Schema Markup for AI Search: The Four Types That Matter
Schema markup is a labelling layer that tells machines what your content means instead of making them guess. Schema.org defines hundreds of types, but a normal business needs four. This covers which four, how to implement them, and the mistakes that quietly break them.
What is schema markup?
Schema markup is structured data added to a page that describes its content in a machine-readable vocabulary. A human reading your page can tell that "Paid Media" is a service and "PHP 45,000" is a price. A parser cannot, unless you label them.
Use JSON-LD: a block of JSON inside a <script type="application/ld+json"> tag in your <head>. Google recommends it, and because it sits separate from your visible HTML it survives redesigns instead of being scattered through your markup.
Why does it matter more for AI search than for Google?
Because an answer engine has a higher bar than a search engine. Google can rank a page on signals like links and relevance without deeply understanding it. An AI assistant composing an answer has to be confident enough about a claim to attribute it to you.
Structured data lowers that risk. When your page explicitly declares "this is an Organization, based in Quezon City, that provides these Services, and here are five Question and Answer pairs", the model is not inferring anything. It is reading a labelled record.
That is also why the highest-value type for AI visibility is the one nobody bothers with.
Which four types actually matter?
1. Organization
Who you are, where you are, what you do. This is what allows an AI engine to answer "agencies in Quezon City" with any confidence.
Include name, url, description, email, and critically address and areaServed. Without geographic properties you are invisible to location-qualified questions, which are usually the highest-intent ones.
2. Service
One per thing you sell, on its own page. Include name, description, provider, serviceType, and an offers block if you can state a pricing approach.
Pricing is where most businesses go quiet. "Custom quotes, project-based" is a perfectly good value for offers.description, and it is far better than nothing. A model asked "how much does this cost" needs something to work with.
3. FAQPage
This is the highest-leverage type for AI visibility, and the most commonly skipped.
FAQPage hands the engine pre-formed question and answer pairs in exactly the structure it wants to produce. You are not asking it to extract an answer from prose. You are giving it the answer, already paired with its question.
One rule: the questions and answers in your markup must also appear as visible text on the page. Marking up content users cannot see is a guidelines violation and can earn a manual penalty.
4. Article or BlogPosting
For content pages. Include headline, description, datePublished, dateModified, author and image.
author matters more than people expect. In an era where a lot of published content is machine-generated, a named author with a real identity is a credibility signal. Link the author to a Person with a url pointing at a genuine bio.
What breaks structured data in practice?
Five recurring failures:
- Marking up content that is not on the page. The most common penalty trigger.
- Escaping mistakes that break the JSON. A value containing
</script>can terminate the block early. Escape angle brackets in emitted JSON. - Stale dates.
dateModifiedthat never updates tells engines your content is abandoned. - Missing geographic data. No
addressorareaServedmeans no local answers. - Testing pasted code instead of the live URL. Your templating may be mangling output. Always validate the real page.
That second one is a genuine security issue as well as a correctness one. If your schema is generated from user-editable fields, a title containing a closing script tag can break out of the JSON block entirely.
How do you check it is working?
Two tools, used together:
- Google Rich Results Test for eligibility and Google-specific warnings
- Schema.org validator for strict vocabulary correctness
Run both against the live URL. Then check your own pages the way an engine would: view source and confirm the JSON-LD is present in the initial HTML rather than being injected by JavaScript after load. Structured data added client-side is unreliable for AI crawlers, many of which do not execute JavaScript.
What schema cannot do
It cannot manufacture authority. Marking up a thin page as an Article with an author does not make it credible; it makes it a well-labelled thin page.
Structured data describes content that already exists. If the underlying content does not answer anything clearly, no amount of markup will get it cited. Fix the content first, then label it. That order is covered in how to get cited by ChatGPT.
Frequently Asked Questions
What to do next
Audit what you already emit. View source on your homepage and a service page and see which of the four types are present. Most sites have none, or an Organization block missing its address.
Add the missing ones in the order above, then pair it with llms.txt to complete the machine-readable layer. The full sequence sits in our guide to AI visibility in the Philippines.
Frequently asked questions
Does schema markup help with AI search?
Yes, indirectly but meaningfully. Structured data tells machines what each part of your page means rather than making them infer it, which makes your content easier to extract and safer to attribute. It is not a ranking factor on its own.
Which schema format should I use?
JSON-LD, placed in a script tag in the head. Google recommends it and it keeps markup separate from your HTML, so it is far easier to maintain than Microdata or RDFa.
Do I need every Schema.org type?
No. Schema.org has hundreds of types and most sites need four: Organization, Service, FAQPage and Article or BlogPosting. Adding obscure types adds maintenance burden without adding value.
Will schema markup improve my Google rankings?
Not directly. Structured data does not boost rankings by itself. It makes rich results eligible and makes your content easier for machines to parse, which affects click-through and citation rather than position.
How do I test my structured data?
Use Google's Rich Results Test for eligibility and the Schema.org validator for correctness. Test the live URL rather than pasted code, so you catch problems introduced by your templating.
Sources
Related reading
AI Visibility in the Philippines: How to Get Your Business Cited by ChatGPT, Perplexity and Google AI
AI visibility means getting your business cited inside AI answers. How Philippine businesses earn citations in ChatGPT, Perplexity and Google AI Overviews.
How to Get Your Business Cited by ChatGPT
ChatGPT cites sources it can extract and attribute with confidence. Here is the practical checklist for getting your business named in AI answers.
What Is llms.txt and Should Your Site Have One?
llms.txt is a plain-text summary of your site written for AI models. Here is what it does, what goes in it, and whether it is worth your time.