Structured Data for AI Search in 2026: Does Schema Help?

Structured data for AI search is schema.org markup, usually written as JSON-LD, that states your business facts in a form machines can read without guessing. It helps engines understand who you are and connect your website to your other profiles, but Google says its AI features need no special schema, and no engine documents markup as a route to citations.

This article is part of our complete guide: AI Search Optimization.

• Google needs 0 special schema.org types for AI Overviews and AI Mode, yet calls structured data worth keeping for rich results (Google Search Central's guide to generative AI search).
• Google's AI features page names 1 structured data practice among its SEO basics: the markup must match the visible text on the page (Google's AI features documentation).
• Google accepts 3 formats, JSON-LD, Microdata and RDFa, recommends JSON-LD, and does not guarantee that correct markup shows in results (Google's general structured data guidelines).
• In November 2025, Microsoft Bing wrote that schema-marked product pages, FAQs and comparison tables help AI systems interpret content (Bing Webmaster Blog on AI search and conversions).
• FAQ rich results stopped appearing in Google Search on 7 May 2026, so FAQ markup no longer changes how a result looks (Google Search Central's documentation updates).

What does structured data actually tell an AI engine?

Structured data tells an engine what each fact on a page is: your business name, your address, your logo and your profiles on other websites.

A human reader sees "Open 7am to 4pm, closed Sundays" and understands it at once. A machine reading the same line has to work out that it describes opening hours, for which business and at which location. JSON-LD removes that guesswork by repeating the facts in a fixed vocabulary, so an engine reads "openingHoursSpecification" and knows exactly what the values mean.

The vocabulary comes from schema.org, the shared set of types that Google's documentation is built on, and the markup sits in the page's code where visitors never see it. Google describes structured data as a standard format for giving information about a page and classifying its content, and says it uses the markup it finds to understand page content and to gather information about the wider world, including companies. That second use is the one that matters for AI answers: an engine that is sure which business a page belongs to can attach the right facts to the right name.

Think of it as labelling rather than persuading. Markup cannot make a page more useful, more trusted or more quotable than its visible text already is. What it can do is remove doubt about identity, which matters most when your business shares a name with another one, trades in several places or has changed its address or hours.

What do Google and Bing say about schema and AI answers?

Google says structured data is not required for its AI features and no special markup exists for them, while Bing says schema-marked content helps AI systems interpret pages.

Google's guidance is the most detailed any engine has published. Its guide to generative AI search says structured data is not required and there is no special schema.org markup to add, but that it is a good idea to keep using it as part of overall SEO because it helps with eligibility for rich results. The same guide says Google Search does not use AI text files, extra markup or Markdown versions of pages, and that a page only needs to be indexed and eligible to show with a snippet to be considered.

Google adds one condition that applies directly to AI work. Its AI features documentation lists making sure your structured data matches the visible text on the page as an SEO practice that still applies to AI Overviews and AI Mode. Markup that claims facts the page does not show is therefore not a hidden advantage; it is a mismatch Google's own guidance tells you to avoid.

Microsoft goes slightly further in its wording. In a November 2025 post on the Bing Webmaster Blog, Fabrice Canel and Krishna Madhavan wrote that structured content such as schema-marked product pages, FAQs and comparison tables helps AI systems interpret and summarise content. They link clearer interpretation with a better chance of being summarised and cited, which is a different claim from a ranking boost, and Bing publishes no list of types that Copilot prefers.

The other engines publish nothing on the subject. OpenAI's crawler documentation explains OAI-SearchBot, GPTBot and ChatGPT-User and how robots.txt controls them, and Perplexity's crawler documentation covers PerplexityBot, Perplexity-User and firewall settings, but neither page mentions structured data. Anything you read about ChatGPT or Perplexity rewarding particular schema types is someone's observation or guess, not documented behaviour.

Why is schema markup not a shortcut to AI citations?

