Schema markup for AI overviews: what actually works in 2025

TL;DR: Schema markup doesn't directly trigger AI Overview citations, but it makes your content much easier for Google's systems to parse and trust. Types like FAQPage, HowTo, Article, and Product hand AI systems clean, extractable facts. Pages with relevant schema get cited more consistently because the signal cuts ambiguity about what a page claims and who's behind it.

What is schema markup and why does it matter for Google AI Overviews?

Schema markup is structured data you add to a webpage, usually in JSON-LD format, that tells search engines what your content means instead of making them guess. A recipe page can say it's a recipe, name the author, list the ingredients, and specify the cook time. A product page can declare its price, availability, and reviews. Instead of Google inferring those things from your prose, schema hands them over explicitly.

Google AI Overviews, the answer-style summaries that now sit at the top of many results pages, depend on Google's ability to pull trustworthy, specific facts from the web [1]. When a page has no structured data, the system does more interpretive work. When a page has accurate, complete schema, the system finds what it needs faster and with more confidence.

This matters more than it did two years ago. Google's Search Generative Experience turned into AI Overviews across most English-language queries by mid-2024, and coverage kept growing through 2025 [2]. The pages getting cited tend to share a few traits: clear entity definitions, factual specificity, authorship signals, and structured data that says the same thing the page says.

Schema is less a magic ranking factor than a way to reduce friction. AI systems work probabilistically. Anything that removes uncertainty about your content's meaning, your brand's identity, or your data's accuracy nudges the odds that your content gets selected and cited.

Does schema markup directly affect whether you appear in AI Overviews?

Honest answer: there's no public confirmation from Google that schema is a direct ranking signal for AI Overviews specifically. Google has said structured data earns rich results and helps it understand content [3], but the company hasn't made a clean causal claim about AI Overviews and schema the way it has for featured snippets.

What the evidence shows is correlational and plausible. A 2024 analysis by Authoritas found that pages appearing in AI Overviews were more likely to use structured data than pages that didn't appear, even after controlling for domain authority [4]. The mechanism makes sense. AI Overview generation leans on Google's Knowledge Graph and its grasp of entities, relationships, and facts. Schema feeds that graph.

So schema probably won't get you into AI Overviews on its own. A page with perfect schema and thin content won't beat a substantive, well-cited page with no schema. But schema on a genuinely good page tips the balance. It's the difference between a page Google can confidently quote and one it's less sure about.

For queries with clear factual answers, how-to steps, product comparisons, or event details, schema makes your answers machine-readable at exactly the level AI extraction wants. For opinion-heavy or narrative content, schema still helps with entity identification, but the lift is smaller.

Which schema types matter most for AI Overview visibility?

Not all schema is equal here. Some types map directly onto the kinds of content AI Overviews pull from. Others are useful for rich results but less relevant to how AI systems extract answers.

FAQPage is probably the highest-leverage type right now. AI Overviews often extract question-and-answer pairs, and FAQPage schema structures those pairs explicitly. If your page answers the specific question a user typed, FAQPage schema makes that match unambiguous [3].

HowTo is similar for procedural queries. Step-by-step instructions with named steps, durations, and tools give AI systems the exact format they need to surface a process answer.

Article and NewsArticle matter for topical authority. These types let you declare the author (link to an author entity with Person schema), the publication date, the headline, and the publisher. Google's quality systems care about E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness), and Article schema with a properly defined author entity is a direct signal toward that [9].

Product and Offer schema help commercial queries. Price, availability, review aggregate scores, and return policy become extractable facts rather than buried text. AI Overviews for product comparison queries pull from this kind of structured data.

Organization and WebSite schema on your homepage establish your brand as a known entity. This feeds the Knowledge Graph entry for your company, which matters when AI systems decide whether your brand is a credible source. Pair Organization schema with a SameAs property pointing to your Wikipedia page, Wikidata entry, LinkedIn, and other authoritative profiles.

BreadcrumbList helps AI systems read your site's topical hierarchy. It's not flashy, but it adds to the overall picture of what your site covers and how authoritatively.

