Schema for AI Agents & Taskability: Using Actions, Offers, and Verification Markup

Schema for AI Agents & Taskability

AI assistants increasingly choose experiences they can execute, not just links to pages. To be selected, your site must expose taskable capabilities: clear Actions that agents can call, Offer and pricing data agents can use to transact, and verification signals that prove identity and trust.

This article explains the schema patterns that matter, shows concrete JSON‑LD examples for Actions and Offers, and gives a verification and testing checklist so you can make your pages reliably discoverable and executable by assistants.

Key schema patterns and JSON‑LD examples

Make taskability explicit using potentialAction on relevant entities. Common actionable types include SearchAction, ReserveAction, OrderAction, and ApplyAction. Combine these with Offer (or PriceSpecification) so assistants can evaluate cost/availability before executing.

Example: SearchAction (site search)

{
  "@context": "https://schema.org",
  "@type": "WebSite",
  "url": "https://example.com/",
  "potentialAction": {
    "@type": "SearchAction",
    "target": "https://example.com/search?q={search_term_string}",
    "query-input": "required name=search_term_string"
  }
}

Example: ReserveAction with Offer (booking or appointment)

{
  "@context": "https://schema.org",
  "@type": "Service",
  "name": "Therapy Session",
  "provider": {
    "@type": "LocalBusiness",
    "name": "Example Clinic",
    "url": "https://example.com"
  },
  "offers": {
    "@type": "Offer",
    "price": "120.00",
    "priceCurrency": "USD",
    "availability": "https://schema.org/InStock",
    "url": "https://example.com/book"
  },
  "potentialAction": {
    "@type": "ReserveAction",
    "target": {
      "@type": "EntryPoint",
      "urlTemplate": "https://example.com/book?service=therapy&slot={slot}",
      "actionPlatform": [
        "http://schema.org/DesktopWebPlatform",
        "http://schema.org/AndroidPlatform",
        "http://schema.org/iOSPlatform"
      ]
    },
    "result": {
      "@type": "Reservation",
      "name": "Therapy Session Reservation"
    }
  }
}

Implementation notes: keep URL templates stable, document the parameter names (for example slot or date), and make sure endpoints return consistent, machine-readable success/failure states. Agents prefer predictable, programmatic flows (CORS, documented status codes, and authenticated endpoints where appropriate).

Verification, safety, and rollout checklist

There is no single "verification" property that guarantees an assistant will run a sensitive task. Instead, build layered provenance and trust signals so integrators and assistants can validate your identity and policies.

• Identity and provenance: include sameAs, consistent identifier (brand IDs, business registry numbers), publisher/provider metadata, and accurate dateModified timestamps.
• Trust signals: expose Review, AggregateRating, Award, and when applicable ClaimReview for fact-checked assertions. Represent third‑party badges and accreditation as structured data where possible.
• Offer fidelity: ensure offers have correct price, priceCurrency, availability, and an url that maps to the exact action/booking endpoint.

Testing & monitoring checklist

1. Validate JSON‑LD with Schema.org guidance and the major validator tools (Rich Results / Schema validators).
2. Simulate agent flows end‑to‑end: call EntryPoint URLs, test parameter behavior, and confirm responses for success and failure paths.
3. Instrument observability: log assistant-triggered requests, monitor error rates, and track conversions that originate from agent flows.
4. Roll out incrementally: expose read-only Actions (SearchAction) first, then add booking/payment Actions after identity and fraud protections are in place.

Conclusion: To increase the chance assistants will pick your experience, make actions explicit with potentialAction, provide precise Offer and pricing data, and layer verification signals so agent integrators can trust your service. Publish stable URL templates, document parameters and response formats, and monitor behavior after launch.