AI is transforming online commerce. “Orders from AI search on Shopify increased 15x in 2025,” according to Shopify President Harley Finkelstein, though that growth comes off a small base.
The opportunity behind it is large: McKinsey estimates agentic commerce could reach $3 trillion to $5 trillion globally by 2030. If you don’t make your website AI agent-friendly, you will fall behind. Structured data is essential to that effort.
This guide covers what structured data is and how to implement it on your website. We also show you how to test whether your site is actually readable to agents today.
What is structured data, and why do AI agents need it?
Structured data is how your page speaks the agent’s language. It labels and organizes webpage information so machines and AI can understand it.
Consider a service described in natural language as: “Monthly plan is $79. Pay annually, get two months free.”
A human reads that instantly. An AI agent must tokenize the sentence, parse grammar, infer the payment structure, and calculate totals. It usually succeeds, but inaccuracy is possible.
Structured data eliminates guesswork. Instead of forcing agents to parse natural language, you provide information already labeled and ready to read, using agreed-upon labels from Schema.org.
Structured data can be formatted as JSON-LD, Microdata, or RDFa. JSON-LD (JavaScript Object Notation for Linked Data) is recommended.
For the pricing example above:
{
"@context": "https://schema.org",
"@type": "Offer",
"priceCurrency": "USD",
"priceSpecification": [
{
"@type": "UnitPriceSpecification",
"price": "79",
"priceCurrency": "USD",
"billingDuration": "P1M",
"name": "Monthly plan"
},
{
"@type": "UnitPriceSpecification",
"price": "790",
"priceCurrency": "USD",
"billingDuration": "P1Y",
"name": "Annual plan (2 months free)"
}
]
}This script tells the AI agent everything about your pricing: two plans, their prices, currencies, and billing frequency, all labeled explicitly.
What is Schema.org?
Schema.org is a shared vocabulary for structured data, created jointly by Google, Microsoft, Bing, and Yahoo in 2011. It defines a standard list of content types and properties, such as Product, Offer, LocalBusiness, and FAQPage, so that any machine, search engine, or AI agent that knows Schema.org can instantly understand what each label means.
Which schema types matter most for AI agents
When an AI agent lands on your page, it looks for specific information to complete the task a user gave it. The schema types that matter are the ones that answer that task directly.
Here is what each business should prioritize.
Identity and contact
- Organization: tells an agent who you are: legal name of the organization, logo, address, contact information, and company identifiers.
- LocalBusiness: adds physical location data: address, phone number, open hours, ratings, directions, and actions to book appointments or order items.
- Person: used for individuals, professionals, or authors; establishes a named entity with credentials and contact details.
These three schema types answer the “who are you and how do I reach you” question in a single, machine-readable block.
One field is worth adding to your Organization markup on its own: sameAs. It links your entity to authoritative profiles elsewhere, so an agent can confirm it is looking at the right company and not a similarly named one. Point sameAs at stable, canonical profiles like your LinkedIn page, your Wikidata entry, and your Crunchbase profile.
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "Acme Analytics",
"url": "https://www.acmeanalytics.com",
"sameAs": [
"https://www.linkedin.com/company/acme-analytics",
"https://www.wikidata.org/wiki/Q00000000",
"https://www.crunchbase.com/organization/acme-analytics"
]
}Products and services
- Product: information about your product, including price, availability, and review ratings.
- Offer: attaches price, currency, and availability to a Product so an agent can confirm it’s in stock before recommending it.
- Service: covers non-physical offerings: what the service does, who provides it, and where it’s available.
Together, these schema types give an agent everything it needs to evaluate, compare, and recommend what you sell.
Conversational Q&A
- FAQPage: maps common questions to direct answers. AI agents are fundamentally question-answering systems, so this format aligns precisely with how they process and surface information.
- HowTo: breaks a process into numbered steps. Agents can pull individual steps to answer “how do I” queries without reading your full article.
These two schema types turn your content into direct answers, the format agents prioritize when users ask questions.
One caveat on FAQPage: Google removed FAQ rich results from search listings on 7 May 2026, so this markup no longer earns you an expanded SERP snippet. FAQPage remains a valid Schema.org type that AI systems still parse, so the reason to use it now is machine comprehension, not a rich result.
Actions
- potentialAction: It tells an agent what can actually be done here: book, order, search, or contact.
potentialAction is in a category of its own. The next section covers exactly how to implement it and why it’s the most direct signal you can send to an action-capable agent.
