Structured data for AI shopping agents
Structured data is the entire interface between an ecommerce store and an AI shopping agent. This page covers the formats agents extract, the order they try them in, how the fallbacks degrade, and how much of your catalogue has to be covered before it counts.
Structured data for AI shopping agents is machine-readable product markup — primarily schema.org JSON-LD, with microdata and Open Graph as degraded fallbacks — that lets an agent turn a product page into a comparable record without rendering it.
Extraction order and how it degrades
Agents try formats in a predictable order and stop at the first that yields a usable record. Each fallback loses fields, and losing fields is what removes you from a comparison.
| Format | Typical fields recovered | Verdict |
|---|---|---|
| schema.org JSON-LD | Full record: identity, offer, ratings, shipping, returns | The target. Everything else is damage control. |
| Microdata / RDFa | Identity and price, usually no policy or identifiers | Parsed, but you lose the fields that win comparisons. |
| Open Graph | Title, image, sometimes price | Enough to name the product, not to recommend it. |
| Raw HTML heuristics | Unreliable; frequently wrong price or currency | Treated as low confidence; often discarded. |
| Nothing | None | The page is not a product as far as the agent is concerned. |
Feeds are not a substitute
Merchants often assume a Google Merchant Center feed covers them. It does not. A feed serves the surfaces that ingest that feed; an agent fetching your product page directly sees the page, and if the page has no structured data the feed does not help.
Feeds and on-page structured data should be generated from the same source of truth so they cannot drift. Where they disagree, the page is the one an agent trusts, because it is the one it fetched.
Coverage matters more than perfection
One immaculately marked-up flagship product does nothing if the other 400 SKUs are blank. Agents sample; the probability that they hit your good page is proportional to how many good pages you have.
Measure coverage as the share of sampled product URLs that yield a complete record. In our 120-store benchmark the pattern was bimodal — stores tended to be near-complete or near-empty, largely determined by whether the platform generated markup by default. Shopify stores hit 59% coverage; custom builds hit 11%.
- Template-level, not page-level
- Fix the product template once rather than editing pages. Anything hand-maintained decays.
- Include variants
- Size and colour differences change price and availability. A single parent record loses variant-constrained queries.
- Validate continuously
- Theme updates, app installs and replatforms silently drop markup. Readiness is a monitoring problem, not a one-off task.
Frequently asked questions
- What structured data do AI shopping agents read?
- Primarily schema.org Product markup as JSON-LD in the page's initial HTML. Microdata, RDFa and Open Graph are parsed as fallbacks but recover fewer fields, and heuristic HTML parsing is treated as low confidence.
- Does a Google Merchant Center feed cover AI shopping agents?
- No. A feed serves the surfaces that ingest it. An agent fetching your product page reads the page. Feeds and on-page markup should come from one source of truth, and where they disagree the page wins because it is what was fetched.
- What percentage of my catalogue needs structured data?
- As close to all of it as your template allows. Agents sample; partial coverage means partial eligibility. Fix the product template rather than individual pages, since anything hand-maintained decays with theme and app changes.
- Does structured data help with classic SEO too?
- Yes — the same Product markup drives rich results and shopping surfaces in conventional search. This is why readiness work is low-risk: it pays off in channels you already measure regardless of how quickly agent traffic grows.
Measure your structured-data coverage
The scan samples product URLs from your sitemap and reports coverage, not just presence.