Structured data and AI citability: what mark-up can and cannot do

Note revised on 25 May 2026. Article originally published in February 2025 — full rewrite following the Google Search Central publication of 15 May 2026.

The slug of this page contains the word "GEO", which the firm no longer recognises as an autonomous discipline. The clarification has been set out in a dedicated note to which it is useful to refer here^[1]. The present text deals with the underlying technical dimension: what structured-data mark-up can, and cannot, do for a company's visibility in generative environments.

What structured data is

Structured data is standardised mark-up, most often in JSON-LD, which makes explicit the nature and structure of a web page's content for the automatic systems that read it. A product page tags a product; an article page tags an article; a person page tags a person. The reference standard — schema.org — has been jointly maintained by Google, Microsoft, Yahoo and Yandex since 2011, and now covers several hundred types and several thousand properties^[2].

For classical SEO, this mark-up has produced observable effects for years: rich snippets in Google results, enriched FAQ display, rating stars on products, breadcrumbs visible on the results page. None of these effects is guaranteed — Google enables them at its discretion — but their triggering measurably improves click-through rates.

For generative environments, the effect is of the same nature but of a different order. The synthesis systems that compose a response from several sources rely on any signal that helps to understand the structure of a content, identify who is speaking, date a piece of information, link a claim to a source. Structured-data mark-up is one of these signals. It is not the only one, and it is not sufficient on its own.

What mark-up is not

Three misunderstandings deserve to be cleared up before any technical work.

Structured-data mark-up is not a secret weapon. Well-tagged pages are not mechanically cited by models. Mark-up improves the readability of content for an automatic system; it does not create editorial quality that does not exist in the visible content.

Nor is mark-up an optimisation technique specific to generative engines. The fundamentals of schema.org mark-up are identical for classical SEO and for readability by LLMs. Implementing JSON-LD because one wants to "shift to GEO" is a mistaken formulation. Careful mark-up should have been in place on any serious site for years.

Finally, mark-up is not a deception technique. Marking a product page as a person page, or conversely, or overstating an entity's attributes in its mark-up, exposes the site to documented sanctions from engines and to loss of eligibility for enriched displays. Generative models, whose corpora progressively integrate observed penalties, end up handling inconsistent mark-up with equivalent vigilance.

The relevant schema types by business profile

Four schema types cover most of the needs of a mid-sized Swiss B2B company. None is new; all benefit from being systematised.

Organization. The base schema that describes the company itself: legal name, address, UID number, sector of activity, founder, management team, presence on professional networks. A company without complete and up-to-date Organization mark-up signals to automatic systems that it does not take care of its own identification — it is a weak signal.

Service or Product. The mark-up of the company's main offerings. For a consulting firm, the Service schema; for a software publisher, the SoftwareApplication schema; for a manufacturer, the Product schema. The areaServed property deserves particular attention for a Swiss company with regional anchoring: explicitly specifying "Suisse romande" or "Canton of Valais" rather than a vague "Switzerland" guides the models' selection in geographically situated responses.

Article and Person. For the site's editorial content — articles, notes, analyses — the Article schema combined with the Person schema of the author. This dual identification reinforces the E-E-A-T signal expected by engines, and it now constitutes an editorial authority signal taken up by synthesis systems. The anonymous author or the collective "editorial team" signature is a weak signal; the nominally identified author with a professional biography is a strong signal.

FAQPage. For pages that effectively deal with frequently asked questions. Not to be overloaded: mark-up only when the questions and their answers actually appear in the page's visible content. FAQPage mark-up on a page that contains none is sanctioned.

Other schemas may be relevant depending on the sector — LocalBusiness for a geographically anchored activity, JobPosting for a careers page, Course for a training organisation, MedicalEntity for the health sector. The guiding principle remains the same: only tag what the page effectively contains.

The common mistakes that invalidate the effort

Four mistakes recur in the technical audits conducted by the firm.

The first is phantom mark-up: tagging present in the code but with no correspondence to the visible content. A Person schema mentioning an executive who appears nowhere on the rendered page. A Service schema describing an offering that the site does not present. These mismatches, detectable by Google Search Console and by validation tools, can result in loss of eligibility for enriched results and weaken the trust granted to the page.

The second is inflated mark-up: a schema overloaded with excessive properties — awards, endorsements, aggregateRating that are invented or unverifiable. This inflation can produce a short-term effect and then turn around, abruptly, at the first automatic inconsistency check.

The third is mark-up that is inconsistent across pages: the same company described with slightly different information across subdomains, languages, product pages. Synthesis systems arbitrate in case of contradiction, and tend to retain the most stable source, not the most recent one. Perfect cross-page consistency is not a cosmetic finish; it is a condition of authority.

The fourth is unmaintained mark-up: a schema set up in 2022 and never re-examined. The evolution of schema.org, changes in the management team, shifts in the scope of activity create gaps that accumulate. An annual review of the mark-up, even a brief one, is a hygiene exercise that technical departments often underestimate.

One dimension among others

Structured-data mark-up is one of the three operational workstreams that the firm identifies in its framework on AI citability. The other two — editorial authority (source transparency, identification of authors, freshness of content) and third-party authority (presence in the sources that models consult as arbiters) — matter at least as much. Perfect mark-up on poor content does not produce citability; strong content that is poorly marked up may be less well selected than it deserves.

The articulation of the three workstreams and the full method for measuring gaps are set out in the Cahier MCVA n°1 — Measuring AI citability after May 2026, the first volume of the Cahiers MCVA collection.

Sources

[1] See GEO vs SEO: why this opposition no longer has any doctrine to stand on, Jérôme Deshaie, rewrite of 25 May 2026. [↩]

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

Jérôme Deshaie is CEO of MCVA Consulting SA, a Swiss firm specialising in strategic consulting on artificial intelligence, based in Valais.