Structured data in 2026: Google, AI and schema

Modular web pages connected by golden structured-data lines on a dark green background.

By Patricia Jarmek SEO 16 October 2026

6 min read

Structured data is often presented as a technical add-on that will “explain everything” to a search engine. It is more useful as a contract between visible content and a machine-readable description: it identifies the author, product, price, publication date and the page’s place in the site hierarchy.

Good markup reduces ambiguity and may make a page eligible for an enhanced search result. It does not compensate for weak content, guarantee a rich result or create a special route into AI answers. In 2026, a smaller accurate and maintainable graph is more valuable than a large graph built on assumptions.

What structured data does — and does not do

[1] recommend JSON-LD, while also supporting Microdata and RDFa. Markup must represent the main user-visible content, stay current and avoid misleading claims. Passing a technical test does not guarantee a rich result.

Keep three layers separate: Schema.org defines a vocabulary. Google documents which types can support a Google feature. The actual result depends on indexing, policies, quality and the query. Valid Service markup can add semantic clarity without creating a dedicated Google rich result. Conversely, a service page should not be labelled Product merely to pursue stars or a price display.

What changed in 2026

[9], added vocabulary for EU Digital Product Passports and ecommerce relationships such as isOftenBoughtWith and specification. This makes those facts expressible in a shared vocabulary. It does not mean Google automatically uses them in search features.

Google also adds and removes its own search features. Its [8] record the removal of FAQ rich-result documentation after the feature was retired in May 2026, alongside updates to some VideoObject properties. Structured data therefore needs an owner, an inventory and a review cadence.

Priority types for a business site and store

Organization, WebSite and BreadcrumbList

On the home page, describe the real organisation, URL, logo, contact details and verified profiles. Google says [4] can help it understand and disambiguate an organisation. Use a stable @id and reference that entity from articles and offers. [6] should reflect a genuine, usable site hierarchy.

Article or BlogPosting

Use Article or the more specific BlogPosting for editorial work. [3] highlights useful properties such as the headline, images, dates and author. datePublished is the original publication datedateModified changes only after a substantive update. Use a real author with a profile URL rather than “Admin”.

Product and Offer

On product pages, mark up only facts customers can see: name, image, currency, price, availability and the real offer. Google distinguishes [5]. For rapidly changing price or availability, key facts should be present in the initial HTML rather than depend entirely on JavaScript. [11]. Never fabricate AggregateRating, import third-party ratings as your own or leave stale prices in JSON-LD.

Person, ProfilePage and FAQ

An author page should connect a name, biography, expertise and verified profiles. This article is by Patricia Jarmek. An FAQ can remain useful to readers, but it should not be added just for a search-result treatment: Google [8]. Do not substitute QAPage. That type is for a single central question with user-submitted answers, not an editorial FAQ list.

There is no special schema for AI answers

Google explicitly says that AI Overviews and AI Mode require no special schema.org type or additional technical requirement. [7], and a page must be indexed and eligible to appear with a snippet. Google’s [13] also says that llms.txt, “AI markup” and artificial content chunking do not improve visibility in Google Search.

Structured data still matters for connecting authors, organisations, products and offers, and for standard rich-result eligibility. AI visibility begins with original, verifiable content that can be crawled and indexed. See SEO or GEO and How to create content that helps AI understand your brand for the broader strategy.

Multilingual markup without contradictions

Each language version needs localised headline, description, inLanguage, URL and text values. The organisation can retain one global @id, while each article has its own URL and @id. hreflang and canonical are not Schema.org properties, but they must agree with the WPML translation relationships. A translated title paired with an English description or the wrong language code undermines the model.

A maintainable implementation process

Build an inventory. Export JSON-LD by template and find duplicate organisations, conflicting authors and stale offers.
Map type to template. Articles, products, categories, authors and the home page need different data.
Choose one source of truth. Price, stock, dates and authorship should come from the CMS or commerce platform, not copied code.
Run two tests. [14] covers supported Google features[10] checks the broader Schema.org graph and syntax.
Inspect the live URL. Search Console URL Inspection shows what Google can retrieve. A clean validator result does not prove indexing or guarantee a rich result.
Monitor change. Review Search Console warnings monthly and official documentation quarterly.

Common failures include duplicate output from the theme, SEO plugin and WooCommerce, hidden or stale claims, invented ratings, dateModified changing on every request, wrong inLanguage and treating validator success as a display guarantee.

Frequently asked questions

Does structured data improve rankings?

It is not a direct ranking bonus. It can help systems understand content and enable certain rich-result features, but appearance is not guaranteed.

Is JSON-LD mandatory?

No. Google supports JSON-LD, Microdata and RDFa, but recommends JSON-LD because it is usually easier to generate and maintain separately.

Should we add FAQPage in 2026?

Not to obtain a Google FAQ rich result, because that feature has been retired. Keep useful FAQs for readers and use markup only when it serves a consistent data model.

Is there schema for AI Overviews?

No. Google does not require special AI schema, AI-readable text files or a separate writing format.