← Guides

Schema markup for AI answers: which types actually earn their keep

Published 12 Nov 2026 · 9 min read

Schema markup helps a machine parse facts it can already read on your page — address, opening hours, services, prices — and it cannot make a thin page substantial or an untrue claim true. Google states that no special structured data is needed to appear in its AI features. The types worth your time are LocalBusiness and its subtypes, Organization, Service, FAQPage, and, with real restrictions, Review and AggregateRating.

What structured data does, and what it does not

Structured data is a translation layer. Your page says "open till half five"; the markup says closes 17:30. Your page says "we cover Mansfield and round about"; the markup says areaServed Nottinghamshire, England. A machine reading the page can now be certain of facts it would otherwise have to infer from prose.

Be clear about the limit. Google's own guidance on its AI features in Search says there is no special schema.org structured data that you need to add, and that what matters is your structured data matching the visible text on the page. Markup is not a ranking lever or a visibility switch. It is an accuracy aid, and its benefit is that a model summarising your sector is less likely to get your hours, location or services wrong.

That is worth having. Most of the damage we see is not absence — it is a model confidently naming a business with the wrong town or a service it stopped offering two years ago.

LocalBusiness: the one to get right

For anything with a service area, this is the highest-value markup on your site. Google requires only two properties, name and address, and recommends geo, telephone, url, priceRange, openingHoursSpecification, aggregateRating and review. Use the most specific subtype available — Google's documentation says to use the most specific LocalBusiness subtype possible.

{ "@context": "https://schema.org", "@type": "Dentist", "name": "Headingley Dental Practice", "url": "https://example.com/", "telephone": "+44 113 496 0000", "priceRange": "££", "address": { "@type": "PostalAddress", "streetAddress": "14 Cardigan Road", "addressLocality": "Leeds", "addressRegion": "West Yorkshire", "postalCode": "LS6 1LJ", "addressCountry": "GB" }, "geo": { "@type": "GeoCoordinates", "latitude": 53.8156, "longitude": -1.5831 }, "openingHoursSpecification": [{ "@type": "OpeningHoursSpecification", "dayOfWeek": ["Monday", "Tuesday", "Wednesday", "Thursday", "Friday"], "opens": "08:30", "closes": "17:30" }], "areaServed": "Leeds, West Yorkshire, England" }

Note addressRegion and addressCountry. They are the difference between a model placing you in Leeds, England and Leeds, Alabama. Town names collide constantly — Boston, Birmingham, Richmond, Perth, Hamilton and Mansfield all exist on more than one continent — and the markup is the cheapest place to settle the question permanently.

areaServed is inherited from Organization and is not one of Google's listed properties for this type, but it is valid schema.org and it states something a service business usually leaves implied. If you travel to customers rather than receiving them, say so there.

Organization: who you are across the web

Organization has no required properties at all. Google's position is to add as many relevant ones as you can, and it uses them to tell your organisation apart from others with similar names. The property that does the most work is sameAs: the links out to your profiles elsewhere.

{ "@context": "https://schema.org", "@type": "Organization", "name": "Thornes Joinery", "legalName": "Thornes Joinery Limited", "url": "https://example.com/", "logo": "https://example.com/logo.png", "description": "Bespoke joinery and fitted furniture for homes in Mansfield, Nottinghamshire, England.", "foundingDate": "2009-04-01", "vatID": "GB123456789", "sameAs": [ "https://www.linkedin.com/company/example", "https://uk.trustpilot.com/review/example.com", "https://www.checkatrade.com/trades/example" ] }

sameAs is where structured data meets the thing that actually moves the needle. A model summarising a sector leans heavily on what third parties say about you, and sameAs is how you tell it which third-party profiles are yours. It does nothing if those profiles are empty, and a great deal if they are not.

Service: what you sell, one page at a time

Service is not a Google rich result type, so nothing visible will happen in Search. It earns its place anyway, because it attaches an unambiguous machine-readable statement of what a given page is about to the page itself. Put one on each service page, not a pile of them on the homepage.

{ "@context": "https://schema.org", "@type": "Service", "name": "Sash window restoration", "serviceType": "Sash window restoration", "description": "Draught-proofing, cord replacement and repair of timber sash windows in period properties.", "provider": { "@type": "LocalBusiness", "name": "Thornes Joinery", "url": "https://example.com/" }, "areaServed": { "@type": "AdministrativeArea", "name": "Nottinghamshire, England" } }

serviceType, provider and areaServed are the three that matter: what, who, where. If the page itself does not state all three in ordinary sentences as well, fix the page first. Markup that describes something the reader cannot see is against Google's guidelines and, more to the point, is a strong hint that the page is too thin to quote.

FAQPage: still useful, for a different reason now

The honest update: FAQ rich results are gone. Google restricted them to well-known government and health sites in 2023, removed them from Search entirely on 7 May 2026, and the type no longer appears in its rich results gallery. Anyone still selling FAQ schema as a way to win extra space in Google is selling something that does not exist.

