What schema markup does for AI search, and what it doesn't

Schema markup will not get ChatGPT to cite your business, and Google has said in writing that its AI features do not need it.

That is the short version. The longer version matters because schema still does two useful jobs, and agencies are still selling it as a third job it has never been shown to do.

We have spent most of 2026 being asked some form of "do we need schema for AI?" by Northampton businesses who have read that structured data is the key to AI visibility. So we went back to what Google publishes, what Google staff have said in public, and the handful of tests that measured anything.

This article separates those three things. It ends with the two schema types we think a UK small business should actually add, and a copy-and-paste JSON-LD example for each.

What schema markup does for Google, with evidence

Schema markup earns rich results and helps Google tell your business apart from others with a similar name, and both of those are documented. Google's Organization structured data page, updated 8 September 2026, says the markup helps Google "better understand your organization's administrative details and disambiguate your organization in search results". It also controls which logo appears in your knowledge panel.

The LocalBusiness page is similarly specific. It lists knowledge panels, business carousels and opening hours as the features the markup can feed. Only name and address are required. Everything else, including openingHoursSpecification, geo and priceRange, is recommended.

Article markup is the third one with a documented payoff. Google says it can improve how titles, images and dates show in Search and Google News, and it tells Google who wrote the piece through author.name and author.url. The guidance is strict: the name field holds the name only, with no job title or "posted by".

These are modest, visual benefits. None of them is a ranking signal. Barry Schwartz has been recording Google saying so for years, and nothing in the 2026 documentation changes that.

What Google says about schema and its AI features

Google's own guide on generative AI features says structured data is not required and there is no special markup to add. The guide, Optimizing your website for generative AI features on Google Search, was published on 15 May 2026 and updated on 10 July. The line is clear: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add."

The same page says you do not need to "create new machine readable files, AI text files, markup, or Markdown to appear in Google Search", and lists llms.txt files among the things to skip. A 15 June entry in Google's documentation changelog confirms Search ignores llms.txt entirely.

Google does add that structured data is still "a good idea" as part of your overall SEO. That is the Organization and LocalBusiness benefit from the previous section, nothing more.

John Mueller gave a more personal answer on Reddit in January 2026, which Search Engine Roundtable wrote up. Asked whether extensive schema helps large language models understand an entity, he said "yes, no, and it depends", and that some features "thrive with structured data" while many types are wishful thinking. His sharpest line was aimed at the hype:

"Your 'best geo insurance comparison site' isn't going to rank better by adding insurance markup."

John Mueller, Google, on Reddit, via Search Engine Roundtable, January 2026

He flagged the whole reply as personal opinion, not official guidance. Even so, it matches the documentation.

What the published tests found

The largest test to date found that adding schema did not move AI citations in any useful direction. In May 2026 Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026, matched each against three control pages on other domains with similar citation levels, and measured citations 30 days either side.

Source: Ahrefs, May 2026.

The +43% figure is the one to remember, because it shows how a bad study gets written. Pages that added schema did get 43% more AI Mode citations afterwards. So did the control pages that added nothing. Once you subtract the controls, the gain is 2.4%, which Ahrefs says is noise.

Ahrefs also found that AI-cited pages were almost three times as likely to carry JSON-LD as uncited pages, across 6 million URLs. That correlation is what most "schema helps AI" claims rest on. Well-run sites tend to have schema and tend to get cited, and the controlled test shows the markup itself is not why.

The caveats are honest ones. Every page in the sample already had 100 or more AI Overview citations, so the test says nothing about a page nobody has heard of. All schema types were pooled. The window was 30 days.

The smaller tests point the same way. In September 2025, Julio Guevara built two pages for a product that Gemini and ChatGPT had never seen, one with visible text plus schema, one with schema only. He ran extraction prompts hundreds of times. His verdict: "this only worked on the page with information visible as text".

Mark Williams-Cook ran a stranger experiment, written up by Search Engine Journal in June 2026. He put a fictional company address only inside a deliberately invalid JSON-LD block, with a made-up @type and properties that do not exist. ChatGPT and Perplexity both returned the address.

His point: the models were reading it as text in the page source, "not parsing it as schema". Valid markup earned nothing that broken markup did not.

The one test that went the other way

Search Engine Land's three-page experiment is the strongest evidence for schema helping AI Overviews, and its author does not call it proof. Molly Nogami built three near-identical single-page sites: one with well-implemented schema, one with poor schema, one with none. Only the well-implemented page appeared in an AI Overview, at position three for its query.

The no-schema page was never indexed at all, which the article cannot explain. With three pages and one run, Nogami writes: "We don't consider these results to be absolute proof that well-implemented schema plays a role in AI Overview presence." An earlier analysis by the same team of 100 healthcare sites found only a slight, non-significant correlation.

Bing is the one engine that has said schema helps its models. Fabrice Canel of Bing told an SMX Munich audience that schema helps Microsoft's LLMs understand content, as Search Engine Land reported. There is no public figure attached to that, and no equivalent statement from OpenAI, Anthropic or Perplexity.

