- Schema is not a ranking factor in AI search. Google states structured data is not required for generative AI features, and there is no AI-specific schema type.
- Schema is a machine-understanding tool. It removes ambiguity about what a thing is, who published it, and how entities relate. That is a different job from ranking.
- Correlation is not causation. Studies finding cited pages carry more structured data are describing well-maintained sites, not proving a lever.
- The types that earn their keep are
Organization,Person,Product,LocalBusinessandBreadcrumbList— because each feeds a system with a documented consumer. - The type most oversold is
FAQPage, whose rich-result eligibility Google restricted in 2023 and never restored for most sites.
"Add schema and you'll get picked up by ChatGPT" is one of the most confidently repeated claims in the AI search industry, and it is not supported by the documentation of any of the systems it describes.
What makes this hard to argue about is that the advice is not useless — it is misattributed. Structured data does real work. It just does not do the work it is being sold to do, and the difference matters when you are deciding what to spend a development sprint on.
This article separates three things people run together: whether schema influences ranking, whether it influences machine understanding, and what the correlation evidence can legitimately support. Then it goes type by type, because the honest answer differs sharply between Organization and FAQPage.
1. The claim, and where it came from
The claim usually arrives in one of three forms:
- "LLMs read JSON-LD, so schema is how you feed them."
- "Cited pages have more structured data, therefore schema drives citations."
- "There's a new AI schema type you need to add."
The third is simply false — no such type exists, and Google says so directly. The first is half-true in a way that misleads: a language model processing your page may well encounter your JSON-LD, but "the model can see it" is not the same as "the retrieval system ranks on it." The second is the interesting one, and it is where most of the confusion lives, so section 6 deals with it properly.
The origin of all three is a reasonable intuition: structured data is machine-readable, AI systems are machines, therefore structured data must be how the machines are fed. It sounds right. It survives because almost nobody checks it against the primary documentation.
Google's generative AI optimization guidance lists over-focusing on structured data among the things that are not required for generative AI features, and states there is no special schema.org markup for AI. It recommends structured data for rich results — a different, older, well-documented benefit.
2. Schema vs ranking: what the documentation says
Take the three surfaces separately, because they have different amounts of published evidence.
Google AI Overviews and AI Mode
These run retrieval-augmented generation on top of Google's core ranking systems. So the question "does schema help me appear in AI Overviews?" reduces to "does schema help me rank in Search?" — and Google has answered that one consistently for a decade: structured data is not a ranking factor. It determines eligibility for rich results, which is a presentation change, not a position change.
There is an indirect path worth naming honestly, because it is the strongest version of the pro-schema argument: rich results can lift click-through rate; behaviour and brand familiarity feed longer-term signals; a page that earns more engagement may perform better over time. That chain is real but long, and every link in it is weaker than the one before. It is not "schema → ranking."
ChatGPT search
OpenAI documents which crawlers do what, and says nothing about structured data as a retrieval input. What it does document is the thing that actually gates you: crawler access. Blocking OAI-SearchBot removes you from ChatGPT's search answers regardless of how perfect your JSON-LD is.
I have audited sites with immaculate structured data that were invisible in ChatGPT because of one line in robots.txt. The schema was not the problem, and no amount of it would have been the fix.
Perplexity
Perplexity attributes at passage level — it lifts specific claims and links them. That rewards clean, extractable page structure. Note what that means: the structure that matters is your HTML, not your JSON-LD. A well-formed table beats a Product block for a system trying to quote a specification, because the table is the thing on the page.
Ranking decides whether you are retrieved. Understanding decides whether what you are is unambiguous once retrieved. Schema operates almost entirely on the second. People sell it as the first.
3. Schema vs machine understanding: the real job
Here is the case for structured data, made without overclaiming.
Natural language is ambiguous and machines resolve ambiguity expensively. A page containing "Jayant Solanki, Indore, SEO" requires inference to determine whether that is a person, a business, an author byline, or a mention of someone else. The same facts in Person markup with a stable @id and a sameAs array require none.