Schema markup is not a shortcut because AI engines quote visible passages that answer a question, and markup holds facts, not answers.

When ChatGPT, Perplexity or Google AI Mode writes a reply, it cites pages whose text supports what it says. A JSON-LD block can confirm that a bakery is in Leeds and opens at 7am, but it cannot explain which sourdough suits a toaster or how far ahead to order a wedding cake. Those answers live in your paragraphs, headings and tables, so a page with perfect markup and thin text has nothing worth quoting.

Google is also explicit that markup guarantees nothing. Its general guidelines say Google does not guarantee that structured data will show up in search results, even when it is valid, and its AI guidance repeats that indexing and serving are never guaranteed. If valid markup does not even guarantee a rich result, it cannot be relied on to place a business inside an AI answer.

Studies from SEO vendors claiming that pages with schema are cited more often appear regularly, but they are small, their samples are chosen by the vendors, and engines change their systems faster than the studies are repeated. Owners often observe that well-marked websites get cited, yet those websites usually also have clear answers, consistent listings and mentions elsewhere, so the markup alone cannot be credited. The honest position is that schema supports understanding and the rest of the page earns the citation.

Which schema types matter most for a business?

Organization or a LocalBusiness subtype on the homepage matters most, followed by Article on guides, Product on product pages and BreadcrumbList across the website.

The table below sets out what each common type states and why it still earns its place in 2026. The order reflects how directly each type describes the business itself, which is the fact AI answers most need to get right. A website does not need every type: a plumber with five service pages needs far less than an online shop with five hundred products.

Type	Where it goes	What it states	Why it is worth keeping
Organization	Homepage or one About page	Name, website, logo, contact details, sameAs profiles	Ties the business to its official profiles and can set the logo Google shows
LocalBusiness subtype (Bakery, Plumber, Dentist)	Homepage or each location page	Address, phone, opening hours, map coordinates	States the facts local questions depend on; Google requires name and address
Article or BlogPosting	Each guide or post	Headline, author, publication and update dates	Makes the author and the date of a claim explicit
Product	Each product page	Name, price, availability, reviews	Supports shopping results and keeps stock facts current
BreadcrumbList	Every page	Where the page sits in the website	Shows how pages relate to each other
FAQPage	Pages with a visible FAQ	Questions and answers already on the page	No rich result since May 2026, so keep it only if it matches the visible FAQ

Two cautions apply to this table. Google's local business documentation recommends review and rating markup only for websites that collect reviews about other businesses, so a bakery should not mark up its own testimonials in its LocalBusiness block. And FAQPage markup is now purely descriptive in Google: the visible FAQ still gives engines self-contained answers to lift, which is where its value lies.

Two narrower questions follow from this one and each has its own guide. The first is how to add Organization schema with sameAs links, the properties to fill, where the block goes and how to test it. The second is which LocalBusiness schema helps a local business in AI answers, from choosing the right subtype to matching your opening hours with your Google Business Profile.

How should structured data match your page and your listings?

Structured data should repeat exactly what the page shows and what your listings say, because a mismatch is a reason for an engine to doubt every source.

Google's general guidelines set the rule plainly: the markup must be a true representation of the page content, and you should not mark up content that readers cannot see. A LocalBusiness block that lists Sunday hours the page never mentions, or a service area wider than the one in your text, breaks that rule. Spammy or misleading markup can lead to a manual action, which Google says removes a page's eligibility for rich results, so the risk is real even though rankings are not affected.

Consistency outside the page matters just as much. AI answers about a business draw on its website, its Google Business Profile, Bing Places, directories and press coverage, and when these disagree an engine has to choose one. A plumber in Nairobi who moves premises and updates the website but not the profile can find an AI answer giving the old address, whatever the markup says.

The practical fix is to keep one master list of your facts: legal name, trading name, address, phone, email, hours, services and profile links. Change every copy on the same day whenever one fact changes, starting with the visible page, then the markup, then the profiles. Markup that is generated from the same fields as the visible page, as many website builders and content systems do, removes the risk of the two drifting apart.