Here's a quick comparison of which types do what:

| Schema type | Primary AI benefit | Query types it helps with | |---|---|---| | FAQPage | Exact Q&A extraction | Informational, navigational | | HowTo | Step extraction | Procedural, how-to | | Article / NewsArticle | Authorship, freshness signals | News, guides, analysis | | Product / Offer | Price, availability, reviews | Commercial, comparison | | Organization | Brand entity, trust | Brand, navigational | | Person | Author credibility (E-E-A-T) | Any content with named author | | BreadcrumbList | Topic hierarchy | All types | | Speakable | Audio/voice extraction | Voice search, AI assistants |

Schema types and their primary AI Overview benefit

| | | |---|---| | FAQPage | 95 | | HowTo | 90 | | Article / NewsArticle | 80 | | Organization | 78 | | Product / Offer | 75 | | Person (author) | 72 | | Speakable | 65 | | BreadcrumbList | 55 |

Source: Authoritas AI Overviews Study 2024; Google Developers Structured Data documentation

How do you implement schema markup correctly for AI search?

JSON-LD is the format Google explicitly recommends, and it's the one AI systems handle most cleanly [3]. It sits in a script tag in the page head or body, separate from your HTML content, so you can update it without touching your visible copy. Microdata and RDFa still work, but they're more fragile and harder to maintain.

A few implementation principles matter specifically for AI visibility.

First, match the schema to what the page actually says. This sounds obvious, but it's where most errors happen. If your FAQPage schema lists six questions and your page only covers four, Google's systems catch the mismatch and may discount the schema entirely. Accuracy beats completeness.

Second, nest your schema properly. An Article that references a Person (the author) that references an Organization (the publisher) gives Google a connected graph of entities instead of isolated fragments. Each entity you define well reinforces the others.

Third, use the description property liberally. This one is underused. A detailed, accurate description field in your Organization or Article schema gives AI systems direct text to work with when they write summaries that reference your brand or content.

Fourth, keep your schema current. A Product schema showing a price from two years ago, or an Event schema for a date that's passed, signals stale data. Google's freshness evaluation applies to structured data too.

Fifth, validate everything. Google's Rich Results Test (search.google.com/test/rich-results) checks your schema against the current spec. Schema.org's validator catches structural errors. Run both on every page that gets structured data, and build validation into your deployment pipeline so broken schema doesn't ship silently.

For teams managing schema at scale, most CMS platforms now have structured data plugins. WordPress has Yoast SEO, Rank Math, and Schema Pro. Shopify handles Product schema natively. But plugins give you templates, not perfection. Review auto-generated schema on your most important pages by hand rather than trusting the plugin got it right.

What are the best practices for schema markup and Google AI Overviews SEO in 2025?

The best practices that matter in 2025 have shifted a little from what was standard two years ago. Here's what the current evidence and Google's own guidance support.

Go beyond the minimum. Early advice said add the required fields and stop. For AI visibility, recommended and optional properties matter. In FAQPage, for example, adding acceptedAnswer with a properly formatted text property and dateCreated on each answer gives AI systems richer data to work with.

Build entity stacks. AI systems care about entities and relationships more than keywords. Define your brand as an Organization with a proper sameAs array. Define your authors as Person entities with knowsAbout and credential properties. Define your products as Product entities with Offer, AggregateRating, and Brand relationships. The more completely you define your entities, the stronger your footprint in the Knowledge Graph.

Keep schema faithful to your visible copy. Google can and does cross-reference structured data claims against the visible text of the page [10]. If your FAQPage schema claims an answer that isn't reflected in your page copy, that's a red flag. Write your content first, then write schema that describes it accurately.

Use Speakable schema for voice and assistant contexts. Speakable (schema.org/speakable) marks specific sections of a page as suitable for text-to-speech and AI assistant extraction. It's under-adopted but explicitly useful in the AI-first context. It tells Google's systems: this paragraph is the clean, quotable version of our answer [3].

Prioritize pages that already rank. Schema won't pull a page from position 30 into an AI Overview. It helps pages already in consideration. Find which of your pages appear on page one for target queries, then make sure those pages have complete, accurate schema. That's where the marginal return is highest.

For teams wanting to measure this systematically, AI search visibility metrics and KPIs are worth understanding before you start an implementation project, so you know what a win looks like.

