When an AI agent evaluates a product page, it skips the design and reads the data layer underneath. If a store's Product schema is incomplete or absent, the agent cannot reliably quote a price, confirm availability, or recommend the product — so it defaults to a competitor whose data it can trust.
That much is well understood. What surprised us when we pulled the numbers from our latest batch of scans was how cleanly stores divide on the most important dimension: structured data is binary — you either have it, or you do not.
What we found
We ran our full AI-readiness pipeline on 8 e-commerce stores across fashion, beauty, footwear, accessories, and home goods. The readiness score has three components — structured data (40% weight), AI preference (40%), and crawlability (20%). The structured data score, which measures whether product pages expose a valid JSON-LD Product block, told the story.
| Group | Structured data score | Count |
|---|---|---|
| Top performers | 80–91 / 100 | 4 stores |
| Zero-score stores | 0 / 100 | 4 stores |
Not a single store scored between 1 and 79. The line is invisible until you cross it — and once you cross it, you are either passing comfortably or missing entirely.
What 0 looks like
The four stores with a structured data score of zero share the same root cause: the crawler found no valid Product schema on any product page. In two cases the store was not serving JSON-LD at all. In the other two, the Product schema was present on some pages but corrupted, incomplete, or blocked behind JavaScript that the crawler could not render. To an AI agent the result is identical — the product's price, availability, brand, and description are unreadable.
These are not small operations. They include well-known fashion brands with millions in monthly traffic.
The overall readiness scores for this group ranged from 28 to 40 out of 100. Even where crawlability was perfect (one store scored 100 on that axis), the zero on structured data dragged the total below any useful threshold. In AI commerce, being findable but unreadable is barely better than not being found at all.
What 80+ looks like
The four stores that scored well on structured data averaged 88 out of 100 — a comfortable pass. They all included the essential fields identified in our earlier breakdown: name, price with priceCurrency, availability as a full schema.org URL, brand, image, and either sku or gtin. Their median consistency score was 100, meaning every product page used the same template — no drift between listings.
Their overall readiness scores ranged from 61 to 72 out of 100, limited not by structured data but by the other categories.
The standout was a beauty store that reached 72 overall (89 on structured data, 100 on crawlability). The AI preference score pulled it down — but that category improves automatically when a competitor URL is provided for comparison, rather than relying on a neutral baseline.
The hidden gap: policy data
Even among the top performers, two common fields were missing from every Product block: shippingDetails and hasMerchantReturnPolicy. The stores that scored 89 on structured data still scored 0 on policy — because the return window, shipping cost, and warranty terms were described in prose on a separate page rather than included in the product's JSON-LD.
An AI agent comparing two identical products at the same price prefers the store that can prove its return policy in machine-readable form. This is a one-field fix in the structured data template, yet almost no store in our sample had it.
The simple win: llms.txt
5 of the 8 stores we scanned had no llms.txt file — a single text file at the root of the domain that tells AI crawlers where to find structured content. Unlike robots.txt, which is about blocking or allowing, llms.txt is about inviting. Its absence leaves it to the agent to guess which pages matter.
This is a 10-minute deployment — one file, no code change, no risk — that directly improves crawlability. For the stores that were missing it, fixing that alone would move their crawlability score from 40 to at least 80.
What the data says
AI readiness is not a gradient you improve point by point. For most stores, it is a binary: either Product schema is present and valid, or it is absent and the store is invisible to AI buyers. The ones in the middle — stores with partial, inconsistent, or JavaScript-rendered schema — are scored as zero because that is how AI agents see them.
The takeaway is blunt, but it is the one the data supports: before you optimise your AI preference or tune your crawlability, confirm you have a valid Product schema on every product page. That one check separates a store that can be recommended from one that cannot.
Run a free scan of your store at prefero.me — we will tell you exactly where you stand on structured data, crawlability, and AI preference, with every issue linked to a fix.