SEO · 10 min read

Schema markup for AI Overviews: 2026 guide

Across 730 AI citations analyzed, attribute-rich schema scores 61.7% vs 41.6% for generic. Full JSON-LD stack, copy-paste, validated.

Jérémy Wagner

Jérémy Wagner

Founder · Odyssée

Follow on LinkedIn
Schema markup for AI Overviews: 2026 guide

Why schema markup shifted from SEO to AI signal in 2026

Schema markup is no longer just for Google rich results. It now feeds AI engines directly to decide which pages to cite in their answers. ChatGPT Search, Perplexity, Google AI Overviews and Gemini all parse JSON-LD tags to identify the content type, its author, its dates and its trust signals.

Three numbers to set the scope:

61.7%

AI citation rate for pages with attribute-rich schema

Growth Marshal, Feb. 2026

730

real AI citations analyzed across 1,006 pages and 75 queries

Growth Marshal, Feb. 2026

+20 pts

gap between attribute-rich schema and generic schema

Growth Marshal, Feb. 2026

Source: Growth Marshal study, February 2026.

The conclusion is brutal. Sites that ship surface-level markup, like an empty Article or an Organization with no properties, get no measurable lift compared to sites with no schema at all. What pays is attribute-rich markup: filled prices, ratings, specs, sameAs, jobTitle, image at the right ratio.

What the research says about 730 real AI citations

The strongest study to date is the one by Kurt Fischman, published February 22, 2026. It analyzes 1,006 pages ranking on 75 queries, and the 730 AI citations generated by ChatGPT and Gemini on those same queries. Three findings break the schema marketing folklore.

  1. Generic schema brings nothing. Empty Article, Organization, BreadcrumbList does not increase the AI citation rate. The lift over no markup is not statistically significant.
  2. Attribute-rich schema changes everything. Product with prices, ratings and specs filled, or Review with identified author, hits 61.7% AI citation rate vs 41.6% for minimal implementations.
  3. Google rank still dominates. Each lost SERP position cuts AI citation odds by about 24%. Schema acts as an amplifier, not as a substitute for ranking.

The 4 critical schemas for a blog article

For an editorial article, four schema types cover the bulk of signals AI engines look for. Each plays a distinct role in the trust chain that ChatGPT, Perplexity and Google rebuild before deciding to cite you.

01Pilier 1

Article

Identifies the content type, author, publication and modification dates, cover image. Foundation of LLM extraction.
02Pilier 2

Person (author)

Builds the author entity with jobTitle, sameAs LinkedIn and knowsAbout. Decisive E-E-A-T signal for LLMs.
03Pilier 3

FAQPage

Marks each Q/A as an independent entity. Format directly reusable by AI engines in their synthetic answers.
04Pilier 4

BreadcrumbList

Provides the navigation path Home → Blog → Article. Helps AI engines locate the page in the site hierarchy.

On top of these four, Organization is declared once at the site level (in the layout.tsx), and every Article and Person references it via @id. This is the entity graph LLMs walk through to validate the consistency of a source. See our author page for a concrete Person + sameAs example deployed in production.

The full JSON-LD stack: copy-paste code

Here is the implementation used on this blog. It runs in production on every article and has been validated through Rich Results Test. The code is TypeScript / React, but the JSON-LD structure is portable to WordPress, Webflow or any backend that can inject a <script type="application/ld+json"> into the <head>.

Article + nested Person

{
  "@context": "https://schema.org",
  "@type": "Article",
  "@id": "https://your-site.com/blog/slug#article",
  "headline": "Exact article title",
  "description": "140-160 character meta description",
  "image": "https://your-site.com/blog/cover.jpg",
  "datePublished": "2026-04-27",
  "dateModified": "2026-04-27",
  "inLanguage": "en-US",
  "isAccessibleForFree": true,
  "articleSection": "SEO",
  "keywords": "schema markup, JSON-LD, AI Overviews",
  "author": {
    "@type": "Person",
    "@id": "https://your-site.com/authors/slug#person",
    "name": "First Last",
    "jobTitle": "Founder",
    "url": "https://your-site.com/authors/slug",
    "image": "https://your-site.com/authors/slug.jpg",
    "sameAs": [
      "https://www.linkedin.com/in/slug/",
      "https://github.com/slug"
    ],
    "worksFor": { "@id": "https://your-site.com/#organization" }
  },
  "publisher": { "@id": "https://your-site.com/#organization" },
  "mainEntityOfPage": {
    "@type": "WebPage",
    "@id": "https://your-site.com/blog/slug"
  }
}

Three properties pay a direct return in AI citations: sameAs on the author (LinkedIn mandatory), image in 1200×630 landscape format, and a real dateModified (no fake freshness).

