Agentic Commerce Checklist: 12 Machine‑Readable Facts Every App Page Must Emit
Written by AppWispr editorial
Return to blogAGENTIC COMMERCE CHECKLIST: 12 MACHINE‑READABLE FACTS EVERY APP PAGE MUST EMIT
AI agents (and the stores they recommend) will prefer product pages that publish canonical, machine‑readable facts. This post gives a short audit you can run in 48–72 hours and 12 concrete JSON‑LD snippets you can drop into product pages so agents can identify, compare, and — critically — take action (add to cart or checkout). The checklist prioritizes facts agents actually use: unique identifiers, offer shape, evidence blocks, callable checkout, freshness indicators, and seller identity.
Section 1
Why agents need machine‑readable facts (and what breaks fast)
Agents (multi‑step automated assistants) decide to recommend a product based on explicit, verifiable facts they can ingest: canonical identifiers, current price and currency, availability, seller, and how to complete the purchase. If facts are absent, inconsistent, or injected only by client JavaScript, agents either ignore the page or fall back to less reliable data sources.
Practical failure modes you can test in minutes: (1) the page shows a sale price visually but JSON‑LD omits price or priceCurrency, (2) availability is in text but not the Offer.availability property, or (3) there is no identifier (GTIN/SKU/@id) so the agent can’t dedupe variants. Any of these will make an agent avoid recommending your product.
- Agents prefer structured Offer/ Product — not scraped HTML only.
- Missing priceCurrency or price makes comparisons impossible.
- No unique identifier prevents reliable mapping across sources.
Section 2
The 12 facts: audit checklist (what to emit and why)
Below are the 12 machine‑readable facts agentic systems expect. Implement these as JSON‑LD Product + nested Offer blocks (or as clearly typed microdata/RDFa) placed in the server HTML. Each fact is paired with the practical reason agents need it and a minimal implementation note.
Treat this checklist as a hard audit: mark each item Pass/Fail on live product pages before deploying changes.
- 1. canonical @id — page‑level persistent URI so agents can dereference the entity.
- 2. name & description — short canonical name plus a single authoritative description.
- 3. primary image (URL) — absolute, high‑quality image URL.
- 4. brand & manufacturer — normalized brand name or Organization @id.
- 5. identifiers (GTIN/UPC/EAN, MPN, sku) — exact identifiers for SKU mapping.
- 6. offers.price & priceCurrency — numeric price and ISO currency code in Offer or PriceSpecification.
Sources used in this section
Section 3
The remaining 6 facts: availability, evidence, actions, and trust signals
7. availability (ItemAvailability) and inventory hints — InStock, OutOfStock, PreOrder. Agents use standardized enums (schema.org ItemAvailability) to reason about fulfillment windows. Include expected lead times where possible.
8. priceValidUntil / validFrom — time bounds for current price or promotion. Agents will ignore stale sale prices; use future UTC timestamps to signal active promos.
9. aggregateRating & reviewCount — numerical rating and count; agents use these for quality filtering (don’t invent numbers; if you lack reviews, omit).
10. seller (Organization) and returns/shipping policy pointers — explicit seller identity, seller home page @id, and references to policies so agents can vet trustworthiness.
- 11. evidenceBlock(s) — a small, machine‑readable provenance section: source URLs, timestamps, and optionally cryptographic proof URIs or order history references so agents can verify claims.
- 12. potentialAction / CheckoutAction or OrderAction endpoint — a callable URL or API description (ActionHandler) the agent can follow to add to cart or invoke a checkout flow.
Section 4
48–72 hour implementation plan (practical, staged)
Day 0–1: Audit and quick wins. Crawl a representative sample of product pages (or open a few SKUs) and check for price, priceCurrency, availability, and sku/GTIN in server HTML. These four are the highest‑impact facts for agent recommendation and for Google/Merchant compatibility.
Day 2: Add Product + nested Offer JSON‑LD blocks to templates. Keep JSON‑LD inlined in server HTML (not injected by JS). Use canonical @id URIs (your product page URL or a GS1 Digital Link for GTINs) so agents and crawlers can dereference the entity.
- Day 3 (optional but recommended): Add potentialAction for checkout + evidenceBlock. If you cannot expose a public order API, include an OrderAction with a web form URL as the target and clear parameters (sku, quantity).
- Quick validation: run Google Structured Data Testing and schema.org validator and spot‑check agent behavior with small agent scripts or a headless agent runner.
Sources used in this section
Section 5
Concrete JSON‑LD snippets (drop‑in templates and tips)
Below are minimal, copy‑paste JSON‑LD templates for the highest‑priority objects. Use these as a starting point and expand fields as you have them. Place the JSON‑LD inside a <script type="application/ld+json"> tag in the server‑rendered HTML of each product page.
Implementation tips: keep values canonical and machine‑typed (e.g., use a GS1 Web URI for GTIN where available), ensure price and availability match visible content, and set priceValidUntil in ISO‑8601 UTC to prevent stale promotions being used by agents.
- Snippet examples: Product + Offer (price, currency, availability), Organization seller node, potentialAction OrderAction with target URL, evidenceBlock pattern (simple array of source objects).
- Don’t forget to declare @context: "https://schema.org" and use standard types (Product, Offer, Organization, AggregateRating, Review, OrderAction/BuyAction).
Sources used in this section
FAQ
Common follow-up questions
Do I need to add all 12 facts to start benefiting agents?
No. Start with the highest‑impact facts: price, priceCurrency, availability, and a canonical identifier (sku or GTIN). These four enable pricing and availability comparisons. Add seller identity and a callable potentialAction next. Treat the rest as incremental: evidence blocks and checkout endpoints increase trust and automation.
Should JSON‑LD be server‑rendered or can it be injected with JavaScript?
Server‑rendered JSON‑LD is strongly preferred. Many crawlers and agent pipelines ingest the initial HTML; client‑injected markup risks being missed or out of sync. If you must inject, ensure progressive enhancement and test with the exact agent or crawler you expect to use.
What is potentialAction / CheckoutAction and how safe is exposing it?
potentialAction is a schema.org Action you can attach to Product or Offer to describe how to purchase (e.g., an OrderAction with a target URL or an API endpoint). It does not itself perform transactions — it documents the action an agent can take. If exposing a direct API, require authentication or tokenized flows for real purchases; agents will use the documented target as a programmatic affordance.
How do evidence blocks differ from reviews or ratings?
Evidence blocks are machine‑readable provenance statements: source URLs, timestamps, or transactional proof (e.g., order receipts, verification pages). Reviews and ratings are consumer feedback signals. Evidence blocks help an agent verify claims like price history, stock levels, or authenticity.
Sources
Research used in this article
Each generated article keeps its own linked source list so the underlying reporting is visible and easy to verify.
How To Add Product Snippet Structured Data | Google Search Central
https://developers.google.cn/search/docs/appearance/structured-data/product-snippet?hl=en
Schema.org
Schema.org Actions
https://schema.org/docs/actions.html
Schema.org
identifier - Schema.org Property
https://schema.org/identifier
Product data specification - Google Merchant Center Help
https://support.google.com/merchants/answer/7052112?hl=en-GB
GS1
GS1 Web URI Structure Standard
https://www.gs1.org/standards/Digital-Link/1-0
Referenced source
Structured Linked Data as a Memory Layer for Agent-Orchestrted Retrieval (preprint)
https://arxiv.org/abs/2603.10700
Next step
Turn the idea into a build-ready plan.
AppWispr takes the research and packages it into a product brief, mockups, screenshots, and launch copy you can use right away.