Schema for AI Overviews: The Complete Technical Checklist

Schema is the fastest-moving signal in AI Overview optimization. Implement it correctly and you can see citation rate improvements within 2–6 weeks — before you’ve written a single new word of content.

In my experience auditing sites for AI search readiness, schema is the most consistently undervalued fix. I’ve reviewed sites with excellent content and near-zero AI citations — the missing layer was almost always structured data. The content was citable. Google just couldn’t read it efficiently enough to pull from it.

This is the complete implementation checklist: every schema type that matters for AI Overviews, every required field, the validation process, and the common errors that silently kill citation chances.

It’s part of the Generative Engine Optimization system I use on production sites. Start here before any content changes — schema fixes are prerequisite to everything else.

Why schema matters for AI Overviews

Google’s AI Overviews use a chunk-retrieve-synthesize pipeline. For your content to be retrieved in step 2 and extracted cleanly in step 3, the pipeline needs to know:

• What type of content this is (Article? FAQ? How-to procedure?)
• Who produced it (named author, with external verification)
• When it was last confirmed accurate (dateModified, not just datePublished)
• What the key questions and answers are (FAQPage schema)

Schema provides all four answers in machine-readable format. Without it, the pipeline has to infer all of this from the HTML — which it does, but with lower confidence. Lower confidence → lower citation likelihood.

Schema type 1 — Article

Required on every blog post and long-form page.

{
  "@context": "https://schema.org",
  "@type": "Article",
  "@id": "https://yourdomain.com/blog/post-slug/#article",
  "headline": "Article Headline Here",
  "description": "Meta description text here",
  "datePublished": "2026-05-23",
  "dateModified": "2026-05-23",
  "author": {
    "@type": "Person",
    "@id": "https://yourdomain.com/#person"
  },
  "publisher": {
    "@type": "Organization",
    "@id": "https://yourdomain.com/#organization"
  },
  "mainEntityOfPage": {
    "@type": "WebPage",
    "@id": "https://yourdomain.com/blog/post-slug/"
  },
  "image": {
    "@type": "ImageObject",
    "@id": "https://yourdomain.com/images/post-image.jpg",
    "url": "https://yourdomain.com/images/post-image.jpg",
    "width": 1200,
    "height": 630
  }
}

The field that matters most: dateModified

dateModified must reflect when the content was last substantively updated — not just the publish date. Sites where dateModified never changes after initial publish look stale to AI Overviews even when the content is current.

Rule: when you make a substantive content update (adding new information, not fixing a typo), update dateModified. This signals recency to the AI retrieval pipeline.

The field most often wrong: author

author should reference your Person entity via @id, not repeat the full Person object on every article. This creates the linked entity graph that AI systems use for attribution. A plain string like "author": "Bryan Collins" is worse than a proper entity reference.

Schema type 2 — FAQPage

Required on every article with a FAQ section.

{
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "What schema types are most important for AI Overviews?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "In order of impact: Article schema with accurate dateModified, FAQPage schema with PAA-sourced questions, Speakable schema on direct-answer blocks, Person schema on the About page with sameAs, and BreadcrumbList on every page."
      }
    }
  ]
}

Two rules that make or break FAQPage schema:

Rule 1 — Questions must come from People Also Ask data. Open Google, search your target keyword, expand the PAA boxes, and use those real questions as your FAQ questions. Invented questions don’t match what users are actually searching — and AI Overviews are more likely to extract answers to real user queries than to questions nobody asks.

Rule 2 — Schema questions must match visible page content. If your FAQPage schema lists six questions but only four are visible in the HTML, Google treats that as inaccurate structured data. Every question in the schema must have a visible, readable answer on the page.

Schema type 3 — Person

Required on your About page. Also referenced from every Article via author: { "@id": "..." }.