How do you add and test structured data?

You add structured data as a JSON-LD script in the page's code, then test it with Google's Rich Results Test and fix every error before relying on it.

Most content systems handle the writing for you. Popular WordPress SEO plugins, many Shopify themes and many website builders output Organization, Article and Product markup from fields you already fill in, so the first task is to find out what your website prints today rather than to add more. Two plugins that both print an Organization block can produce conflicting copies, which is worse than one clean block.

Google's Rich Results Test checks a live address or a pasted snippet, reports valid items, warnings and errors, and supports 31 rich result types including Organization and Local business. It tests as a smartphone by default and shows whether robots.txt let it fetch the page, which also catches a page Google cannot reach at all. For schema.org types Google does not use for rich results, the Schema Markup Validator at validator.schema.org checks the syntax against the full vocabulary.

Testing once is not enough. Prices, opening hours and staff change, and markup that was right at launch can be wrong a year later without anyone noticing. Put a check of the markup on the same list as your yearly review of hours and contact details, and retest any page whose facts change.

Where does structured data fit in an AI search plan?

Structured data belongs after the two jobs that decide citations: letting the engines' search bots reach your pages and opening each page with a clear answer.

A page that OAI-SearchBot, Claude-SearchBot or PerplexityBot cannot fetch will never be cited, however good its markup. The same is true of a page whose text never answers a question directly, because there is no passage to quote. Only when both are in place does markup add value, by making sure the engine attaches the quoted answer to the right business.

You can see where your pages stand before touching any code. Scan your website with the AI Search Visibility Checker to find out whether each engine's search bot is let in, which question each page can win and which passage an AI would quote. Each page gets an AI Citation Score out of 100 across five areas, and entity and trust is one of them at 15 points, alongside reach, the answer, evidence and specific content.

For the classic search side of the same work, the AI Website SEO Tools check includes a Sharing and rich results area among its six. Running both shows whether a weak result comes from access, content or markup. That saves you from adding schema to fix a problem it cannot solve, such as a blocked bot or a page with no clear answer.

Questions business owners ask about schema and AI search

Do I need schema markup to appear in Google AI Overviews?

No. Google says structured data is not required for its generative AI features and there is no special schema to add. A page needs to be indexed and eligible to show with a snippet, and its markup, if any, should match the visible text.

Does ChatGPT read JSON-LD?

OpenAI does not say. Its crawler documentation covers which bots exist and how robots.txt controls them, not how ChatGPT search reads a page's code. Put every important fact in the visible text as well, so it is available whatever an engine reads.

Is FAQ schema still worth adding in 2026?

Only where it matches a real FAQ on the page. Google stopped showing FAQ rich results on 7 May 2026, so the markup no longer changes your result, but Bing names FAQs among the structures that help AI systems interpret content. The visible questions and answers matter more than the code.

Can wrong structured data harm my website?

Wrong markup rarely harms rankings, but it can cost you features and spread errors. Google can remove rich result eligibility for misleading markup, and outdated hours or addresses in markup add one more conflicting source for an engine to weigh. Fix or remove markup you cannot keep accurate.

Which format should I use for schema markup?

Use JSON-LD where your website allows it. Google supports JSON-LD, Microdata and RDFa and recommends JSON-LD because it is easiest to add and maintain. It sits in its own script block, so editing the visible design does not break it.

Your next step: check the facts before the markup

Start by writing down the facts every engine should know about your business: name, address, phone, hours, services and the profiles that belong to you.

Compare that list with your homepage, your Google Business Profile and your main directory listings, and correct any that disagree before you look at code. Then test your homepage in the Rich Results Test to see which markup already exists, add Organization or the right LocalBusiness subtype if it is missing, and retest whenever a fact changes. When the markup is clean, spend the rest of your time on the visible answers, because those are what engines quote.