Schema markup for AI search 2026

Schema markup in 2026 isn’t just a Google ranking signal — it’s the primary structured-data layer AI engines parse to extract citations. FAQPage alone lifts citation rate roughly 3x. Most operators are still running 2018-era schema setups and wondering why AI engines skip them. This article is the implementation guide I wish I’d had when building 500k.io.

I run 500k.io solo at $9,500 MRR / $114K ARR / 22.8% to my $500K target. Separately I co-founded The Kreators AI with Jack — about $45M of client revenue ($10M Meta on my side). Both businesses depend on AI citation visibility for top-of-funnel. The schema setup below is what runs on 500k.io now and validates clean in Google Rich Results Test, Schema.org Validator, and every AI engine’s structured-data parse.

Why schema matters more in 2026 than 2023

In 2023, schema markup was a “nice to have” for Google rich snippets. Maybe a 5-10% CTR lift on SERP. Marginal.

In 2026, schema is the primary structured-data interface AI engines use to:

• Extract facts cleanly from your content
• Disambiguate entities (you vs another similar brand)
• Decide whether to cite a specific URL versus paraphrasing without attribution
• Build the entity graph that determines AI search rankings

The mechanism: AI engines parse JSON-LD schema reliably. They parse HTML semantics partially. They parse JS-rendered content least reliably. The path of least resistance for the engine is the JSON-LD block — so the engine biases toward content that gives it clean structured data.

Per Profound’s 2025 GEO benchmark, articles with FAQPage schema get cited at ~3.4x the rate of articles without. That’s not a margin — that’s a different game.

The 7 schema types you need

1. Article (every blog post / journal entry)

The foundation. Every long-form piece of content gets Article schema.

Required fields:

• @type: Article
• headline
• datePublished
• dateModified (critical for freshness signal)
• author (Person reference)
• publisher (Organization reference)
• image
• mainEntityOfPage

Optional but recommended:

• description
• articleSection (e.g., “AI Coding”)
• keywords
• wordCount

2. FAQPage (any article with 4+ questions)

The single highest-leverage GEO schema. If you have 4+ Q&A pairs in an article, mark them up.

Structure:

{
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "What is X?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "X is Y. It does Z."
      }
    },
    ...
  ]
}

Pro tips:

• Each answer 40-80 words. Long enough to stand alone as a citation, short enough to read on SERP.
• Don’t repeat the question in the answer. Engines extract the answer; the question is context.
• Real questions readers would ask, not made-up “what is the importance of X” filler.

3. BreadcrumbList (every page)

Helps engines understand site hierarchy. Lift on AI citations modest but real.

{
  "@type": "BreadcrumbList",
  "itemListElement": [
    {"@type": "ListItem", "position": 1, "name": "Home", "item": "/"},
    {"@type": "ListItem", "position": 2, "name": "Journal", "item": "/journal"},
    {"@type": "ListItem", "position": 3, "name": "Schema Markup 2026", "item": "/journal/schema-for-ai-search-2026"}
  ]
}

4. Person (every author)

The E-E-A-T signal. Engines disambiguate authors via Person schema with sameAs links.

{
  "@type": "Person",
  "@id": "/about#maxime",
  "name": "Maxime Le Morillon",
  "url": "/about",
  "jobTitle": "Founder, 500k.io",
  "worksFor": {"@id": "https://thekreators.ai#org"},
  "sameAs": [
    "https://x.com/theKreators_ai",
    "https://linkedin.com/in/maximelemorillon",
    "https://www.wikidata.org/wiki/Q[YOURID]"
  ]
}

The sameAs array is what tells engines “this person is the same entity across these platforms.” Without it, your name is ambiguous.

5. Organization (homepage + globally)

Defines the publisher entity.

{
  "@type": "Organization",
  "@id": "/about#org",
  "name": "500k.io",
  "url": "https://500k.io",
  "logo": "https://500k.io/logo.png",
  "parentOrganization": {"@id": "https://thekreators.ai#org"},
  "sameAs": [...]
}

The parentOrganization field is how I wire 500k.io to The Kreators AI. Engines understand the relationship.

6. HowTo (any tutorial-format content)

For step-by-step content. Steps become individually-citable units.