What that buys you, specifically:
- Disambiguation. Which "Apex Consulting" is this, of the eleven with that name?
- Attribute extraction without parsing. Price, availability, opening hours and SKU as declared values rather than inferences from marketing prose.
- Relationship declaration. This article was written by this person, who works for this organisation, which offers this service in this area.
- Cross-source identity.
sameAsties your site to your LinkedIn, your Google Business Profile and your GitHub, so systems can consolidate rather than treat them as separate entities.
None of that is a ranking argument. All of it is worth having. A system that has to guess what you are will sometimes guess wrong, and when it does, you don't get a penalty — you get quietly omitted from a comparison, or merged with someone else's business.
4. Type by type: what each one actually buys you
The general advice "add schema" is close to useless because the types differ enormously in value. Here is my honest ranking.
| Type | Documented consumer | Worth it? |
|---|---|---|
Organization | Knowledge panel, entity consolidation | Yes — foundational |
Person | Author entity, knowledge panel | Yes — for author-led sites |
Product | Merchant listings, shopping surfaces | Yes — highest commercial value |
LocalBusiness | Local pack, Business Profile alignment | Yes — for physical or service-area businesses |
BreadcrumbList | Breadcrumb rich result | Yes — cheap and stable |
Article / BlogPosting | Article rich results, author linkage | Moderate |
Service | No dedicated rich result | Moderate — clarity only |
FAQPage | Restricted since 2023 | Low — and widely oversold |
Organization
The one to get right first. It declares that your business exists as an entity with a name, a canonical URL, a logo and a set of verified profiles. Everything else in your graph should reference it by @id.
Get right: a stable @id that never changes; name matching what you actually trade as; url pointing at the canonical hostname; logo as a real, indexable image; sameAs covering profiles you control; alternateName covering the variants people search.
Common failure: declaring Organization on the homepage with one @id and a different one on interior pages, so nothing consolidates. Pick one identifier and use it everywhere.
Person
For consultants, authors and anyone whose name is part of the offer. Same discipline: one @id, referenced from every article's author. This is the mechanism by which a body of work accrues to a person rather than to a collection of unconnected pages.
Worth stating: Person markup does not create authority. It makes existing authority attributable. If nothing off-site corroborates the person, the markup describes a claim nobody else is making.
Service
There is no service rich result, so the payoff is purely semantic — declaring what you offer, to whom, and where. I include it, at low priority, because "SEO consultant serving India and the UAE" as declared data is less ambiguous than the same phrase in a hero heading. Do not expect a visible result from it.
Product
The highest-value type for commerce, and the one where the AI argument is strongest — not because of ranking, but because AI shopping surfaces consume product data directly. Include offers with price, priceCurrency and availability; plus sku, gtin, brand, image and aggregateRating where genuine.
Your feed, your schema and your rendered page must agree. When an agent quotes a price the checkout then contradicts, you have manufactured a support ticket. Disagreement between the three is the single most common product-data fault I find, and fixing it beats adding new properties.
LocalBusiness
Essential for anyone with a physical location or a service area. address, geo, openingHoursSpecification, telephone, priceRange, and areaServed for service-area businesses.
The decisive factor is not the markup — it is consistency between the markup, your Google Business Profile and your citations. Identical NAP across all three is worth more than any property you could add to one of them.
Article and BlogPosting
headline, datePublished, dateModified, author by reference, publisher by reference, and a real image. Two cautions: dateModified must reflect an actual substantive edit, and author should point at a Person @id, not be a bare string.
BreadcrumbList
Cheap, stable, still produces a visible rich result, and tells a machine where a page sits in your hierarchy. There is no argument against it.
sameAs
Not a type but a property, and the most underrated item on this list. It is the explicit statement "this entity and that profile are the same thing" — the mechanism for consolidating an identity scattered across a site, a LinkedIn page, a Business Profile and a GitHub account.
Only list profiles you control and that are genuinely you. Padding sameAs with directory listings you don't own is noise at best.
