Schema for Search and AI: What It Helps With and What It Doesn't
Schema markup is one of the most talked-about and most misunderstood tools in AI search advice. It's genuinely useful, but not for the reasons a lot of current content suggests. This guide sets out what schema actually does, what it's documented to help with, and where the claims made about it go further than the evidence supports.
What schema is
Schema, more precisely referred to as structured data using the Schema.org vocabulary, is a standardised way of labelling the content on a webpage so that machines can interpret it more reliably than they can from unstructured text alone. Rather than a search engine having to infer that a particular block of text is a business's opening hours, structured data can explicitly state it, in a consistent, machine-readable format.
It's typically implemented as JSON-LD, a small script added to a page that describes the entities and facts on that page: a business's name and address, a product's price, an article's author and publish date, and so on. It doesn't change what a visitor sees on the page; it adds a parallel, structured description underneath.
Schema.org support is not the same as Google Search support
This distinction is easy to miss and worth being explicit about. Schema.org defines hundreds of vocabulary types covering almost anything a webpage might describe. Google, however, only supports a much smaller, defined subset of these for specific Search behaviours and visual features, and Google directs developers to its own Search Central documentation, not Schema.org alone, as the authority for what Google Search actually does with a given type.
That means a page can carry perfectly valid Schema.org markup for a type Google doesn't have a corresponding Search feature for. The markup isn't wrong, it just doesn't produce a visible outcome in Google specifically. This matters because a lot of schema advice treats "it's in Schema.org" and "Google supports it" as the same thing, when they aren't.
What Google's documentation says schema does
Google's own documentation distinguishes between required, recommended, and optional properties depending on the specific structured data feature. Organisation structured data currently has no required properties; Google recommends adding whichever properties are genuinely relevant. Many other supported rich-result types do have required properties, and missing them makes a page ineligible for that specific feature; LocalBusiness, for example, requires at minimum a name and an address to be eligible for its associated rich result.
Google's guidance for its AI-powered Search features, AI Overviews and AI Mode, states that these features are rooted in Google's core Search systems and don't require special AI-specific structured data. Schema isn't a separate mechanism for "AI visibility" distinct from how it's always worked for Search.
What schema doesn't do
This is where a lot of AI search advice overstates things.
There's no primary platform guidance establishing generic Schema.org markup as a direct ranking, citation, or recommendation signal across AI-generated answers. For Google specifically, this can be stated firmly: Google explicitly says no special Schema.org markup is required for AI Overviews or AI Mode. OpenAI's current publisher guidance for ChatGPT Search likewise doesn't identify Schema.org markup as a requirement for inclusion; its guidance instead focuses on allowing OAI-SearchBot to crawl a site.
It doesn't compensate for poor or unclear content. Structured data describes what's already on a page. It can't invent clarity that isn't otherwise there. A vague page with excellent schema markup is still a vague page.
It isn't a guaranteed path to rich results, even where a type is supported. Meeting the technical requirements for a given structured data type makes a page eligible for the associated feature, not automatically entitled to it; Google's own systems still decide whether and how to display it.
It isn't a generic ranking factor. Google has publicly stated structured data itself isn't a generic ranking signal, though it can help Google understand pages and enable additional Search experiences where relevant.
What schema genuinely helps with
None of the above means schema isn't worth doing. It's genuinely useful for a few specific, documented reasons:
Reducing ambiguity for machines. Structured data provides explicit, machine-readable information about what a page represents, reducing how much a search system needs to infer solely from unstructured content. This supports the broader goal of entity clarity, which we cover in more detail in our Entity SEO guide.
Enabling specific, supported visual features in Google Search. Where a business's content matches the requirements for a Google-supported type, correctly implemented schema can make it eligible for the corresponding rich result.
Supporting entity and Knowledge Panel information. Google has expanded its supported Organisation properties over time (including name, alternateName, legalName, sameAs, address, and contact details), specifically to help it disambiguate a business and represent its details more accurately in features like knowledge panels.