How to implement the potentialAction property
In the agentic era, potentialAction is one of the most important schema types you can and should implement. For an agent working on a user’s behalf, completing the task given by the user is the goal, and potentialAction is how your website can facilitate that task.
When an agent lands on your page, it looks for what it can do there. potentialAction answers that question directly. Schema.org defines several action types, each telling the agent a different kind of task is possible:
Transactions:
OrderAction: the agent can place an order on the user’s behalfBuyAction: the agent can initiate a direct purchaseQuoteAction: the agent can request a price quote for the userRentAction: the agent can arrange a rental or leasePreOrderAction: the agent can secure a product before it is available
Bookings and reservations:
ReserveAction: the agent can book a table, slot, or appointment
Search and discovery:
SearchAction: the agent can run a search on your site using the user’s query
Communication:
ContactAction: the agent can submit an inquiry on the user’s behalfSubscribeAction: the agent can sign the user up for a newsletter or service
Each action type includes a target property pointing to the URL where the action is handled: a booking form, a checkout page, or a contact form.
Today, most agents call the target URL and route the user there with the relevant context pre-filled. The user completes the final step.
As agentic capabilities mature, agents will call URLs programmatically and complete actions entirely on users’ behalf, requiring no manual steps. If you run a store, this is the same groundwork behind getting your online store ready for agentic commerce.
How to implement structured data
How you implement structured data depends on the platform your site is built on. Most CMS platforms handle the basics automatically and give you a straightforward way to add custom JSON-LD for more advanced schemas like potentialAction and Offer.
Here is how it works across the most common ones.
WordPress
Plugins like Yoast SEO, Rank Math, and Schema Pro generate and inject JSON-LD automatically based on your content. For complex schemas like potentialAction, Rank Ranger Schema Markup Generator to build the JSON-LD, then paste it into your theme’s <head>.
Shopify
Most Shopify themes include basic Product and Offer schema by default. For additional schemas like potentialAction, you will need a third-party app like Schema Plus. Or use a generator to build the JSON-LD and add it directly into your theme’s <head>.
Webflow
Webflow automatically generates schema markup from visible page content on static pages, so basic schemas are handled without any code. For more control, use a schema generator to build your JSON-LD and paste it into the <head> section of your page settings.
Wix
Wix automatically adds preset schema markup to product, booking, blog, and event pages. It also generates LocalBusiness markup when you add a business name and location.
For all other page types and advanced schemas like potentialAction, use a schema generator to build your JSON-LD and paste it into the Advanced SEO tab in your page settings. Note that Wix only accepts JSON-LD and has a 7,000-character limit per markup.
Custom or headless sites (Next.js, Nuxt, Gatsby, Astro)
Use a schema generator to build your JSON-LD, then place it inside a <script type="application/ld+json"> tag in the <head> of your HTML, server-rendered. This gives you the most control and is the most reliable method for AI agent readability.
Schema stacks to implement by business type
| Business type | Schema types to implement |
|---|---|
| Local service businesses | LocalBusiness, Service, FAQPage, potentialAction (ReserveAction / ContactAction) |
| E-commerce and product businesses | Product, Offer, Organization, potentialAction (OrderAction) |
| Content and informational sites | Article, FAQPage, HowTo, BreadcrumbList |
| Insurance agencies | InsuranceAgency, Service (one per coverage type), LocalBusiness, Review / AggregateRating |
For an insurance agency, stack InsuranceAgency with a separate Service block for each coverage type you sell (auto, home, life, commercial), LocalBusiness for your office details, and Review or AggregateRating for client ratings. Insurance is a high-trust, regulated space where a wrong quote or a misattributed policy carries real consequences, so machine-readable identity, services, and genuine reviews matter more here than they do for a low-stakes site: they help an agent represent the agency accurately.
Note: JSON-LD must be server-rendered and placed in the HTML <head>. If it is injected after page load by a JavaScript framework or plugin, most AI agents will never see it. They read the raw HTML response and do not wait for the page to finish rendering.
How to test whether AI agents can read your site
Testing this costs nothing and takes less than 15 minutes. Run through these three steps whenever you add or update schema markup.
Step 1: Validate your structured data
Go to Schema Markup Validator and enter your URL or snippet. It parses your page the way a machine does and flags issues. If it can’t find your markup, an AI agent won’t find it either.