{
  "@type": "HowTo",
  "name": "How to write a CLAUDE.md file",
  "step": [
    {"@type": "HowToStep", "name": "Step 1", "text": "..."},
    {"@type": "HowToStep", "name": "Step 2", "text": "..."}
  ]
}

Pro tip: HowTo schema gets de-emphasized in some Google searches as of 2024. But AI engines still weight it. Use it for tutorials.

7. Product or Review (for tool reviews / comparisons)

For articles that review or compare specific tools.

{
  "@type": "Review",
  "itemReviewed": {
    "@type": "Product",
    "name": "Beehiiv"
  },
  "author": {"@id": "/about#maxime"},
  "reviewRating": {
    "@type": "Rating",
    "ratingValue": "4.5",
    "bestRating": "5"
  }
}

For comparison articles (like Beehiiv vs ConvertKit), use Product schema for each compared product.

The @graph wiring (the pro move)

The amateur approach: separate <script type="application/ld+json"> tags for each schema type.

The pro approach: one JSON-LD block with @graph containing all entities, cross-referenced by @id.

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "WebSite",
      "@id": "https://500k.io#website",
      "url": "https://500k.io",
      "name": "500k.io",
      "publisher": {"@id": "https://500k.io#org"}
    },
    {
      "@type": "Organization",
      "@id": "https://500k.io#org",
      "name": "500k.io",
      "url": "https://500k.io",
      "logo": "https://500k.io/logo.png",
      "parentOrganization": {"@id": "https://thekreators.ai#org"}
    },
    {
      "@type": "Person",
      "@id": "https://500k.io/about#maxime",
      "name": "Maxime Le Morillon",
      "sameAs": ["...", "...", "..."]
    },
    {
      "@type": "Article",
      "@id": "https://500k.io/journal/schema-for-ai-search-2026",
      "headline": "Schema markup for AI search 2026",
      "datePublished": "2026-05-18",
      "dateModified": "2026-05-18",
      "author": {"@id": "https://500k.io/about#maxime"},
      "publisher": {"@id": "https://500k.io#org"},
      "isPartOf": {"@id": "https://500k.io#website"}
    },
    {
      "@type": "FAQPage",
      "@id": "https://500k.io/journal/schema-for-ai-search-2026#faqs",
      "mainEntity": [...]
    },
    {
      "@type": "BreadcrumbList",
      "itemListElement": [...]
    }
  ]
}

Why this matters: AI engines (and Google) parse the @graph as a single coherent entity description. Cross-references via @id let the engine understand “the author of this Article is the same Person who works for the parentOrganization referenced in the Organization.” The full entity picture compounds.

Most CMS plugins ship separate scripts. They miss the @id cross-references. Custom JSON-LD beats plugin output for this.

How to implement on common stacks

Astro

Generate JSON-LD in a server-rendered component. I use a seo.ts utility that builds the @graph from frontmatter + page context. Output goes into <Layout> head as a single <script>.

Next.js

metadata API in App Router supports JSON-LD. Use a server component to generate and inject.

WordPress

Yoast or Rank Math handle basic schema. For @graph wiring with Person sameAs, write a custom JSON-LD function in functions.php or use Schema Pro. Plugins fall short on the relationships.

Static HTML / Hugo

Direct JSON-LD in template <head>. Hugo’s data files make this clean.

The 3 mistakes that kill AI visibility

Mistake 1 — Stale dateModified

Articles with datePublished: 2024-03-12 and no dateModified updates get de-weighted. Engines assume the content is stale.

Fix: every quarter, review pillar articles. Update one or two stats, bump dateModified. Cost: 30 min per article. Returns: maintained citation share.

Mistake 2 — Missing Person schema with sameAs

Articles by anonymous or barely-identified authors lose citation rate. The fix: Person schema on every article + sameAs linking to verified social/Wikidata.

If you don’t have a Wikidata entry, get one. 60 minutes. Single biggest entity-graph signal you can add.

Mistake 3 — Schema in JS-rendered content

If your JSON-LD is injected by client-side JavaScript, AI crawlers often miss it. View-source the page. If <script type="application/ld+json"> doesn’t appear in raw HTML, fix that this week.