Making relationships explicit. Properties like sameAs allow a business to explicitly connect its website to other verified profiles, giving search systems an explicit link rather than requiring them to infer the connection.
Documented, indirect, and unproven: three different kinds of schema claim
A lot of confusion around schema comes from blending three genuinely different categories of claim together:
Documented. Google uses structured data as explicit information about page content and entities, and to enable specific, supported Search features. This is confirmed directly in Google's own developer documentation.
Indirect. Better-defined entities and clearer relationships may contribute to a more coherent, machine-readable representation of a business overall. That's a reasonable inference, but it isn't the same as any platform documenting a direct AI-citation boost from schema specifically.
Unproven. Claims such as "adding schema will increase ChatGPT citations," "FAQ schema improves AI recommendations," or "more schema means better AEO performance" are repeated often but aren't backed by platform documentation. They shouldn't be presented as established mechanisms.
Keeping these three categories separate is a useful discipline when evaluating any piece of schema advice, including this one.
Common schema types worth knowing
Organisation — Google-supported for organisation details and disambiguation. Some properties are used to distinguish a business from similarly named organisations; others can influence visual elements like the logo shown in Search or a knowledge panel.
LocalBusiness (a subtype of Organisation) — Google-supported, with name and address required for rich-result eligibility. Relevant for businesses operating from a specific physical location.
Article — a Google-supported structured data feature, covering properties like author, publish date, and headline for blog posts, guides, and news content.
BreadcrumbList — Google-supported for presenting a page's position within a site's navigation hierarchy.
Service — valid Schema.org vocabulary for describing a specific service a business offers, but not currently a standalone type in Google's Search rich-result gallery. Not every valid Schema.org type has a corresponding Google Search feature.
FAQPage — still a valid Schema.org type, but its associated Google Search rich result stopped appearing on 7 May 2026, and Google subsequently removed its FAQ rich-result documentation. FAQPage markup shouldn't be implemented with the expectation of gaining a dedicated Google Search appearance.
A note on validation and misconceptions
Structured data must accurately represent the visible content of the page it's on. Google's guidelines are explicit that markup describing something different from what's actually on the page, or added purely to try to manipulate a search feature, can result in a page losing eligibility for rich results or, in more serious cases, facing manual action.
Critical errors in structured data can make markup ineligible for the relevant Google rich result, while warnings may still leave a page eligible but flag missing recommended information. Google recommends its Rich Results Test for validating Google-supported Search features, and the general Schema Markup Validator for broader Schema.org validation.
More markup isn't automatically better. Google prioritises relevant, accurate, and sufficiently complete structured data; adding irrelevant types or properties purely to increase the amount of markup provides no documented advantage and may breach Google's structured data guidelines. Schema and AEO aren't the same thing either. Schema is one input into the broader picture of entity clarity and technical accessibility that AEO is concerned with, not a synonym for it.
An example
A restaurant's website might list its opening hours clearly in the page's visible text, but if that same information isn't reflected in structured data, a search engine still has to interpret it from unstructured text, which is more error-prone, especially where hours vary by day or include exceptions for holidays. Adding accurate LocalBusiness structured data with correctly formatted opening hours doesn't change what a visitor reads on the page; it gives Google an explicit, structured version of the same information to draw from with more confidence.
Common mistakes
Implementing schema and treating it as done. Structured data needs to be kept accurate as business details change; stale schema (an old address, discontinued services) can create the same kind of ambiguity as inconsistent information elsewhere.
Assuming schema improves AI citation likelihood. As covered above, there's no documented basis for this specific claim, despite how often it's repeated.
Treating Schema.org validity as equivalent to Google Search support. A type being valid in Schema.org doesn't mean Google has a corresponding rich result for it.
Treating FAQPage schema as a growth tactic. Since the associated Google Search rich result is gone, this specific type no longer delivers the Search visibility benefit many businesses implement it for.
Neglecting validation. Structured data with critical errors can be ineligible for its associated rich result entirely, silently providing no benefit.