Does schema markup actually help you show up in AI search? An honest look at the evidence

Schema markup helps Google understand your business and can earn rich results, but there is no reliable evidence it directly lifts you in AI Overviews or ChatGPT, and some AI pipelines strip the JSON-LD before the model reads the page. Do schema well because it disambiguates your entity and feeds Google’s Knowledge Graph. Do not treat it as an AI-visibility lever, because it is plumbing, not a magic citation switch.

This page answers one narrow question: does adding schema actually change whether you show up in AI search? It is deliberately not a how-to. For which types to use and how to mark up a service-area business correctly, see the LocalBusiness schema guide. For the wider picture of how AI engines pick local businesses, see AI search for local businesses. Here we stick to the does-it-help debate and the evidence behind both sides.

Figures current as of 17 July 2026. AI-search behaviour changes month to month, so each figure below is attributed to a named source with its own date. Re-check them before you act on them.

Does schema help AI Overviews?

No, not directly. Google’s own documentation says there is no special markup that gets you into AI Overviews or AI Mode, and structured data is not required for either. Schema helps you indirectly by making you eligible for rich results in ordinary Search and by clarifying your entity for the Knowledge Graph, which is one of the inputs Google’s AI features draw on. So the benefit is real but second-hand.

Google’s AI features documentation (last updated 10 December 2025) states plainly that “there’s also no special schema.org structured data that you need to add” to appear in AI Overviews and AI Mode, and that the same people-first fundamentals that win in Search are what win in AI features. The same document still recommends structured data as part of an overall strategy because it keeps you eligible for rich results. Read those two statements together and the position is consistent: schema is worth doing, but not as an AI growth hack. You can read it yourself at developers.google.com/search/docs/appearance/ai-features.

This matters because the hype pieces often imply a direct line from JSON-LD to an AI citation. There is no documented mechanism for that on Google’s side. What schema does is help machines parse who you are, what you do and where you do it, so the indirect path runs through better understanding, not a dedicated AI feed.

Do LLMs even read your JSON-LD?

Often they do not. Many AI extraction and HTML-to-markdown pipelines convert a page to plain text before the model sees it, and that conversion routinely discards the JSON-LD block sitting in the page head. So markup you can see in the source is frequently gone by the time an answer engine reads the content. This is the detail most “schema for AI” posts skip.

An independent experiment by the AI-search monitoring tool Otterly.ai, published 23 March 2026, tested this directly by asking each major answer engine to fetch and report the schema markup from specific URLs. Their finding: 6 of 7 platforms were unable to fetch or correctly interpret the schema when asked. Only Gemini returned the JSON-LD accurately. Google’s AI Mode partially read it but hallucinated, inventing a “Service” schema that was not on the page. Their conclusion was blunt: schema is an SEO lever, not a GEO growth lever. The write-up is at otterly.ai/blog/schema-markup-real-impact-ai-search.

The same experiment did record a large rise in AI Overview citations after schema was added, but it attributed that to schema lifting ordinary rankings and rich-result eligibility, not to any AI reading the JSON-LD directly. That is the honest shape of the indirect benefit: the markup helped through Search, not through a back channel into the answer engine.

That is one experiment, not a law of physics, and pipelines differ. There is a credible counter-signal too. Microsoft’s Fabrice Canel, a principal product manager on Bing, said at SMX Munich in March 2025, as relayed by attendees rather than in a verbatim transcript, that schema markup helps Microsoft’s large language models, which power Copilot, understand content. As of mid-2026 that remains the only first-party, on-the-record statement from a major answer engine that schema feeds its models. So the honest summary is mixed: at least one engine says it uses schema, most independent tests cannot show a direct lift, and several pipelines drop it before reading.

Where that leaves you: the facts that you most want an AI engine to repeat about you, your trades, your towns, your hours, your phone number, should be in the visible page text, not hidden in markup the parser may throw away. Schema can mirror those facts. It should not be the only place they live.

What does schema actually do for local AI visibility?

For a local trade, clean LocalBusiness plus Service plus accurate areaServed markup, all matching the visible text, helps machines parse who you are, what you fix and where you cover. That is genuine value: it disambiguates your entity and keeps your name, address, area and services consistent across the systems that feed Google’s Knowledge Graph, which is also where your rich results come from. What it does not do is rescue a thin page. A bare service-area page with immaculate schema still loses, because the schema is describing content that was not worth citing in the first place.

