AI answers for reasons their owners didn’t choose and can’t see.
How AI assistants actually read an ecommerce store
An AI assistant answering a shopping question, or an AI-written search result, works from two machine-readable layers, and neither is the part a human looks at.
The first is structured data, the schema markup that states the facts in a format a machine can lift directly: price and availability, the return terms (the MerchantReturnPolicy type), the shipping terms (OfferShippingDetails). The second is extractable structure, whether the page uses semantic headings and clean content blocks that let a language model separate the actual product description from template noise like navigation and footers, and turn the page into something it can summarize. A page can be flawless to the human eye and close to blank to both.
The consequence is competitive, not cosmetic. If a machine cannot reliably read your return policy, your shipping terms, your availability, or a clear summary of the page, it has less to cite about you than a competitor that exposes those details, so it cites the competitor instead. The benchmark measured both layers across all 80,000 stores, and that is the lens for everything below.
The facts AI can’t find, and the pages it can’t read
Across the whole sample, the machine-readable layer is mostly missing.
Return-policy schema, the structured statement of how returns work, is absent on 33,148 stores. That is more than 40% of the sample and the second most common failure of any kind in the entire benchmark. Structured shipping data is missing on 25,515 stores. And 25,683 stores give a language model nothing clean to extract, with no clear summary or hierarchy it can turn into an answer.
The effect is quiet and easy to miss. When an assistant is asked “can I return this” or “how fast does it ship,” a store with none of this isn’t wrong in the answer. It is simply absent from it, while a competitor that did build the layer gets named instead.
AI-readiness is mostly a story about platform age
Sort the data by platform and the pattern is almost entirely about when the platform was built. Return-policy schema is the clearest case.
| Platform | Stores missing return-policy schema |
|---|---|
| Shopware 6 | 2.7% |
| Shopify | 35.9% |
| Magento 2 | 61% |
| PrestaShop | 66% |
Share of live stores in the benchmark sample missing return-policy (MerchantReturnPolicy) markup.
A store on PrestaShop is roughly 24 times more likely to be missing this markup than a store on Shopware. That is not a sign that PrestaShop is worse, it is a sign that the platform does not add the markup for you and most merchants never do. Shopware, the newest of the four, ships it by default and misses it on fewer than 3% of its stores. The content layer skews a little differently: Shopify, modern but heavily template-driven, fails the language-model summary check on 8,191 stores, slightly more than Magento’s 7,821, a reminder that a polished template does not guarantee extractable structure.
The through-line is that modern platforms inherit semantic maturity and older ones have to add it by hand. It is the same correlation the wider benchmark found between speed and AI-readiness: the platforms built for modern performance tend to ship modern semantics in the same package.
The scores, with one heavy caveat
As category health, AI-readiness lands like this. The figure is the share of stores with no detected issue, so higher is cleaner. <figure> <img src=”/guides/img/benchmark-ai-readiness.png” alt=”AI-readiness clean-rate by platform: Shopware 88.4%, Shopify 84.1%, Magento 79.7%, PrestaShop 76.8%.”> <figcaption>BelVG benchmark of 80,000 stores, scanned March–April 2026.</figcaption> </figure>
| Platform | AI-readiness |
|---|---|
| Shopware 6 | 88.44% |
| Shopify | 84.06% |
| Magento 2 | 79.73% |
| PrestaShop | 76.77% |
Read it as baseline health, not platform quality, and with the caveat that matters most in this whole article: Shopware’s lead is mostly recency. It was built in an era when JSON-LD schema was already a standard expectation, so it inherits modern semantics that the older frameworks are perfectly capable of and simply were not shipping when they were designed. The score measures defaults and age at least as much as it measures the platform.
What it means for your store
AI search and AI shopping assistants are a distribution channel, and like every channel it has an entry requirement. This one is the machine-readable layer, and most stores have not built it. That is the genuinely useful finding: the reason a store is missing from AI answers is rarely the product or the design. It is a layer of structured data and page structure that was never added, on a platform that never added it for you.
It is also mechanical work, well understood, and largely the same on every platform. We cover the how in the AI search optimization guide and the structured data guide; this article is about how common the gap is, not how to close it. The full methodology sits in the 2026 platform benchmark, alongside the most common problems we found across the same 80,000 stores.
This benchmark is the same audit we run on individual stores, so the scan behind these numbers is the one you can point at your own. The one thing the data cannot tell you is which side of the line your store is on.