FAQPage
The type most oversold and the one I'd deprioritise. Google restricted FAQ rich results in 2023 to a small set of authoritative government and health sites; for everyone else the visible benefit disappeared and has not returned.
It retains modest value as declared question-answer structure, and I still ship it on articles that genuinely have an FAQ section — including this one. But it is a tidy-up, not a strategy, and anyone selling FAQ schema as an AI visibility lever is selling a 2022 tactic.
5. Entity relationships and the Knowledge Graph
This is where structured data stops being a checklist and starts being architecture.
An isolated block of markup describes one page. A graph describes a business: this organisation, this person who works for it, these services it offers, this area it serves, these articles that person wrote. Systems that build entity representations care far more about the relationships than about any single node.
Mechanically, this means using @graph with stable @id values and referencing rather than repeating:
{
"@context": "https://schema.org",
"@graph": [
{ "@type": "Person", "@id": "https://example.com/#me",
"name": "…", "sameAs": ["…"] },
{ "@type": "Organization", "@id": "https://example.com/#org",
"founder": { "@id": "https://example.com/#me" } },
{ "@type": "BlogPosting", "@id": "https://example.com/post#article",
"author": { "@id": "https://example.com/#me" },
"publisher": { "@id": "https://example.com/#org" } }
]
}Repeating the author's name as a string on forty articles creates forty unconnected assertions. Referencing one @id forty times creates one entity with forty works attached. Same effort, different outcome.
Markup is a claim about yourself, and self-claims carry limited weight. Knowledge Graph inclusion follows corroboration from independent sources — press, profiles, citations, mentions. sameAs helps a system connect what already exists. It cannot manufacture the thing being connected.
6. What the correlation research does and doesn't show
Several 2025–2026 studies report that pages cited by AI systems carry structured data more often than pages that aren't. That finding is real and worth taking seriously. It is also the most misused statistic in this field.
What it supports: structured data is associated with a higher likelihood of citation.
What it does not support: adding structured data causes citation.
The gap between those is the whole argument, and there are at least four reasons the correlation would appear even if schema did nothing at all:
- Confounding by site quality. Sites with correct JSON-LD tend to have technical staff, working CI, server-side rendering, sane bot rules and maintained content. Any of those could produce the citation advantage.
- Confounding by ranking. Cited pages tend to rank; ranking pages tend to be on well-maintained sites; well-maintained sites tend to have schema. The chain runs through ranking, not through markup.
- Platform-generated markup. Shopify, WordPress with a decent SEO plugin, and most modern CMSs emit schema by default. Its presence may indicate "uses a mainstream CMS competently" rather than any deliberate act.
- Measurement bias. Studies vary in how they sample prompts and how they define a citation. Small differences there move the reported effect substantially.
None of this means the studies are wrong. It means the finding is descriptive. The only design that could establish causation is a controlled experiment — comparable pages, structured data added to one group, citation rates tracked over time. I have not seen one published at credible scale, and until someone runs it, "schema improves AI citation" is a hypothesis wearing a statistic as a disguise.
If a vendor cites a correlation study as proof that their schema product will get you cited, they are either not reading the study's own limitations section or hoping you won't.
7. How I'd implement it
Given all of the above, structured data still belongs in your build — sized to what it actually does.
Tier 1 — foundational
OrganizationorPersonwith a stable@id, used sitewidesameAslisting every profile you genuinely controlBreadcrumbListon every page below the root- NAP identical across schema, Google Business Profile and citations
Tier 2 — by business type
- Commerce:
Productwithoffers,sku,gtin,brand,availability— matching feed and page - Local:
LocalBusinesswithopeningHoursSpecificationandareaServed - Publishing:
Article/BlogPostingwithauthorandpublisherby reference - Services:
Servicenodes linked to the organisation
Tier 3 — optional
FAQPagewhere a real FAQ exists — expect no rich resultHowTofor genuinely procedural contentDatasetif you publish original data
Hygiene
- Validate with Google's Rich Results Test and the Schema.org validator
- Never mark up content that isn't visible on the page
- Keep
dateModifiedtruthful - One
@idper entity, reused — never regenerated per page - Re-validate after any template change; schema breaks silently
8. Frequently asked questions
Does schema markup help me rank in ChatGPT?