Picture a boiler-repair firm covering Uckfield, Lewes and the villages between. The useful version of their schema looks like this:

• LocalBusiness with the real trading name, the one genuine business address and the one real phone number, never a fake street address invented per town.
• Service describing the actual job, “boiler repair”, tied to areaServed for the towns and parishes they truly cover.
• areaServed and contact details that match the visible service-area text on the page word for word, so a reader and a parser see the same thing.
• No self-serving review stars, because Google has not shown review rich results from a business’s own markup since 2019, and adding them invites a manual action.

The order matters. Write the page a Lewes homeowner would find useful, then mark up what is already there. Schema that describes a real, specific page is hygiene worth having. Schema bolted onto a near-empty template is a liability, because mismatched or invisible markup risks a structured-data manual action. This is the same reason near-duplicate location pages fail on their own merits, covered in duplicate content on location pages.

So is there a “schema for AI” trick worth chasing?

There is no separate AI-only schema, no llms.txt hack and no chunking trick that substitutes for being genuinely useful and indexable. Google’s guidance is that the work which earns AI visibility is the same people-first SEO that has always mattered: genuinely helpful content on technically healthy pages, with clear and consistent signals about who you are. Schema is one of those signals, not a shortcut around them.

The practical test is whether your trust and expertise are demonstrable on the page itself, the part of E-E-A-T that AI engines and human readers both reward. We cover that for trades in E-E-A-T for local businesses. And if you are wondering how much of the page can be AI-assisted before it tips into the thin, scaled territory Google penalises, that line is drawn in how much AI content is acceptable.

Where the Townsmith engine fits

Townsmith builds real, editable location pages, scores each one from 0 to 100, and flags the thin ones before you publish through an advisory pre-publish gate. It emits the LocalBusiness, Service, areaServed and BreadcrumbList markup the page actually warrants, generated server-side from your own facts with no AI and no external calls in the free plugin. The point is the order described above: substance first, then schema that mirrors the visible text exactly. The optional Pro add-on can turn that advisory gate into one you enforce, so weak pages cannot go live until they clear the threshold.

That is deliberately the modest claim. The engine produces page-accurate schema and tells you which pages are too thin to publish. It does not promise an AI citation, because nothing honestly can. If you want to see how the scoring and schema output work, the features overview and the documentation describe it without the hype. Townsmith complements Rank Math, Yoast, AIOSEO and SEOPress rather than replacing them, so your existing SEO plugin keeps owning the base graph.

Common questions

Will adding schema get me into AI Overviews?

Not on its own. Google states there is no special markup required for AI Overviews or AI Mode and no AI-specific schema to add. Schema can help indirectly by making you eligible for rich results and by clarifying your entity for the Knowledge Graph, but it is not a direct route into an AI answer.

If schema might be stripped, should I bother with it at all?

Yes, for ordinary Search. Schema still earns rich results, disambiguates your business and keeps your details consistent across systems. Treat it as extraction hygiene for Google Search, and make sure the same facts appear in the visible page text so an AI engine can read them even when the JSON-LD is discarded.

Should I add review stars to my own pages to stand out in AI results?

No. Google has not shown self-serving review rich results from a business’s own markup since 2019, and adding AggregateRating to your own pages can trigger a structured-data manual action. Star ratings still come from Product markup and from third-party platforms such as Google, Yelp and Trustpilot, not from rating yourself.

Does more schema mean better AI visibility?

No. Piling on schema types does not raise your odds of being cited, and markup that does not match the visible content is a risk rather than a benefit. Clean, accurate LocalBusiness and Service markup that mirrors a genuinely useful page is the whole job. The page has to earn the citation first.

What is the single highest-value thing I can do instead?

Write a page a real customer in your town would find useful, with your specific services, areas and proof of experience in plain text, then mark up what is already there. Good people-first SEO is the optimisation. Schema is the hygiene layer on top of it, not a replacement for it.

About the author

Written by Stephen Evans, who builds the Townsmith Local Pages Engine. The claims above link to their primary sources so you can check them, and the figures are dated because AI-search behaviour keeps moving. More about the project is on the about page.