If you want a structured audit of which pages need schema work and how your brand shows up in AI answers, Spawned's AI visibility audit surfaces those gaps, ranked by traffic opportunity.

How does Google use structured data when generating AI Overviews?

Google hasn't published a detailed technical spec of how its large language models interact with structured data during AI Overview generation. What it has shared publicly, and what you can infer from how AI Overviews behave, fills in a reasonable picture.

Google's AI Overview systems draw from its index, which holds structured data alongside page text [1]. When it generates a response, the system retrieves candidate pages and pulls out relevant content. Pages with structured data hand the system pre-parsed facts. Instead of reading a paragraph to find a product's price, the system reads the price property directly.

The Knowledge Graph is the connective tissue. When you implement Organization schema with proper SameAs links, Google updates its Knowledge Graph entry for your brand. When AI Overview generation looks for credible sources on a topic your brand covers, that Knowledge Graph entry is part of what decides whether your content is a candidate.

Google's documentation on AI Overviews says it cites sources it finds reliable, relevant, and fresh [1]. Structured data contributes to all three. Author and organization schema feed reliability signals. Freshness comes from accurate date properties. Relevance comes from the topical specificity that schema enables.

There's also the angle of generative engine optimization, which looks at how to structure content more broadly so AI systems can extract and cite it. Schema is one layer of that, alongside content structure, entity definition, and citation patterns.

What does the research say about schema and AI search citation rates?

Nobody has great controlled experimental data here. The closest published work comes from a handful of SEO analytics firms that analyzed large samples of AI Overview citations and compared the structured data rates of cited versus non-cited pages.

The Authoritas 2024 AI Overviews study analyzed thousands of queries and found that URLs appearing in AI Overviews were more likely to use structured data, particularly FAQPage and HowTo types, than URLs in organic results that weren't cited [4]. The study didn't establish causation, and the effect sizes varied by query type.

A Semrush analysis of AI Overview citation patterns from late 2024 found that roughly 80% of AI Overview sources came from pages already ranking in the top 10 organic results [6]. That means schema's role is mostly about separating already-competitive pages, not pulling uncompetitive pages into citations.

For AI SEO more broadly, the research consensus is that content quality and topical authority are the dominant factors. Schema is a multiplier on content that's already strong, not a substitute for it.

One honest caveat: the AI Overview system changes often. Google rolled out AI Overviews more aggressively in early 2025, pulled back on certain query types after accuracy complaints, then expanded again [2]. Any specific correlation numbers from six months ago may not hold today. The directional finding, that structured data helps AI systems extract and trust your content, has stayed consistent even as the specifics shift.

Are there schema types that don't help with AI Overviews?

A few schema types are basically decorative for AI Overview purposes, even though they still earn rich results in standard search.

VideoObject schema is useful for video carousels and video-rich results in Google, but AI Overviews rarely cite video content directly. If your content lives mostly in video with little text, schema alone won't make up for the missing extractable text.

SiteLinksSearchBox schema helps with branded search navigation but has no known effect on AI Overview citation rates. It's still worth implementing for user experience. Just don't expect it to move the needle here.

Recipe schema is a special case. It generates great rich results, and Google does sometimes cite recipe pages in AI Overviews for cooking queries, but the properties that matter most there are ingredient lists and step instructions, which is exactly what HowTo-style properties capture anyway.

The types to deprioritize describe your site's technical architecture rather than its content. SiteLinksSearchBox, SearchAction, and ReadAction schema are for navigational behavior, not content extraction.

Spend your implementation effort on schema that describes what you claim, who made it, and what it's about. Those are the properties AI systems need to select and cite your content.

How does schema markup connect to broader AI SEO strategy?

Schema is one layer in a larger system. Picture AI search as a stack: your content's factual quality and specificity at the bottom, entity definition and Knowledge Graph presence above that, structured data marking up your content above that, then technical accessibility (crawlability, page speed, clean markup) at the top. You need all the layers, and they reinforce each other.

Entity definition is where a lot of brands underinvest. Your brand needs a Wikipedia page or Wikidata entry, a Google Business Profile if applicable, consistent NAP (name, address, phone) data across the web, and structured data that connects all of these via SameAs properties. Without that entity foundation, even perfect schema does less, because Google's systems don't have a strong prior belief about who you are.