There is no documented evidence that it does. OpenAI does not list structured data as a retrieval input. What demonstrably governs your presence in ChatGPT search is crawler access — specifically whether OAI-SearchBot can reach your pages — and whether your content exists in server-rendered HTML.
Is structured data required for AI Overviews?
No. Google states structured data is not required for its generative AI features and there is no AI-specific schema type. AI Overviews retrieve using core Search ranking, so ordinary retrievability is the prerequisite, not markup.
But studies show cited pages have more schema. Doesn't that prove it works?
It proves an association. Structured data co-occurs with technical competence, server-side rendering, maintained content and good rankings — any of which could produce the citation advantage. Establishing causation needs a controlled experiment, and I'm not aware of one published at credible scale.
Which schema type should I add first?
Organization or Person with a stable @id and an honest sameAs array. Everything else in your graph should reference it. Get identity right before adding types.
Is FAQ schema still worth adding?
Only marginally. Google restricted FAQ rich results in 2023 to a narrow set of authoritative sites, and the visible benefit disappeared for everyone else. Add it where a genuine FAQ exists; don't build content around it.
Can schema get my business into the Knowledge Graph?
Not by itself. Markup is a self-claim. Knowledge Graph inclusion follows independent corroboration — press, profiles, citations and mentions. sameAs helps systems connect evidence that already exists; it can't substitute for it.
Should I use JSON-LD, Microdata or RDFa?
JSON-LD. Google recommends it, it keeps markup out of your presentation HTML, and it is far easier to generate, validate and diff.
Can incorrect schema hurt me?
Marking up content that isn't visible on the page violates Google's structured data policies and can cost rich-result eligibility or trigger manual action. Invalid markup is usually ignored rather than penalised — but it also isn't doing anything, which is its own cost.
Where I'd land
Structured data is worth implementing, worth maintaining, and worth validating after every template change. It is not worth buying as an AI ranking product, and it is not where I would spend the first week of a project.
If your pages aren't in the index, aren't server-rendered, or are behind a WAF rule that blocks the crawlers that matter, no amount of JSON-LD will help — those are the problems that actually gate AI visibility, and I've written about both in how to get cited by AI search and making your site agent-ready.
Fix retrievability first. Then make your entity unambiguous. In that order.
Structured data that describes your business, not just your pages
I'm Jayant Solanki — an SEO, GEO and automation strategist working with eCommerce, local-service and global brands. I build entity graphs that consolidate rather than fragment, and I'll tell you when markup isn't your problem.
A structured data engagement typically covers:
- Entity graph design with stable identifiers across every template
- Product, LocalBusiness and Article implementation matched to your business type
- Feed, schema and rendered-page consistency for commerce
- NAP and
sameAsconsolidation across every profile you control - Validation in CI, so a template change can't silently break the graph
Ranked #1 for "metal buildings" · +30% YoY organic traffic · Top-3 map pack across 10+ keywords
Marketer Companion parses and validates structured data on any page alongside Search Console and GA4 data — free Chrome extension. Try it free →
Sources
- Google Search Central, Optimizing your website for generative AI features on Google Search
- Google Search Central, Introduction to structured data markup
- Google Search Central, Structured data general guidelines
- Google Search Central Blog, Changes to HowTo and FAQ rich results (August 2023)
- Schema.org, Getting started with schema.org
- OpenAI, OpenAI crawlers
Jayant Solanki
AI-Ready SEO, GEO & AIO strategist based in Indore, India, working with eCommerce, local service and global brands across India, the UAE and the US. Ranked a client #1 for a 45,000/month commercial keyword; builds the measurement and automation behind the work.
More about Jayant →