Structured data and AI search — five schema markups worth doing

Structured data, meaning schema.org markup, tells a machine what the content of a page means: this is a company, this is a price, this is a question and there is its answer. This markup has been added for search engines for years, for the sake of star ratings and rich results. AI search gave the same work a new meaning: structured information is easier for a language model to pick up correctly, and information picked up incorrectly is the worst outcome in GEO. Here are the markups worth doing and the order to do them in.

Why structured data matters for AI visibility

A language model can read a page without markup, but then it has to infer: is ”from €590” the installation price or the monthly one, does the address refer to the company or its customer, who wrote the page. The inference is usually right, and sometimes it is not. Schema markup removes the guesswork by stating the relationships directly: this price belongs to this service, this organisation is located here, this person is the author.

From a GEO point of view the goal is therefore not just a ”rich result” but the accuracy of the information in answers. A company whose basic details are unambiguous to a machine gets mentioned correctly more often than one whose details have to be inferred from body text.

These markups are worth doing, in this order

1. Organization. The foundation for everything: official name, address, phone, email, logo and sameAs links to the company’s other profiles (LinkedIn, Google Business). This is the markup that binds scattered mentions into a single entity. For a local business the LocalBusiness subtype with opening hours is better still.

2. FAQPage. The single most effective markup for GEO, because it packages content into exactly the shape AI answers move in: question and answer pairs. Every genuine FAQ section is worth marking up, but only a genuine one. Invented questions nobody asks are noise to reader and machine alike.

3. Service or Product. What you sell, what it costs, where it is available. Marking up price information feels bold to many, but pricing questions are exactly what people put to an AI, and the answer goes to whoever’s price is machine-readable.

4. Article. Publication date, modification date and publisher for articles. Freshness is a selection criterion in search-based AI answers, and date markup is how you prove it.

5. BreadcrumbList. Light but useful: it states the site’s structure and the page’s place in it.

Three mistakes that waste the work

The markup lies. Schema claiming a page contains content that is not visible on it is a risk: Google can penalise it, and a contradiction fed to a language model weakens its trust in all your other data too. Mark up only what is genuinely on the page.

The same information with different values in different places. If the phone number in your Organization markup differs from the one in your footer or your Google Business profile, the machine has to choose, and it may choose wrong. One truth, the same everywhere.

Marking up everything possible. Schema.org knows hundreds of types, and tools tempt you to use every one. The benefit comes from the five listed above; the rest are maintenance burden without return for most sites.

How to check where you stand

A quick check takes fifteen minutes. Put your home page, a service page and one article through Google’s rich results test or a schema.org validator. Look at three things: is there an Organization markup and are its details correct, do your FAQ sections have FAQPage markup, and do your articles carry dates. If the answer is yes three times, the foundation is sound and you can move on to content. If not, this is the cheapest fix on the GEO list. On a WordPress site most of the markup is handled by an SEO plugin or the theme without hand-coding.

What structured data does and does not do

Structured data describes a page’s contents in a form a machine reads without interpreting the layout. That is the whole function, and both the promise and the limits follow from it.

What it does: removes ambiguity. Without it, a service reading a page has to work out from position and formatting which text is the price, which is the author and which is the opening hours. With it, those are labelled. On a page where several numbers appear, that difference decides whether the right one gets quoted.

What it does not do: improve the content. Marking up a page that answers nothing produces a well-labelled page that answers nothing. The markup is a description of what is there, and it inherits the quality of what it describes.

The types worth implementing

Dozens of schema types exist and most companies need four or five. These carry the weight.

• Organization. Who the company is, where it is, how to reach it, its identifiers and profiles. One instance, site-wide, and the foundation for anything a service says about you.
• Article. Author, published date, updated date and headline. This is where the identity signals that decide citation actually live.
• FAQPage. Question and answer pairs labelled as such rather than inferred from formatting.
• Product and Offer. For shops, with availability and identifier populated rather than left out.
• LocalBusiness. For companies serving a physical area, with hours and address.

Adding more types beyond these has diminishing returns quickly. Getting these five right and keeping them accurate beats broad coverage that nobody maintains.

Connecting the pieces

The most common weakness is not a missing type. It is a set of correct types that do not reference each other, leaving a service with several disconnected descriptions instead of one picture.

An article should name its author, and the author should point to a page describing them. The organisation should be named as the publisher. A product should belong to the organisation selling it. These links are what let a service conclude that the same company is behind everything, rather than treating each page as an unrelated document.

In practice this means using identifiers consistently and pointing entities at each other rather than repeating names as plain text. Most SEO plugins do this reasonably well if the settings are filled in, which is more often the issue than the plugin.

Where implementations go wrong

Four failures account for most broken markup, and none of them shows up as a visible problem on the page.

The markup disagrees with the page. A price in the data that differs from the price on screen, usually because one was updated and the other was not. This is the most damaging error, because it makes the source unreliable rather than merely incomplete.

Duplicate markup from two sources. A theme and a plugin both output Organization data, with different values. Services then see two versions of the company.

Marked-up content that is not visible. A policy violation and a reason to ignore the page entirely.

Fields left at defaults. A plugin’s placeholder author name or a founding year nobody set. These are worse than absent, because they assert something false.

Checking it without special tools

Two free validators cover this. Google’s rich results test reports what Google recognises, and the schema.org validator reports what is technically valid regardless of any one company’s support.

Run one page of each type rather than the whole site: a service page, an article, a product if you have one, and the front page. Read the warnings and not only the errors, since the missing optional fields that appear as warnings are frequently the ones an answer needs.

Then check the values by eye against the page. Validators confirm the structure is correct; they cannot tell you the price in it is from last year.

How much of a difference to expect

Worth being plain about this, because structured data attracts more enthusiasm than its effect justifies.

On a page with clear headings, an obvious answer and a visible author, markup adds precision rather than transforming the outcome. On a page where the same information exists but the layout is ambiguous, it can be the difference between being parsed correctly and being parsed wrongly. And for shops, where availability and price have to be read exactly, it is closer to a requirement than an enhancement.

The proportionate treatment is to implement the five types properly, keep them accurate, and spend the remaining effort on what they describe.

Frequently asked questions

What does structured data mean?

Machine-readable information embedded in a page’s HTML that states the meaning of the content using the schema.org vocabulary: what is a company, a price, a question, an answer or an article. The most common implementation is a JSON-LD block in the head of the page.

Does schema markup help AI visibility?

It improves the odds that models pick up your information correctly: the name, prices, contact details and question and answer pairs. Markup does not replace good content, but it reduces incorrect mentions and supports becoming a source.

Which schema types matter most?

Organization (or LocalBusiness), FAQPage, Service or Product with prices, Article with dates, and BreadcrumbList. In that order, and only where they match real content.

Does schema markup need to be hand-coded?

Rarely. In WordPress, SEO plugins produce the basic markup automatically, and FAQPage schema for FAQ sections can be generated from the content at theme level. Hand-coding is mainly needed for special cases.

Which schema types matter most for AI visibility?

Organization, Article, FAQPage, and Product with Offer for shops. Beyond those the returns fall off quickly, and keeping five types accurate is worth more than broad coverage nobody maintains.

Does structured data improve rankings?

Not directly. It describes what is already on the page so a machine does not have to infer it from layout. On an ambiguous page that changes whether the right value gets quoted; on a page with no answer on it, it changes nothing.

We generate FAQPage schema automatically from the FAQ sections on our own site, and the same can be done for clients’ sites as part of GEO work. The technical foundation is also described on the SEO service page.