Schema markup is having a second life in GEO pitch decks, usually oversold the same way it was oversold for rich results a decade ago. So let me start with the honest version.
Schema will not get you cited by ChatGPT on its own. It will not trigger FAQ rich results for your business site, because Google restricted those to government and health sites in 2023. And no AI engine lists structured data as a requirement for anything.
Here is what schema actually does in 2026: it makes your entity and your content unambiguous to every machine that reads your pages. In a search world where the machine reasons about you across many retrievals, unambiguous is worth real money. The trick is knowing which 20 percent of schema work delivers all of the value.
TL;DR
- What schema does: Schema markup makes your entity and content unambiguous to machines, which supports AI visibility indirectly rather than forcing citations or triggering rich results.
- The four that matter: For an expertise business, Organization, Person, Article, and FAQPage carry the value, connected into a graph and implementable in an afternoon.
- The FAQ caveat: Google restricted FAQ rich results to government and health sites in 2023, so I add FAQPage schema for the machines that extract its question-answer structure, not for the rich result.
- The one safety rule: Schema must describe what is on the page and never what you wish were on it, because markup that contradicts visible content is the one way structured data actively hurts you.
- Bottom line: Schema in 2026 is cheap entity infrastructure that supports content and authority work but cannot replace it.
What schema can and cannot do for AI search
Cannot: force citations, substitute for content quality, rescue weak rankings, or trigger rich results that your site type no longer qualifies for. Google is explicit that no special structured data is needed for AI Overviews.
Can: assert, in a format built for machines, the facts that retrieval systems otherwise have to infer: this organization is this entity with these official profiles, this author is this person with these credentials, this page is an article published then and updated then, this section is a question with this answer.
Inference is where machines make mistakes, and machines already make plenty about businesses. Schema is how you stop being a guess.
The four schema types that matter, in order
1. Organization
The foundation. Name, URL, logo, description, and crucially sameAs links to your LinkedIn, YouTube, and every official profile. This stitches your scattered web presence into one entity, which is exactly the consistency problem that entity audits keep finding. One block, sitewide, maintained.
2. Person
The E-E-A-T carrier. For an expertise business, the author entity is the asset: name, jobTitle, description, credentials, sameAs to the author’s profiles. Person schema connected to Organization and Article markup is the machine-readable version of “a real accountable expert wrote this,” which is precisely the signal AI surfaces lean toward when choosing sources.
3. Article
The content connector. Headline, author (linked to your Person entity), datePublished, dateModified. The dates matter more than they look: freshness is a citation signal across every AI surface I track, and dateModified is how you assert it honestly. Update the content, bump the date, never the reverse.
4. FAQPage
The extraction mirror. Yes, the rich result is gone for normal sites. Add it anyway, because it encodes your question-and-answer structure in the exact shape answer engines extract, and it costs nothing if your page already has real FAQs. The operative word is real: the markup must mirror visible on-page questions and answers, not invented ones.
Everything else (Product, Service, LocalBusiness, Event, the full gallery) is situational. If you sell products, Product schema is non-negotiable for shopping surfaces. If you are a consultancy, the four above do the heavy lifting.
The one rule that keeps schema safe
Schema describes what is on the page. Never what you wish were on it.
Markup that contradicts visible content (reviews that do not exist, FAQs that appear nowhere on the page, a dateModified that lies) is the one way structured data actively hurts you. It violates Google’s guidelines, and it teaches every machine that reads your site that your assertions cannot be trusted, which is the opposite of the entire point.
Honest, boring, accurate markup compounds. Clever markup gets caught.
Implementation: the practical path
- Use JSON-LD. Google’s recommended format, one script block, trivially validatable. No new microdata in 2026.
- Generate from the page, not from a template. Schema should be derived from actual content. My schema markup generator prompt builds validated JSON-LD from what a page really says, which enforces the honesty rule by construction.
- Validate twice. Syntax with the Rich Results Test or Schema.org validator, then truth: does every claim in the markup appear on the page?
- Wire the graph. Article points to its author Person, Person points to the Organization via worksFor or memberOf, Organization carries the sameAs links. The connections, more than the individual blocks, are what define your entity.
- Maintain dates ruthlessly. A dateModified that updates when content does is a freshness signal. One that updates on every deploy regardless is noise that machines learn to discount.
On this site, all four types are generated at build time from content frontmatter, which is the maintainability pattern I would push any team toward: schema as a build artifact of real content, not a hand-edited liability.
The takeaway
Schema markup in 2026 is entity infrastructure, not a rich-results lottery ticket. Four types carry the value for an expertise business: Organization, Person, Article, FAQPage, connected into a graph and ruthlessly honest about what is actually on the page. Implement them in an afternoon, generate them from real content, and then go spend your remaining energy on the content structure and authority work that schema supports but cannot replace.
Disambiguation is cheap. Being misunderstood by the machines that now answer your buyers is not.
Frequently asked questions
Does schema markup help with AI search visibility?
Indirectly but meaningfully. No AI engine documents schema as a citation requirement, and Google states no special structured data is needed for AI Overviews. What schema does is remove ambiguity: it tells machines exactly what your page is, who wrote it, what your organization is, and how your content is structured. Retrieval systems reward unambiguous sources, so schema supports the entity clarity and trust signals that do drive AI visibility.
Which schema types matter most for AI search?
Four, in priority order: Organization (defines your business entity: name, logo, sameAs links to your profiles), Person (establishes the author entity behind your expertise), Article (connects content to author, dates, and publisher), and FAQPage (mirrors question-answer structure machines extract). These four map directly to the entity definition and authorship signals AI retrieval leans on. Everything else is situational by business type.
Does FAQ schema still give you rich results in Google?
For most sites, no. Google restricted FAQ rich results to authoritative government and health sites back in 2023, so a consultant or business site will not get the expanded FAQ display in classic results anymore. That is not a reason to skip FAQPage schema: it still describes your content structure to every machine reading the page, and the question-answer format it encodes is exactly the shape AI engines extract. Add it for the machines, not the rich result.
Can schema markup hurt my site?
Yes, in one specific way: schema that contradicts the visible page. Marking up reviews that do not exist, FAQs that do not appear on the page, or dates that do not match erodes machine trust and violates Google’s structured data guidelines. The rule is simple: schema describes what is on the page, never what you wish were on it. Honest markup is safe; aspirational markup is a liability.
Should I use JSON-LD or microdata for schema?
JSON-LD. It is Google’s recommended format, it lives in a single script block instead of being woven through your HTML, it is easier to generate, validate, and maintain, and every major consumer of structured data parses it. There is no practical case for new microdata implementations in 2026.
Do AI engines like ChatGPT actually read schema markup?
Retrieval systems fetch your HTML, and schema is part of that HTML, so it is available to them. None of the major AI providers documents exactly how structured data weights their citation decisions. What is documented and observable: entity consistency and clear authorship correlate with citation, and schema is the most machine-legible way to assert both. Treat it as cheap disambiguation insurance, not a ranking lever.
Schema review is one slice of the technical-access work in my AI Search Visibility and SEO Strategy service. Questions about your specific markup? Book a free 30-minute call and we will look at it together.
