Schema Markup for AI Search: Entity SEO That Actually Gets Read
Which JSON-LD types matter for AI answers, how to connect your entity with sameAs, and the mistakes that make schema useless.
Structured data used to be decoration for rich results. In an AI-first web it is closer to a machine-readable business card: it tells a model who published a claim, who the author is, and how that identity connects to the rest of the web.
Start with the entity, not the article
Most sites add Article schema first. That is backwards. If the model cannot resolve who is speaking, the article's credibility is unanchored. Publish Person and Organization nodes once, with stable @id values, and reference them everywhere else.
{
"@context": "https://schema.org",
"@type": "Person",
"@id": "https://example.com/#person",
"name": "Kashi Ahmer",
"jobTitle": "Founder, MKM.Fan",
"url": "https://example.com/about-us.html",
"sameAs": [
"https://github.com/Kashiahmer",
"https://www.linkedin.com/in/example"
]
}The five types worth implementing
| Type | Answers the question | Use on |
|---|---|---|
| Person | Who wrote this? | Home and every article |
| Organization | Who is behind the site? | About page and home |
| WebSite | What is this site, and can it be searched? | Home only |
| Article / BlogPosting | What is this document? | Every post |
| FAQPage | What questions does this page answer? | Pages with real FAQs |
Rules that separate useful from decorative
- One source of truth per entity. Give Person a fixed
@idand reuse it; do not duplicate slightly different copies on forty pages. - Match visible text. Schema that contradicts the page is worse than no schema, and it is a policy violation in Google's eyes.
- Use absolute URLs for
url,imageandsameAs. - Keep descriptions factual. “Founder of MKM.Fan, a platform for publishing content on your own subdomain” beats “world-class visionary”.
- Validate after every change. A single trailing comma breaks the whole block and silently kills every property inside it.
Do not stack five unrelated JSON-LD blocks on one page hoping one of them sticks. Conflicting declarations of the same entity are worse than a single accurate one.
FAQPage: still worth it?
Rich-result display for FAQ markup has narrowed, but the semantic value for retrieval remains. Answer engines read question-answer pairs directly because they map onto the shape of a user query. Keep FAQs genuine — three to six questions you are actually asked — and never mark up invisible text.
A validation routine
- Generate the block with the JSON-LD Generator so quoting and structure are correct.
- Paste it into the Schema Markup Validator and confirm zero errors.
- Check the page in a browser with JavaScript disabled to ensure the JSON-LD is in the raw HTML, not injected later.
- Confirm the entity URL in the schema resolves to a real page with the same name.
Structured data is a trust layer, not a ranking trick. The audit on this site gives it 18 points out of 100 because it is one of the highest-leverage fixes available to a small site.
Frequently asked questions
Do I need schema to appear in AI answers?
No, but it substantially reduces ambiguity about authorship and organisation. Ambiguity is a common reason a model cites a competitor instead.
Can I use multiple @type values?
Yes, an array is valid. Keep it honest: do not claim to be both an Organization and a LocalBusiness unless both are true.
Does schema affect Google rankings directly?
Not as a direct ranking factor, but it improves eligibility for rich results and helps machines interpret the page.