Why this matters for a UK business

A UK business owner is now being sold schema with "AI visibility" on the invoice, and the evidence does not support the price. Ofcom's December 2025 report on how the UK goes online found about 30% of searches now show an AI Overview, 53% of adults say they see them often, and ChatGPT had 1.8 billion UK visits in the first eight months of 2025.

That audience is real, so the pressure to "do something about AI" is real. Schema is an easy thing for an agency to invoice because it is visible in the source code and it can be validated with a free tool. Zach Ball of Page One Power put it bluntly in CMSWire: your next AI budget should not fund schema.

The lever that does show up in the data is the one Google's guide calls non-commodity content, plus mentions of your business on other sites. Schema does not create either. Those two levers are what the survey of 45 generative engine optimisation studies found had the most support, and schema was not among them.

The two schema types worth doing anyway

A UK SME should add LocalBusiness (or Organization) with sameAs links, and Article with a Person author, and then stop. Both have a documented Google benefit, both take under an hour on most CMSs, and both are the kind of plain fact-stating markup that Mueller said machines can use.

LocalBusiness or Organization with sameAs

Use LocalBusiness if customers visit you, Organization if they do not. Google recommends the most specific subtype that fits, so a dentist would use Dentist and an accountant AccountingService. Put it on the homepage or about page once, not on every page.

The sameAs property is where the entity work happens. It points Google at your Companies House listing, your Google Business Profile, LinkedIn and any directory that already knows you. This is what helps Google decide that "Smith & Co" in Northampton is you and not the one in Norwich.

{
  "@context": "https://schema.org",
  "@type": "AccountingService",
  "name": "Example Accountants Ltd",
  "url": "https://www.example.co.uk/",
  "logo": "https://www.example.co.uk/logo.png",
  "telephone": "+441604000000",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "1 Gold Street",
    "addressLocality": "Northampton",
    "postalCode": "NN1 1AA",
    "addressCountry": "GB"
  },
  "openingHoursSpecification": {
    "@type": "OpeningHoursSpecification",
    "dayOfWeek": ["Monday", "Tuesday", "Wednesday", "Thursday", "Friday"],
    "opens": "09:00",
    "closes": "17:30"
  },
  "sameAs": [
    "https://find-and-update.company-information.service.gov.uk/company/00000000",
    "https://www.linkedin.com/company/example-accountants",
    "https://maps.google.com/?cid=000000000000"
  ]
}

Every value must match what the page shows. Google's Article guidance and the LocalBusiness guidance both say so, and the review-star rule is stricter still: a business cannot mark up its own reviews on its own site and expect stars.

Article with a named Person

Article markup on blog posts tells Google who wrote the piece and when, which is the only authorship signal Google documents. The author.url should go to a page that "uniquely identifies the author", so a bio page on your own site or a LinkedIn profile. Google says sameAs on the Person works as an alternative.

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "What schema markup does for AI search, and what it doesn't",
  "datePublished": "2026-10-24T09:00:00+01:00",
  "dateModified": "2026-10-24T09:00:00+01:00",
  "image": "https://www.example.co.uk/images/schema-ai.jpg",
  "author": {
    "@type": "Person",
    "name": "Vijay Rathee",
    "url": "https://www.example.co.uk/about/vijay-rathee",
    "sameAs": ["https://www.linkedin.com/in/example"]
  },
  "publisher": {
    "@type": "Organization",
    "name": "Example Accountants Ltd",
    "logo": {
      "@type": "ImageObject",
      "url": "https://www.example.co.uk/logo.png"
    }
  }
}

Include a timezone in the dates. Google says that without one it assumes Googlebot's timezone, which is not yours.

Schema type against confirmed benefit

Most schema types have no documented benefit at all in 2026, and two that agencies still recommend have been switched off. This table only lists what we could trace to a Google document or a controlled test.

"None measured" means exactly that. Nobody has published a controlled test showing a lift, and the one large test found nothing.

What to do this month

If you have a website and no schema, the order of work is short, and it sits inside the same technical SEO afternoon as your other fixes.

1. Add LocalBusiness or Organization markup to your homepage with at least three sameAs links, and check it in Google's Rich Results Test.
2. Add Article markup with a Person author to your blog template, with a bio page for each author.
3. Remove any FAQPage or HowTo markup that a plugin added, unless the questions are also visible on the page word for word, in which case leave it and expect nothing.
4. Put every fact you want an AI engine to repeat, including prices, addresses and opening hours, in visible HTML text. That is the only place every test agrees the models read.
5. Spend whatever budget was earmarked for "AI schema" on a page only your business could have written.

We add the first two to every site we build as part of the technical SEO work, and we do not sell the rest. If an agency proposes a schema project as the route to AI visibility, ask them which controlled test they are relying on. If the answer is the Ahrefs correlation, they have not read past the headline.

The decision rule we use is this: markup is worth adding when a Google document names the feature it feeds, and not otherwise. Schema that clarifies who you are is cheap and documented. Schema sold as a citation lever is neither.

If you want a second opinion on what your site is sending to Google and the AI engines, our generative engine optimisation audit starts with the visible page, and you can get in touch here.