Schema Markup for AI Search: the 5 Types That Matter, and What They Actually Do

# What schema actually is, in one minute

Structured data is a block of machine-readable facts about your page, written in a vocabulary search engines already agree on. It does not change what a human sees. It changes what a machine can state with confidence.

The vocabulary is schema.org, a shared standard maintained with input from Google, Microsoft, Yahoo and Yandex. The format everyone now uses is JSON-LD, a script block in your page's head. Google's own introduction to structured data is the reference implementation and worth reading before you pay anyone to install it.

The practical difference is this. Without it, an engine reads "CalnetCorp - web design in Brisbane from $2,499" and has to infer that CalnetCorp is an organisation, that Brisbane is a service area, and that $2,499 is a price rather than a phone extension. With it, you have said all three in a form that cannot be misread. That is the entire mechanism. Everything else in this guide is about what that mechanism is and is not good for.

# The correlation everybody sells you

Roughly 71% of the pages ChatGPT cites contain structured data, and about 65% of the pages Google's AI Mode cites do. Those numbers come from SE Ranking's analysis of AI citations, and they are the single most quoted statistic in the AI search industry right now.

You can see why. If you are selling an AI visibility service, "71% of cited pages have this and yours does not" is a very easy sentence to say. It is also true.

What it is not is evidence that schema causes citations. Well-maintained sites tend to have structured data. Well-maintained sites also tend to have better content, more links, clearer information architecture and someone whose job it is to care. Any of those could be doing the work. The correlation tells you that cited pages are the kind of pages that have schema. It does not tell you that adding schema makes a page cited.

That distinction stayed academic until someone ran the experiment.

# The experiment that should change how you buy this

Ahrefs identified 1,885 pages that added JSON-LD structured data between August 2025 and March 2026, matched them against 4,000 control pages that did not, and measured AI citations in the 30 days before and after. If schema drove citations, this is where it would show.

It did not. Per Ahrefs' published results, ChatGPT citations rose 2.2% and Google AI Mode 2.4%, both statistically indistinguishable from zero. Google AI Overviews went the other way, down 4.6%, described as a small but statistically significant decline. Their conclusion, in their words, is that adding schema produced no major uplift on any platform.

There is one important limit on that finding, and it cuts in your favour. The study looked at pages that were already being cited heavily. The authors say so, and they explicitly allow that schema "might still play a role in helping pages get crawled, parsed, or indexed in the first place." Nobody has tested whether schema helps a page that no engine has ever surfaced. That is the situation most Australian small businesses are actually in, and it means the honest answer for a site starting from zero is "unproven", not "useless".

Seventy-one per cent of ChatGPT-cited pages carry structured data. But when Ahrefs added schema to 1,885 pages and compared them against 4,000 controls, citations moved 2.2% on ChatGPT and fell 4.6% on AI Overviews. Schema correlates with being cited. It does not appear to cause it.

# So what does schema actually buy you?

Three things, all real, none of them "more AI citations". Being clear about which is which is the difference between spending sensibly and being sold a correlation.

1. Eligibility for rich results. This one is not in dispute, because it is mechanical. Review stars, FAQ dropdowns, price and availability, breadcrumb trails, event dates: Google's search gallery lists what each type unlocks, and none of it appears without the markup. No schema, no rich result. That is a direct, verifiable return.

2. Entity disambiguation. Schema is how you tell an engine which organisation you are, who runs it, where it operates and what it sells, in a form that does not depend on the engine guessing correctly from your prose. This matters far more than most people expect, and the next-but-one section is a live example from our own site.

3. Possibly, initial discovery. Untested, per the limit above. Worth having on that basis alone, given how cheap it is once it is done properly.

Notice that all three are worth money and none of them require the citation claim to be true. If a proposal in front of you leads with citations rather than these, the person writing it has read the 71% and not the experiment.

# The five types that matter for an Australian service business

Organization, LocalBusiness, Service, Person and FAQPage. Almost everything else is either irrelevant to a service business or a refinement of one of these five.

Organization is the entity home. It states that your business exists, what it is called, where to find it, and what other profiles belong to it. Every other type on your site should reference it rather than restating it.

LocalBusiness extends Organization with the things a local search needs: address, geographic coordinates, service area, opening hours. If you serve customers in a place, you need this.

Service describes what you actually sell, and pairs with an Offer to state the price. Publishing a price in markup is a decision, not a default, and it is the right one if you compete on transparency.

Person is the one most businesses skip and should not. It gives a named human an identity an engine can resolve, connected outward to their other profiles. On a site whose credibility rests on a named expert rather than a logo, this is not decoration.

