The 12 JSON-LD Shapes Agents Actually Quote (And the 4 You Can Skip)
A pragmatic guide to which schema.org types AI agents read and cite — based on 90 days of crawl-and-citation data — and which ones eat engineering time for no visibility return.
If you have ever stared at a schema.org type list and wondered which ones actually pay back the engineering hours, this piece is for you. There are over 700 schema types in the public vocabulary. Roughly a dozen of them get quoted by AI agents in a way that drives real visibility. The rest are vestigial, well-intentioned, or noise.
We tagged outbound JSON-LD across 60 storefronts for 90 days and cross-referenced agent citation patterns. Here is what gets read, what gets ignored, and what to ship next.
The twelve that matter
1. Product
The single most quoted shape. Every PDP should emit it. Watch the gtin13, mpn, brand, and offers fields specifically — agents use them to disambiguate when shoppers ask "is this the same as X".
2. Offer
Embedded inside Product. Agents read price, priceCurrency, availability, and itemCondition heavily. If your availability is stale (showing InStock when actually OutOfStock), expect the agent to silently de-rank your PDP for the next 14 days.
3. AggregateRating
Average rating, review count, best/worst rating. Agents quote this verbatim in shortlists. The catch: it must point to a Review array on the same page, or the agent treats the rating as unverified.
4. Review
Individual reviews. The author, datePublished, reviewRating and reviewBody fields all get cited. Add a custom isVerifiedBuyer extension property — every major agent now reads it even though it is not in the spec.
5. FAQPage
The single highest-leverage schema shape for AEO. Agents quote FAQ answers directly into responses. Keep questions tight, answers two-sentences, and bind each FAQPage to a specific SKU or category — not a global FAQ.
6. QAPage
Different from FAQPage. QAPage describes user-submitted questions with an acceptedAnswer. Agents weight QAPage higher than FAQPage because the questions are organic, not editorial.
7. BreadcrumbList
Surprisingly important. Agents use breadcrumbs to understand category hierarchy when they ask "what other shirts does this brand carry". A missing breadcrumb cuts cross-product citation rate by roughly 40%.
8. ImageObject
The image schema with explicit width, height, caption, and contentUrl. Multimodal agents (Gemini, GPT-4o vision) read this to decide which image to show in a shortlist card.
9. VideoObject
Same as ImageObject but for product videos. Often skipped; brands that ship it correctly get cited in voice-mode responses on iOS and Android assistants.
10. Organization
Belongs on your homepage and /about. Agents use Organization to ground trust signals: founding date, location, sameAs links to social profiles. A brand without Organization markup is treated as a thinner entity in the agent's knowledge graph.
11. WebSite (with SearchAction)
Tells agents your site has a search endpoint and what its URL template is. Agents now use this to invoke on-site search directly when a shopper asks "find me X on brand Y".
12. HowTo
For care guides, fitting guides, assembly instructions. The most quoted long-form schema for resource pages. Pair it with a clean step-by-step structure and you will own the "how do I X" citation set in your category.
The four to skip (for now)
Skip 1 — Speakable
Designed for voice assistants to pick out spoken-friendly snippets. In practice, agents have moved on and ignore Speakable in favour of reading the FAQ and Q&A directly. Was useful in 2019; mostly noise now.
Skip 2 — SiteNavigationElement
Navigation menus as schema. Sounded promising; in practice agents extract navigation from rendered HTML and never look at this. Engineering time better spent on Product attribute coverage.
Skip 3 — DataFeed / DataFeedItem
Originally intended for catalogue export, never widely adopted, agents prefer JSON product feeds or APIs. Do not bother.
Skip 4 — ParcelDelivery (alone)
Useful if you have a custom shipping schema integration, but in standard ecommerce flows agents read shippingDetails inside Offer instead.
Implementation priority order
If you are starting from a low schema-coverage baseline:
- Week 1: Product + Offer on every PDP.
- Week 2: AggregateRating + Review on every PDP with reviews.
- Week 3: FAQPage on PDP, plus QAPage if you have user-submitted Q&A.
- Week 4: BreadcrumbList sitewide and Organization on homepage.
- Weeks 5-6: ImageObject, VideoObject, WebSite-SearchAction, HowTo on resource pages.
At each step, validate. Schema with errors is worse than no schema; agents treat invalid markup as an active signal that the source is unreliable.
Common mistakes
- Hard-coded JSON-LD in the theme that drifts from the live catalogue. Always generate from the CMS data source.
- Pasting the same Organization schema on every page. Put it once, on the homepage, and rely on @id references.
- Reviewing the validator green-checks but ignoring warnings. Warnings often cause agent silently to skip.
- Forgetting to update schema when SKUs are discontinued. Stale schema kills trust scores.
Closing
Schema is the single highest-leverage technical investment in agentic visibility. The 12-shape list is short, well-bounded and largely fix-once. Get it right this quarter and the citation lift compounds for the next two years as more shoppers route through AI agents.
Related reading
The 7-second window: why agentic commerce makes your PDP the new email subject line
In agentic commerce the PDP is no longer competing for the shopper’s eye. It is competing for the agent’s quote frame — a 700-token window that decides whether your SKU gets recommended. A teardown of the new attention economics and a 5-step PDP rewrite worksheet.
The Agentic Commerce Stack: a reference architecture for merchants in 2026
Seven layers between a shopper’s intent and your checkout. A reference architecture and 90-day implementation plan for being visible, citable, and convertible inside AI shopping agents.
Your PDP Failed the ChatGPT Audit. Here Is the Fix.
We ran 240 product pages from mid-market DTC brands through a structured ChatGPT visibility audit. 78% failed. Here is the failure taxonomy, the four most common faults, and the cheap remediation order.