Schema Markup for Google AI Overviews: 4 Types That Matter

Schema markup is the single biggest structural signal Google’s AI Overviews uses to decide which pages to cite. Here are the four schema types that matter, with working JSON-LD templates for each.

Why schema matters for AI Overviews specifically

AI Overviews are generated, not retrieved. Google’s model scans candidate pages, extracts the answer chunks, and composes a summary with citations. Schema tells the model three things it can’t easily infer from raw HTML: what type of content this is, who wrote it, and when it was last updated. Pages with proper schema are cited at roughly 2× the rate of pages without — because the model trusts them more.

Schema 1 — FAQPage (highest ROI)

FAQ schema maps question-based headings to machine-readable Q&A pairs. AI Overviews cite FAQ blocks at roughly 2× the rate of equivalent prose content.

When to use

Any page with 3+ question headings. Pillar pages, comparison posts, how-to guides, product FAQs.

Template

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [{
    "@type": "Question",
    "name": "Your question verbatim",
    "acceptedAnswer": {
      "@type": "Answer",
      "text": "Your answer in 40-80 words. Self-contained. No pronouns that require external context."
    }
  }]
}

Rules

• 5–10 Q&A per page. Fewer = thin signal. More = dilution.
• Answers 40–80 words each. Self-contained.
• Question wording should match real user prompts (check GSC + Google Autocomplete).
• Visible FAQ on page MUST match schema exactly. Mismatches trigger structured-data errors.

Schema 2 — Article (E-E-A-T metadata)

Article schema gives Google the metadata its model uses to judge source authority: named author, publish date, modified date, publisher. Without it, your content is treated as “anonymous web page.”

Template

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "Your H1 exactly",
  "description": "Meta description, max 155 chars",
  "datePublished": "2026-01-15T00:00:00+00:00",
  "dateModified": "2026-04-18T00:00:00+00:00",
  "author": {
    "@type": "Person",
    "name": "Jane Doe",
    "url": "https://yoursite.com/about/jane-doe/",
    "sameAs": ["https://linkedin.com/in/jane-doe"]
  },
  "publisher": {
    "@type": "Organization",
    "name": "Your Brand",
    "logo": {
      "@type": "ImageObject",
      "url": "https://yoursite.com/logo.png"
    }
  },
  "mainEntityOfPage": "https://yoursite.com/the-actual-url/"
}

Rules

• Author type should be Person (for individual byline) not Organization when possible — Google trusts individuals more for E-E-A-T.
• dateModified must be updated on every content refresh (at least every 60–90 days).
• sameAs should point to the author’s LinkedIn / Twitter / published work — builds entity recognition.
• mainEntityOfPage must match the canonical URL.

Schema 3 — HowTo (for procedural queries)

HowTo schema marks a page as a step-by-step process. Google AI Overviews preferentially cite HowTo-marked content for “how to X” queries.

When to use

Any page with explicit numbered/ordered steps. Do NOT use on narrative content that isn’t a process.

Template

{
  "@context": "https://schema.org",
  "@type": "HowTo",
  "name": "How to X in Y minutes",
  "totalTime": "PT30M",
  "step": [
    {
      "@type": "HowToStep",
      "position": 1,
      "name": "First step name",
      "text": "What the reader does in step 1. Self-contained."
    }
  ]
}

Rules

• 5–10 steps. Fewer loses signal. More = likely not a single process.
• totalTime in ISO 8601 duration format (PT30M = 30 minutes, P1D = 1 day).
• Each step.text must be actionable — verb-first, concrete.
• Visible steps on page must match schema.

Schema 4 — Dataset (for downloadable assets)

Dataset schema marks a page as hosting structured data. Google AI Overviews preferentially cite dataset-marked content for data-driven queries (“average X cost,” “benchmark data for Y”).

Template

{
  "@context": "https://schema.org",
  "@type": "Dataset",
  "name": "Dataset title",
  "description": "1-2 sentence description of what the data contains.",
  "creator": {
    "@type": "Organization",
    "name": "Your Brand"
  },
  "license": "https://creativecommons.org/licenses/by/4.0/",
  "distribution": [{
    "@type": "DataDownload",
    "encodingFormat": "text/csv",
    "contentUrl": "https://yoursite.com/path/to/file.csv"
  }]
}

Validation workflow

1. Google Rich Results Test — paste URL. Zero errors. Warnings are OK.
2. Schema.org Validator — stricter, catches type errors Google ignores.
3. GSC Enhancements report — monitor over time for schema errors after deploy.
4. Re-test every 90 days — schema spec evolves; AI Overviews sensitivity changes.

Common schema mistakes that kill AI Overview citations

• FAQ schema without visible FAQ. Google penalizes hidden schema as deceptive. Visible Q&A must match markup.
• Stale dateModified. If your modified date is 9 months old, the AI treats content as stale regardless of actual edits.
• HowTo on non-procedural content. Marking a narrative post as HowTo is fraud detection bait.
• Missing author.sameAs. Anonymous authors are a weak E-E-A-T signal.
• Duplicate schema across pages. Each page needs its own unique Article schema with the correct URL and title.

Do NOT use these for AI Overviews

• Speakable schema — designed for voice search, not AI Overviews.
• ClaimReview — unless you are a certified fact-checking org.
• VideoObject alone — pair with Article, not standalone.
• BreadcrumbList alone — useful addition, but insufficient on its own.

Implementation priority

1. Week 1: Deploy Article + FAQPage on top 10 pages. Validate all. Zero errors.
2. Week 2: Add HowTo where applicable. Add Dataset for any downloadable assets.
3. Week 3+: Monitor GSC Enhancements. Refresh dateModified quarterly. Re-validate.