{
  "@type": "Person",
  "@id": "https://yourdomain.com/#person",
  "name": "Your Name",
  "url": "https://yourdomain.com",
  "description": "Your expertise description here",
  "jobTitle": "Your Title",
  "knowsAbout": ["topic 1", "topic 2", "topic 3"],
  "sameAs": [
    "https://www.linkedin.com/in/yourprofile/",
    "https://www.youtube.com/@yourchannel",
    "https://www.amazon.com/author/yourpage"
  ],
  "image": {
    "@type": "ImageObject",
    "url": "https://yourdomain.com/images/headshot.jpg"
  }
}

The field that does the most entity work: sameAs

The sameAs array connects your on-site identity to your external profiles. This is what AI models use to cross-reference your entity across sources — building confidence that “Bryan Collins” on your website is the same person as “Bryan Collins” on LinkedIn.

Every profile in the sameAs array must be active, accessible, and use your name consistently. Dead or inconsistent profiles weaken the entity signal.

Schema type 4 — Speakable

Required on direct-answer blocks and FAQ answer sections.

{
  "@type": "WebPage",
  "speakable": {
    "@type": "SpeakableSpecification",
    "cssSelector": [
      ".direct-answer",
      ".faq-answer",
      "article > p:first-of-type"
    ]
  }
}

Speakable marks which sections of your page are optimized for text-to-speech delivery and AI extraction. Google’s documentation describes it as enabling “voice actions” — but in practice, it functions as a signal to AI retrieval pipelines about which content blocks are most extraction-ready.

Implementation: add CSS classes to your direct-answer blocks, FAQ answer containers, and opening paragraphs. Add the Speakable schema referencing those classes. Validate that the selectors match actual elements in the rendered HTML.

This is the most underimplemented schema type in AI Overview optimization as of 2026. Early adoption is a genuine competitive advantage.

Schema type 5 — HowTo

Required on any page with step-by-step process content.

{
  "@type": "HowTo",
  "name": "How to Implement FAQPage Schema",
  "step": [
    {
      "@type": "HowToStep",
      "position": "1",
      "name": "Pull PAA data for your target keyword",
      "text": "Search your target keyword in Google and expand the People Also Ask boxes. Document 6–10 real questions users are asking."
    },
    {
      "@type": "HowToStep",
      "position": "2",
      "name": "Write visible FAQ answers on the page",
      "text": "Write a 50–100 word direct answer to each PAA question. These must be visible in the HTML — not generated only by schema."
    }
  ]
}

HowTo schema is a high-value extraction target for how-to queries specifically. If you have articles structured as numbered steps, HowTo schema makes that structure machine-readable and significantly increases extraction likelihood on process queries.

Schema type 6 — BreadcrumbList

Required on every page of the site.

{
  "@type": "BreadcrumbList",
  "itemListElement": [
    {
      "@type": "ListItem",
      "position": 1,
      "name": "Home",
      "item": "https://yourdomain.com/"
    },
    {
      "@type": "ListItem",
      "position": 2,
      "name": "Blog",
      "item": "https://yourdomain.com/blog/"
    },
    {
      "@type": "ListItem",
      "position": 3,
      "name": "Article Title",
      "item": "https://yourdomain.com/blog/article-slug/"
    }
  ]
}

Simple implementation with high structural signal value. This tells the AI pipeline exactly where the page sits in the site hierarchy — useful context for attribution and relevance scoring.

The validation process

After implementing schema, every page type goes through this validation:

1. Google Rich Results Test — paste the URL. Check for errors (must fix) and warnings (should fix). Screenshot the validation.

2. Content alignment check — manually verify that every schema property reflects what’s actually visible on the page. FAQPage questions match visible FAQ. HowTo steps match visible numbered steps. dateModified matches the actual last update.

3. Entity reference check — verify that author in Article schema uses @id pointing to your Person node. Verify that Person schema sameAs URLs are all active and accessible.

4. Repeat for every page type — not just one representative page. Each page type (blog post, about, homepage, category) should be validated separately.

Common errors that silently kill citations

Frequently asked questions

Related: Authority Signals for AI Search · Author Entity Optimization · How to Rank in Google AI Overviews · Why Your Site Isn’t in AI Overviews · The GEO Readiness Checklist · Generative Engine Optimization: The Complete Guide · Get the AI Search Audit ($49)