Structured data: what AI actually reads on your site

What does your site need to contain for AI to read it? Three things: factual content that's usable out of context, complete and honest structured-data markup, and strict consistency between what the code declares and what the page displays. JSON-LD markup is the technical component of this whole. It helps ChatGPT, Perplexity or AI Overviews understand who you are, what you sell and who signs your content — on its own, it creates not a single citation.

This page's slug contains "GEO," a term to which I no longer grant the status of a standalone discipline — the full argument is laid out in GEO vs SEO. This note covers the underlying technical dimension.

What exactly is structured data?

Structured data is standardized markup, most often in JSON-LD format, that makes explicit for automated systems the nature of a page's content: a product, an article, a person, an organization. The reference standard, schema.org, has been maintained since 2011 by Google, Microsoft, Yahoo and Yandex; as of 2026 it covers a few hundred types and several thousand properties^[1].

For classic SEO, the effects have been observable for a long time: rich snippets in Google results, FAQ display, rating stars, a readable breadcrumb trail. None of these is guaranteed — Google enables them at its discretion — but triggering them measurably improves click-through rate.

For generative environments, the effect is of the same nature. A system composing an answer from multiple sources relies on any signal that helps it understand a piece of content's structure, identify who is speaking, date a piece of information, and link a claim to a source. Markup is one of those signals. Not the only one, and not sufficient on its own.

Does a well-marked-up site get automatically cited by ChatGPT?

No, and three misconceptions deserve clearing up before any technical work begins.

Markup is not a secret weapon. A well-marked-up page is not mechanically cited by models: markup improves a machine's ability to read content, it does not create editorial quality that doesn't already exist in the visible content.

Nor is markup a "special AI" technique. The schema.org fundamentals are identical for classic SEO and for readability by large models; setting up JSON-LD "to rank in GEO" is a flawed way to put it. Careful markup should have equipped every serious site for years, and Google's May 15, 2026 doctrine confirmed that there is no separate optimization channel reserved for generative features^[2].

Finally, markup is not a cheating ground. Tagging a product page as a person page, or inflating an entity's attributes in the code, exposes you to documented penalties and loss of eligibility for rich results. Synthesis systems ultimately treat inconsistent markup with equivalent distrust.

The four schemas that matter for a Swiss SME

Organization describes the company itself: legal name, address, business identification number, sector, founder, professional online presence. A company without a complete and up-to-date Organization markup signals to machines that it isn't looking after its own identification.

Service or Product tags the main offerings. For a Swiss company with regional roots, the areaServed property deserves particular attention: specifying "French-speaking Switzerland" or "Canton of Valais" rather than a vague "Switzerland" steers model selection in geographically situated answers.

Article and Person cover editorial content. Dual identification — the article and its named author, with a biography — reinforces the E-E-A-T signal expected by search engines and picked up by synthesis systems. A collective "editorial team" byline is a weak signal; an identified author is a strong one.

FAQPage applies to pages that genuinely address frequently asked questions. The rule is strict: only mark up what appears in the visible content. FAQPage markup on a page that doesn't contain any is penalized.

Other schemas serve depending on the sector — LocalBusiness for a geographically anchored activity, JobPosting for careers, Course for training. The guiding principle doesn't change: only mark up what the page actually contains.

When redesigning mcva.ch, I set up these schemas before even writing the pages: Organization with the business identification number, Person behind every byline, Service for the offerings, Article for every note. This is the technical foundation I replicate in sites designed to be cited.

The mistakes that cancel out the effort

Four mistakes come up regularly, and each one costs more than it delivers.

Ghost markup: code that declares what the page doesn't display — a Person schema mentioning an executive absent from the rendered page, a Service schema describing an offer the site presents nowhere. These mismatches are detected in Search Console and cause loss of eligibility for rich results.

Inflated markup: invented or unverifiable properties — awards, aggregate ratings, endorsements. The short-term effect sometimes exists; the reversal at the first automated consistency check is brutal.

Inconsistent markup: the same company described differently across languages, subdomains or pages. In case of contradiction, synthesis systems arbitrate — and they keep the most stable source, not the most recent one. Cross-page consistency is a condition of authority, not a finishing touch.

Abandoned markup: a schema set up in 2022 and never revisited, while the leadership team, the offerings and the standard itself keep evolving. An annual review, even a quick one, is basic technical hygiene — and it is almost always neglected. The standard moves, your company moves too; the code has to follow both.

Markup is one project out of three

Markup makes things readable; it does not, on its own, make them citable. Two other efforts matter at least as much: editorial authority — sources cited, authors identified, content dated and maintained — and third-party authority — your company's presence in the directories, publications and open databases that models consult as arbiters. Perfect markup on thin content produces nothing; strong content poorly marked up may be picked up less than it deserves.

In a citability-oriented redesign, markup is set up during the content phase, not as a finishing touch. How the three efforts fit together, and how to measure the gaps, is codified in MCVA Notebook No. 1 — Measuring AI citability after May 2026. And to situate this technical work within the full set of decisions an SME must make, the overview is laid out in What should a Swiss SME do about AI in 2026?.

Key takeaways

— JSON-LD markup makes your site readable to machines; it replaces neither editorial quality nor third-party authority. — Four schemas cover the essentials for a Swiss SME: Organization, Service or Product, Article with Person, FAQPage. — Only mark up what the page actually displays: any gap between the code and the visible content ends up costing you.

FAQ

Is schema.org markup paid, or reserved for large sites? No. It's an open, publicly documented standard, usable by any site — from a Valais trust company's showcase site to an e-commerce platform. The real cost is the initial setup time and the annual review.

Does a CMS plugin suffice to set up markup? For the basics, often yes: Article, breadcrumb, FAQ. But Organization, Person and Service require configuration specific to your company — business identification number, service area, real authors — that no default setting knows.

How do I check that my markup works? With Google's rich results test and the schema.org validator, then in Search Console for ongoing monitoring. These tools detect syntax errors and some of the mismatches between code and visible content.

Does markup speed up citation by ChatGPT or Perplexity? No delay or effect is guaranteed — anyone promising otherwise is overselling. Markup improves the readability of your pages for synthesis systems; citation also depends on content quality and your third-party authority.

Is your site readable by AI — and picked up by it? The Score GEO™ pre-audit measures, free of charge, where your company appears — and doesn't appear — in the answers from ChatGPT, Claude, Perplexity, Gemini and Mistral. A quantified answer within a few days. Request a free pre-audit

Sources

[1] Google Search Central, Understand how structured data works. developers.google.com/search/docs/appearance/structured-data/intro-structured-data [↩]

[2] Google Search Central, Optimizing your website for generative AI features on Google Search, published May 15, 2026. developers.google.com/search/docs/fundamentals/ai-optimization-guide [↩]

Jérôme Deshaie is CEO and founder of MCVA Consulting SA, an AI-augmented agency based in Valais. Fifteen years serving major international brands, now working directly with Swiss SMEs. Background.