Step 2: Confirm your JSON-LD is in the server-rendered HTML
Check that your markup appears in the raw HTML an AI crawler actually fetches, not only after client-side JavaScript runs. Open your page, view source, and search for application/ld+json. If it’s there in “view source,” a crawler that fetches your page as GPTBot will see it too. If it only appears in the rendered DOM after scripts run, most agents will miss it, and you need to move it into server-rendered HTML.
Step 3: Ask a live AI a task-style question
Open Claude or ChatGPT with browsing enabled.
Then ask something specific:
- “What are the opening hours and contact details for [business name]?”
- “What products does [site URL] sell and what are the prices?”
- “What services does [business name] offer?”
A well-marked-up page lets the AI answer accurately and immediately. If it hedges, guesses, or says it can’t find the information, your structured data is missing, incomplete, or not being parsed.
This step is the most revealing part of the test because it mirrors exactly how a real user might query an AI about your business.
Step 4: Compare, fix, and repeat
Put the AI’s answers next to your actual schema markup. Identify every field it got wrong or skipped. Fix those fields, redeploy, and run the test again.
While you’re at it, it’s worth checking your overall website accessibility too. A well-structured, accessible site is easier for AI agents to parse. WebYes Accessibility can help you identify and fix accessibility issues that affect both human visitors and AI agents.
Run this test every time your schema changes. It’s the fastest feedback loop available.
What to do next
Structured data is the language AI agents use to understand your site and act on it. Clearer markup makes your site more useful to agents working on behalf of real users.
Structured data is one piece of a larger picture. Semantic HTML and accessible web design work alongside it, including the accessibility tree AI agents rely on to navigate and take action. A well-structured, accessible website benefits not just users with disabilities, but also AI agents, too.
So start with a concrete sequence:
- Validate your schema in the Schema Markup Validator.
- Add
sameAsto yourOrganizationmarkup so agents can pin down your identity. - Confirm your JSON-LD is server-rendered in the HTML, not injected after page load.
- Choose the schema stack that matches your business type and implement it.
Sites with clean, complete structured data today are positioned to participate in the agentic web, whether that means completing purchases, booking services, or answering questions on users’ behalf.
FAQs on AI agents structured data
No. Structured data improves the chance that AI systems can read and understand your content, but it does not force a citation. High-quality, authoritative content is still the primary factor. Structured data removes ambiguity and helps AI systems process your facts accurately, but citation is ultimately at the AI system’s discretion.
Yes. AI systems read and consume structured data to find, parse, and correctly interpret the content on a page. Google’s Search team said in April 2025 that structured data helps its systems understand pages, and Microsoft’s Bing principal PM Fabrice Canel said much the same in March 2025. What structured data does not do is guarantee a ranking or citation lift. The evidence for a direct increase in AI citations is mixed and contested, so treat schema as something that helps agents understand you, not as a guaranteed visibility boost.
Yes. ChatGPT, Perplexity, and Google AI Overviews all use structured data signals to verify entities and extract facts from pages. JSON-LD is the preferred format because these systems can parse it without rendering the full page. Pages without schema give these models less to work with.
No. Focus on pages where an AI agent would need to take action or extract a specific fact: your homepage, product pages, service pages, contact page, and any page with FAQs or how-to content. Adding schema to every page, regardless of type, adds complexity without adding value.
Use JSON-LD built on Schema.org types. The high-value shortlist is Organization, Product / Offer, FAQPage, Service, and LocalBusiness, plus sameAs to link your entities to authoritative profiles. See “Which schema types matter most for AI agents” for what each one does and “How to implement structured data” for adding it on your platform.
They fetch your HTML and read the JSON-LD embedded in it. From there they resolve the entities on the page: who you are, what you sell, the price, and what’s in stock. They then use those labeled facts to answer a user’s question or take an action, instead of guessing at your prose.
Yes. Insurance is a high-trust, YMYL (your money or your life) space, so an agent needs to represent your agency accurately or not at all. Machine-readable identity, clearly labeled services per coverage type, and genuine reviews give agents the facts they need to describe you correctly. See the insurance agency stack in “Schema stacks to implement by business type.”
AUTHOR
Melwyn holds a Bachelor's degree in Computer Science and has been writing professionally for over 4 years. He has a strong background in SEO and is certified in web accessibility from the World Wide Web Consortium (W3C). When he's not writing about website optimisation, you can find him enjoying a game of FIFA or exploring the open world of GTA.
Melwyn Joseph 