The fix: server-render schema. Astro, Next.js (App Router), and most modern frameworks support this natively. WordPress plugins also output server-side. The danger zone is React SPAs without SSR — there schema sometimes lives only in JS.

Validation tools

Before shipping any schema change, validate:

Tool	What it tests	URL
Google Rich Results Test	Google’s interpretation	https://search.google.com/test/rich-results
Schema.org Validator	Pure spec compliance	https://validator.schema.org/
Bing URL Inspection	Bing’s interpretation	https://www.bing.com/webmasters
Schema App Validator	Detailed property checking	Various

Test every new schema before deploy. The cost of a regression is real (lost citations for the duration). The cost of testing is 90 seconds.

What I run on 500k.io

Current schema stack on every article:

Schema	Required?	Notes
Article	✓	Every post
FAQPage	✓	If 4+ FAQs (almost always)
BreadcrumbList	✓	Site-wide
Person	✓	Always Maxime, sameAs to LinkedIn, X, Wikidata
Organization	✓	500k.io with parentOrganization → Kreators AI
WebSite	✓	Once per page
HowTo	Conditional	When tutorial format
Product / Review	Conditional	When tool review

Total schemas per article: 5-7. Wired into a single @graph block. Server-rendered via Astro seo.ts utility.

Validation: every article runs through Google Rich Results Test pre-publish as part of the content engine. Hard gate — articles failing validation don’t ship.

Internal links

• How AI engines actually choose citations — the broader citation logic.
• GEO 2026: how to get cited by ChatGPT and Perplexity — the strategic frame.
• Brand mentions beat backlinks (the 2026 data) — what Person schema sameAs feeds.
• AI SEO playbook 2026 — the holistic strategy.
• The AI content engine: from brief to published in 2 hours — schema validation as part of the workflow.
• How to rank in Perplexity (DR zero reality) — Perplexity-specific tactics.

External sources

• Schema.org documentation — the spec itself.
• Google Search Central — structured data — Google’s interpretation.
• Profound — GEO benchmark studies — the 3.4x FAQPage citation lift data.
• Anthropic — llms.txt and structured data — emerging standards from major AI providers.

What to ship this week

1. Add FAQPage schema to your top 5 articles. 60 minutes total.
2. Add Person schema with sameAs to your author. 15 minutes.
3. Validate every article in Google Rich Results Test. 5 min/article.
4. Audit dateModified across all pillar articles. Bump anything older than 90 days. 30 min.

Five hours of work. Likely 30-50% citation rate lift on those articles within 60 days. Cleanest leverage in technical SEO right now.

FAQ

Is schema markup still relevant in 2026?

+

More than ever. AI engines parse structured data more reliably than HTML. FAQPage schema lifts citation rate ~3x per Profound's 2025 study. The relevant question isn't 'is schema worth it' — it's 'which schemas actually move citations.'

What's the minimum schema set I need?

+

Article, FAQPage, BreadcrumbList, Person, Organization. Five schemas wired into a JSON-LD @graph. That's the floor. Add HowTo, Product, Review, or specific schemas based on content type.

How do I wire schemas together with @graph?

+

Put all schema types into one JSON-LD script tag with a @graph array. Each entity has an @id (URL or fragment). Cross-reference using @id strings. This single-block approach is what Google and AI engines parse cleanest.

Does FAQPage schema really move AI citations 3x?

+

Yes per Profound's 2025 measurement and matching data from Otterly. The mechanism: FAQ structure maps cleanly to the question-answer format AI engines output. Adding 5 well-written FAQs to a pillar article is the highest-leverage GEO intervention you can make.

What about Anthropic's mention of structured.json?

+

Speculative. Anthropic has hinted at supporting a structured.json format alongside llms.txt but as of May 2026 there's no official spec. Don't build for it yet. Watch the official Anthropic announcements.

Can I use plugins to handle schema or do I need to write JSON-LD?

+

Plugins (Yoast, Rank Math, etc.) generate basic schema fine. They fall short on @graph wiring and Person schema with sameAs. For founder-level GEO results, you'll likely need custom JSON-LD generated server-side.