FAQPage marks up genuine questions and answers. It is the one type with a clean, immediate payoff, because the FAQ dropdown in search results is visible and it takes vertical space a competitor cannot then occupy.

What to skip: Product markup on a service, unless you specifically want price rich results and can live with Google's merchant-listing validator asking you for shipping and returns policies you do not have. It is a defensible choice, but make it deliberately.

# Organization first, and why the homepage cannot skip it

Your homepage is the page an engine reads to decide what your business is. If it carries no structured data, that decision is being made by inference. This is the most common and most expensive gap we find, and it is common on sites that are otherwise well built.

The test takes ten seconds. Load your homepage, view source, and search for ld+json. If there are no matches, your entity home is machine-anonymous. Everything else you do in AI search is being built on top of an engine's guess about who you are.

Getting Organization right means: one node, one @id, declared once and referenced everywhere else. Name, URL, logo, the sameAs links out to your other profiles, and the founder if a named person is part of your credibility. The sameAs array is doing more work than it looks like - it is how an engine connects your site to your social profiles and confirms they are the same entity.

# The Australian specifics nobody covers

Australian businesses have an identity problem international schema guides do not address: the legal entity name, the registered business name and the brand are frequently three different strings. An engine that finds all three without being told they are the same thing may conclude they are not.

Three practical points. Publish your ABN and keep it identical everywhere. It is the closest thing Australia has to a unique business identifier, it is publicly verifiable through ABN Lookup, and it resolves ambiguity no amount of prose can. Decide which name is canonical and use it as your Organization name, with the others as alternateName rather than as competing entities. State your service area honestly using areaServed - suburbs and regions you genuinely cover, not an aspirational list, because an engine that finds you claiming forty suburbs and reviewing in one will trust the reviews.

This matters more here than the population would suggest. ABS data on business use of information technology shows how much of the Australian small business sector runs on a single digital presence. When that one presence is ambiguous, there is nothing else for an engine to fall back on.

# What we found on our own site this week

Two structured data faults on our own money pages, both of the kind that no ranking report would ever surface. We publish these because a schema post written by someone claiming a clean sheet is not worth reading.

The first: our /website-in-a-day page declared the business as #org, while eighteen other pages declared the same business as #organization. Two identifiers, one company. An engine following those references has no reason to conclude they are the same entity, which is precisely the confusion structured data exists to prevent. We had split our own entity in half.

The second was worse and more subtle. Our /ai-search-foundation page named a seller by reference - and the node it referenced was defined nowhere on the page. Google Search Console reported the seller as a bare, untyped thing, which is exactly what it was: a pointer to nothing. The page had no Organization markup at all.

Both are now fixed. The point is not that we made mistakes; it is the category of mistake. Neither showed up as a ranking drop, a traffic dip, or a failing audit score. A dangling reference is silent. You find it by reading your own markup as a machine would, or by reading the structured data report in Search Console, and almost nobody does either until something is already wrong.

# How to check yours in ten minutes

Three free tools, in this order, and you will know more about your structured data than most agencies will tell you.

Google's Rich Results Test. Paste a URL. It tells you which types were detected and which are eligible for rich results, and it flags missing required fields. Start with your homepage, then your highest-value service page.

The Schema Markup Validator. Broader than Google's tool because it validates against the full vocabulary rather than only what Google supports. This is the one that catches a malformed node Google silently ignores.

Search Console's structured data reports. The only one of the three that tells you about pages you have not thought to check. It is also where a dangling reference or an invalid object type surfaces, which is how we found ours.

If you would rather work through a structured list than poke at tools, the AI Search Readiness Kit is the same framework we run internally, and structured data is one of its three scored pillars.

# The five mistakes that make schema worthless

Schema is easy to install and easy to install in a way that does nothing. These five account for most of what we find.

1. Marking up something the page does not say. Structured data has to describe visible content, and Google's structured data policies are explicit about it. Review stars for reviews that are not on the page, an FAQ nobody can read: this is a policy violation, not a shortcut, and it risks a manual action.

2. Dangling references. An @id pointing at a node that does not exist. Silent, common, and the exact fault we found on our own site.

3. Inconsistent identifiers. The same entity declared under different @id values across pages. Also ours, also silent.

4. A homepage with nothing. Service pages marked up beautifully while the entity home carries no markup at all. Backwards, and very common.

5. Installing it and never looking again. Prices change, offers retire, staff leave. Structured data that states a retired price is worse than none, because you have now told a machine something false in a format it trusts.

If reading that list produced a specific worry about your own site, a free AI visibility audit will tell you where you actually stand before you spend anything - including what the engines currently say about your business, which is the other half of the picture.