FAQPage schema

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "What is the question?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Full answer of 40 to 80 words, readable outside the article context."
      }
    }
  ]
}

Place it only on pages that actually contain the FAQ. A FAQPage on a page with no visible Q/A in the HTML is treated as spam and triggers a Google manual action.

{
  "@context": "https://schema.org",
  "@type": "BreadcrumbList",
  "itemListElement": [
    { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://your-site.com" },
    { "@type": "ListItem", "position": 2, "name": "Blog", "item": "https://your-site.com/blog" },
    { "@type": "ListItem", "position": 3, "name": "Article title", "item": "https://your-site.com/blog/slug" }
  ]
}

Three levels are enough. The breadcrumb rich result shows up directly under the URL in the Google SERP, which improves CTR by 5 to 10% on the SERPs we have measured.

Organization (declared once, in the layout)

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "@id": "https://your-site.com/#organization",
  "name": "Agency name",
  "url": "https://your-site.com",
  "logo": "https://your-site.com/logo.png",
  "sameAs": [
    "https://www.linkedin.com/company/your-agency",
    "https://github.com/your-agency"
  ]
}

This tag is referenced by every Article (via publisher) and every Person (via worksFor). The result is a coherent entity graph that LLMs walk through whenever you are cited.

Generic vs attribute-rich: the mistake that wipes everything

The nuance that separates an effective implementation from a neutral one comes down to 4 or 5 properly filled properties. The Growth Marshal study shows that sites which mark up by default, without filling the truly informative attributes, get no lift. These 4 properties make the difference on a B2B technical blog.

Ce qui ne sert plus

  • Article without an identified author, or with author.name = "The team" or any generic alias without a dedicated page.
  • Person without sameAs, or with a single LinkedIn link whose profile does not mention the company.
  • FAQPage auto-generated with Q/A copied from other sites or without Q/A visible in the HTML.
  • dateModified bumped every week without any real content update (fake freshness detected by Google).
  • OG image at 800×600 or square format instead of the 1200×630 expected by SERPs.

À faire à la place

  • Author with a stable @id pointing to the site's author page, exact jobTitle, knowsAbout 3-5 topics.
  • Person.sameAs with LinkedIn, GitHub if developer, X/Twitter, personal site if relevant. Minimum 2 live links.
  • FAQPage only with Q/A actually present in the HTML, phrased as real user queries.
  • dateModified updated only on substantial change (new stat, new section, factual fix).
  • OG image at 1200×630 JPG or WebP, under 200 kB, declared in Article.image and openGraph.images.

FAQPage in 2026: what Google changed

Since 2023, Google has restricted the display of FAQ rich results to government and health sites. The official Google documentation states that FAQ rich results are now reserved for sites considered authoritative on government or public health topics.

This does not mean you should remove the FAQPage tag from regular B2B sites. The schema remains useful for two reasons:

  1. AI engines still use it. ChatGPT Search, Perplexity and Claude parse FAQPage to identify Q/A independently of the Google display. This is what allows an AI answer to cite your Q/A directly mid-paragraph.
  2. Bing still shows FAQ rich results on editorial sites, without the gov/health restriction. A correctly tagged blog can still get the rich snippet on Bing, which carries weight on the ChatGPT Search side.
⚠️

Don't confuse the two: the FAQ rich result (the accordion display in Google SERP) has been restricted since 2023. The FAQPage schema itself remains a valid signal for LLMs in 2026. Keep the tag, but stop counting on the Google rich result for most B2B sites.

For an editorial site, the trade-off is simple: keep the tag for the GEO benefit, accept that Google no longer shows the rich snippet, and capitalize on the fact that ChatGPT and Perplexity cite structured Q/A directly.

Deploy the full stack in one day

Deploying a clean JSON-LD stack takes six steps. The goal is not perfection, it is to cover the 4 critical types with truly filled attributes, and to validate every page through Rich Results Test before closing the ticket.