Content specificity matters because AI systems prefer concrete claims over vague generalizations. A page that says "prices vary by provider" is less citable than one that says "prices typically range from $X to $Y per month based on contract length, according to the Bureau of Labor Statistics Consumer Expenditure Survey." Schema can mark up that specific claim, but the claim has to exist in your content first.

For teams tracking how their brand shows up across AI assistants like ChatGPT, Perplexity, and Gemini (more than Google), the dynamics differ, because those systems don't read structured data directly from crawled pages the way Google does. For those systems, the signals that matter are your training data footprint, how often your brand and claims appear in high-quality referenced text, and whether your content has been indexed by the underlying retrieval systems. That's a different optimization path, covered in depth in generative engine optimization.

For Google AI search specifically, schema is one of the clearest, most implementable levers you have.

How do you audit your current schema for AI Overview readiness?

Start with Google Search Console. The Enhancements section shows which schema types Google has detected on your site, how many valid items exist, and where it found errors or warnings [7]. That's your baseline. If Google is finding errors in your current schema, fix those before adding new types.

Next, run your most important pages through Google's Rich Results Test. This tells you which rich result types each page qualifies for, which is a proxy for how cleanly your schema is implemented.

Then do a content-schema alignment check. For each page with schema, compare the structured data properties to what the page actually says. Price properties should match displayed prices. Author names should match bylines. FAQ questions should match the actual questions answered in the content. Mismatches erode trust signals.

After that, look at which pages generate AI Overview impressions for you. Search Console added AI Overview click and impression data in 2024, though the reporting has been inconsistent and granular data isn't always available [7]. Filter your Search Console data by query type and look for pages with high impressions but low AI Overview appearances. Those are candidates for schema improvement.

Finally, benchmark against competitors. Search the queries you want to win. When AI Overviews appear, view the source of pages that get cited. Check their structured data with a browser extension like Schema Markup Validator, or just inspect the page source and look for application/ld+json script tags. You'll see the patterns fast.

For a more automated version of this, AI SEO tools and AI visibility tools can track your AI Overview appearances and flag structured data gaps at scale, which is hard to do by hand across hundreds of pages. Spawned's platform does this tracking automatically and surfaces which schema gaps correlate with missed citation chances for your specific target queries.

What common schema mistakes hurt AI Overview visibility?

The most common mistake is schema that doesn't match page content. Google's quality systems actively check for this. A 2023 Google post on spam policies specifically named structured data misuse as a quality issue that can trigger manual actions [5]. Even short of a manual action, mismatched schema gets ignored or discounted.

The second most common mistake is using schema templates without customizing them. A generic Article schema with no author entity, a placeholder description, and a datePublished that's just the site launch date helps almost nothing. Schema has to be specific to be useful.

Third: forgetting to update schema when content changes. If you change a product price, revise an FAQ answer, or edit a how-to step, the schema needs to change too. Stale schema is worse than no schema for freshness signals.

Fourth: stacking too many schema types on a single page in ways that contradict each other. A page can reasonably be both an Article and a Product page with a review, but if you apply FAQPage, HowTo, Recipe, and Product to a page that's really just one of those things, the signals fight each other.

Fifth: ignoring mobile rendering. Google indexes the mobile version of your pages. If your schema lives in a component that doesn't render on mobile, or if your server-side rendering setup strips structured data from the mobile output, Google never sees it. Always confirm schema presence on mobile via the Rich Results Test's mobile option.

Sixth: treating schema as a one-time task. Schema maintenance belongs in your content operations calendar. Any time you publish new page types, update templates, or migrate platforms, structured data needs a review pass.

Sources

1. Google, 'How Google Search works: AI Overviews'
2. Google Blog, 'AI Overviews and more at Google I/O 2024'
3. Google Developers, 'Structured data markup that Google Search supports'
4. Authoritas, 'AI Overviews Study 2024'
5. Google Search Central, 'Spam policies for Google web search'
6. Semrush, 'AI Overviews Study: How Google's AI search feature works'
7. Google Search Console Help, 'Search Console overview'
8. Google Developers, 'Creating helpful, reliable, people-first content (E-E-A-T)'
9. Google Developers, 'Introduction to structured data markup in Google Search'