The markup is not deprecated, though, and the structure remains genuinely easy for any parser to consume: a question, and a complete answer bound to it.

{ "@context": "https://schema.org", "@type": "FAQPage", "mainEntity": [{ "@type": "Question", "name": "Do you charge for a quotation?", "acceptedAnswer": { "@type": "Answer", "text": "No. Quotations are free and include a site visit anywhere within 20 miles of Mansfield, Nottinghamshire." } }] }

Every question must appear on the page, with that same answer visible to a reader. The value is in the discipline as much as the markup: writing six real questions with complete answers is one of the more reliable ways of making a page quotable.

Review and AggregateRating: the ones with rules

Review requires author with a real name, reviewRating with a ratingValue, and itemReviewed unless it is nested inside the thing being reviewed. AggregateRating requires ratingValue plus at least one of ratingCount or reviewCount. bestRating and worstRating are recommended and default to 5 and 1.

{ "@type": "AggregateRating", "ratingValue": 4.8, "reviewCount": 127, "bestRating": 5, "worstRating": 1 }

Read Google's rule before you add this to your own site. Pages are not eligible for the star review feature if the entity being reviewed controls the reviews about itself, and ratings must be sourced directly from users. Organization review markup is supported for sites that review other organisations, not for a business rating itself. Invented review data is worse than nothing: Google warns that reviews or ratings not by actual users may result in a manual action.

In practice that means your own testimonials page should not carry AggregateRating hoping for stars. Point sameAs at the platforms that do hold your genuine reviews and let those stand as the corroboration.

The three mistakes that make markup worthless

  • Marking up content that is not on the page. Google's policy is blunt: do not mark up content that is not visible to readers of the page, and your structured data must be a true representation of the page content. A plugin that injects an FAQPage block with answers nobody can read is a liability.
  • Invented review data. A rounded 4.9 from an unspecified number of reviewers, or a ratingValue typed into a theme setting. It risks a manual action, and on a page about why you should be trusted it is exactly the wrong lie to be caught in.
  • Stale markup contradicting the page. Opening hours changed on the page but not in the JSON-LD. A phone number that moved. A service you dropped. Contradiction is worse than silence, because it gives a machine two answers and no way to choose.

All three have the same root cause: markup maintained separately from the content it describes. Generate it from the same source your page renders from, and the drift becomes impossible rather than merely unlikely.

How to check what you have

View source and search for application/ld+json. Run the page through Google's Rich Results Test and the Schema Markup Validator, which checks against schema.org rather than Google's subset. Then do the part the tools cannot: read the JSON-LD alongside the rendered page and confirm every single value is both true and visible.

Check your templates rather than one page. Markup is usually generated, so an error on one service page is an error on forty, and a correct homepage tells you very little.

Where this sits in the order of work

Structured data is a finishing move, not a first one. If the answering crawlers cannot reach your site, the markup is irrelevant, so start with whether you are blocking them. If your service pages do not state what you do, where, and for whom in ordinary sentences, markup only describes the gap more precisely.

Done in that order, it is an afternoon's work with a real payoff: fewer wrong facts about you in circulation. To see which facts are circulating now, our free check will show you whether the models name you, and what they say when they do.

Common questions

Does schema markup make an AI assistant recommend me?

No, and anyone promising that is overselling. Structured data makes your facts unambiguous to a machine that is already reading the page. Google says explicitly that there is no special schema.org structured data you need to add for its AI features. It removes a reason not to use you; it does not create a reason to.

Is FAQPage schema still worth adding in 2026?

For rich results in Google, no. Google removed FAQ rich results from Search on 7 May 2026 and the type no longer appears in its rich results gallery. As plain machine-readable question-and-answer pairs it still has value, because a question with a complete answer attached is an easy thing for any parser to lift.

Can I add AggregateRating for my own reviews?

You can add the markup, but Google will not show stars for it. Its review snippet guidance is that the entity being reviewed must not control the reviews about itself, and that ratings must be sourced directly from users. Invented or self-collected ratings risk a manual action, which is a far worse outcome than no stars.

Which LocalBusiness subtype should I use?

The most specific one that fits, which is Google’s own stated guidance. A dental practice uses Dentist, a solicitor uses Attorney under LegalService, a plumber or electrician uses the relevant subtype of HomeAndConstructionBusiness, and a shop uses Store. Fall back to plain LocalBusiness only when nothing fits.

JSON-LD or microdata?

JSON-LD, in a script block in the head or body. Every current example in Google’s documentation uses it, it keeps the markup separate from the layout so a redesign cannot silently break it, and it is far easier to generate from a template than attributes scattered through the HTML.

See whether the assistants name you

Three real customer questions for your trade and town, no signup, about a minute.

Run my free check

Read next