  1. 1

    Audit existing

    1h
    List pages with and without schema. Validate 5 pages via Rich Results Test to spot missing or poorly filled tags.
  2. 2

    Organization schema

    30 min
    Add a single Organization tag at the root layout level, then referenced by every Article and Person via stable @id.
  3. 3

    Person schema on the author page

    1h
    Create or complete the author page with name, jobTitle, sameAs (LinkedIn minimum), knowsAbout 3-5 topics, real photo.
  4. 4

    Article schema on the blog

    2h
    Nest Article + Person + publisher Organization on every article. headline, OG image 1200×630, real dateModified, inLanguage.
  5. 5

    FAQPage and BreadcrumbList

    1h
    FAQPage only if Q/A is visible in the HTML. BreadcrumbList systematic on child pages for navigation.
  6. 6

    Validation and deployment

    30 min
    Re-validate 10 key pages via Rich Results Test. Ship to prod. Watch Search Console > Enhancements for 7 days for warnings.

For a Next.js repo, you will find a complete example in src/components/seo/article-jsonld.tsx of this site. The component takes url, title, description, author and datePublished props, and returns the expected SSR JSON-LD <script>. This is what powers every article on the Odyssée blog.

To go further, the GEO 2026 pillar covers the 7 pillars that combine with schema, and our ChatGPT Search guide details the other levers to activate in parallel. If you prefer to delegate, our SEO & GEO offer includes the full deployment of the JSON-LD stack plus Rich Results Test validation across the entire site.

Questions fréquentes

Which format to use: JSON-LD, Microdata or RDFa?

JSON-LD is the format Google recommends and the one every AI engine prefers. It is cleanly separated from HTML, easier for LLMs to parse, and simpler to maintain in the codebase. Microdata and RDFa remain valid but have been declining since 2020. For a new site in 2026, going straight to JSON-LD is the obvious choice.

How many schemas can be stacked on a single page?

No technical limit. An article can stack Article, Person, FAQPage, BreadcrumbList and Organization with no issue. Google and LLMs parse the full graph. The rule: only declare schemas that match content actually present in the HTML. A FAQPage without visible Q/A triggers a manual action.

Is FAQPage schema still useful in 2026?

Yes, but no longer for Google rich results. Since 2023, Google has restricted FAQ display to government and health sites. The tag remains very useful for ChatGPT Search, Perplexity and Bing: AI engines cite Q/A tagged as independent entities directly. Keep the tag on B2B blogs, just stop counting on the Google display.

How to validate schema markup before shipping?

Use the Rich Results Test by Google by pasting the page URL or the code directly. The tool shows detected tags, warnings and errors. Also validate via the Schema.org Markup Validator for types not covered by Google. In production, monitor Search Console > Enhancements for the first 7 days.

Does schema belong on every page or only on articles?

Every editorial or commercial page benefits from a tailored schema. Article on the blog, Service on service pages, Product on e-commerce, Course on training pages, ContactPage on the contact page. Organization and BreadcrumbList are systematic across the entire site. A page with no schema starts with a measurable handicap on the GEO side.

Can attribute-rich schema make up for poor Google ranking?

No. The Growth Marshal study shows SERP position remains the dominant predictor of AI citations, with a 24% odds drop per lost position. Schema acts as an amplifier on already well-ranked pages, not as a substitute for ranking. Correct strategy: aim for top 5 on the target query with rich schema, not top 50 with perfect markup.

ShareLinkedInX
#JSON-LD#Schema#GEO#AI Search

Published on

Further reading

Articles in the same category

Appearing in ChatGPT Search: the 2026 practical guideSEO

Appearing in ChatGPT Search: the 2026 practical guide

ChatGPT Search passes 400M weekly users and runs on the Bing index. 6 concrete levers to be cited in AI answers. All deployable in 7 days.

April 21, 2026·10 min
GEO in 2026: why SEO is changing everything (and what to do)SEO

GEO in 2026: why SEO is changing everything (and what to do)

ChatGPT, Perplexity and AI Overviews capture 40% of searches. GEO is taking over from traditional SEO. Here's what to do right now to stay visible.

April 15, 2026·11 min

Shall we discuss your project?

30 minutes to scope your needs and tell you exactly what we would do in your position.

Start a project