Structured Data for AI Overviews

What this is about

AI Overviews combine multiple sources into an AI answer above the search results, pulling their content directly from the Google index. This is exactly the same source that classic hits use.

This article assumes the basics from Schema Markup for AI Search. The general SEO foundation is in the article on Structured Data, while the post How AI Overviews Work explains how the AI selects its sources.

Amplifier, not trigger

There is no schema type that guarantees inclusion in AI Overviews, because Google does not name any special structured data for AI functions that you absolutely have to add. Instead, helpful and reliable content on an indexed page is important.

What markup provides is pure precision. It makes it unmistakably clear which number is the price or who wrote the text. The right approach is therefore a clean implementation as a clarity amplifier. You should not hope for a magic switch.

The most effective markups for AI Overviews

Three types contribute the most to visibility. Organization is added as a foundation. All four blocks below are complete and ready to copy.

Adjust URLs, names, and values for your page. Always check them in the validator afterwards so that no errors creep in.

FAQPage: what it still does and what it no longer does

FAQ rich results no longer exist in Google Search because they were removed since May 2026 according to Google's changelog. HowTo results have also disappeared. Therefore, always check Google's gallery of supported types before you plan a display.

However, this does not mean that FAQPage is worthless. The markup still describes in a machine-readable way which text block is a question and which is the answer. Systems look for exactly this pairing when they assemble answers.

So do not expect a colorful search result, but rather a better understanding by the machine. However, the condition always remains that the questions are visible on the page.

1{
2  "@context": "https://schema.org",
3  "@type": "FAQPage",
4  "mainEntity": [
5    {
6      "@type": "Question",
7      "name": "Lösen strukturierte Daten eine AI Overview aus?",
8      "acceptedAnswer": {
9        "@type": "Answer",
10        "text": "Nein. Google nennt für AI Overviews keine zusätzlichen Anforderungen und keine besonderen strukturierten Daten. Markup verstärkt das Verständnis, garantiert aber keine Einbindung."
11      }
12    },
13    {
14      "@type": "Question",
15      "name": "Muss die FAQ sichtbar auf der Seite stehen?",
16      "acceptedAnswer": {
17        "@type": "Answer",
18        "text": "Ja. Markup ohne sichtbaren Inhalt verstößt gegen die Richtlinien für strukturierte Daten."
19      }
20    }
21  ]
22}
Copy

Article and BlogPosting: the freshness signal

The dateModified field is the most honest freshness signal you can set. For topics with a fast half-life, AI answers prefer current sources. This field tells the machine when the text was actually last changed.

The author should be linked to a real profile using sameAs. The publisher points to your Organization.

1{
2  "@context": "https://schema.org",
3  "@type": "Article",
4  "@id": "https://example.com/wissen/artikel#article",
5  "headline": "Strukturierte Daten für AI Overviews",
6  "datePublished": "2026-03-04",
7  "dateModified": "2026-09-18",
8  "author": {
9    "@type": "Person",
10    "name": "Vorname Nachname",
11    "sameAs": "https://www.linkedin.com/in/profil/"
12  },
13  "publisher": { "@id": "https://example.com/#organization" },
14  "mainEntityOfPage": "https://example.com/wissen/artikel"
15}
Copy

Product with Offer and AggregateRating

For commercial queries, this block delivers price, currency, availability, and ratings as pure data, which is much more precise than continuous text.

This data forms the basis for product-related rich results. It also ensures that pricing information is reproduced correctly in AI answers.

1{
2  "@context": "https://schema.org",
3  "@type": "Product",
4  "name": "Beispielprodukt",
5  "sku": "ABC-123",
6  "brand": { "@type": "Brand", "name": "Beispielmarke" },
7  "offers": {
8    "@type": "Offer",
9    "url": "https://example.com/produkt/abc-123",
10    "price": "129.00",
11    "priceCurrency": "EUR",
12    "availability": "https://schema.org/InStock"
13  },
14  "aggregateRating": {
15    "@type": "AggregateRating",
16    "ratingValue": "4.6",
17    "reviewCount": "38"
18  }
19}
Copy

Organization as the foundation of the brand entity

Organization anchors who is behind the source. This block belongs exactly once at the website level and gets a stable @id that all other blocks refer to. This way, the search engine always knows who owns the content.

1{
2  "@context": "https://schema.org",
3  "@type": "Organization",
4  "@id": "https://example.com/#organization",
5  "name": "Beispiel GmbH",
6  "url": "https://example.com/",
7  "logo": "https://example.com/logo.png",
8  "sameAs": [
9    "https://www.linkedin.com/company/beispiel/",
10    "https://de.wikipedia.org/wiki/Beispiel"
11  ]
12}
Copy

Which type fits which page type

Not every markup fits every page. The following table shows you the correct assignment.

SearchAction is intentionally missing from this table because Google removed the Sitelinks search box from the search results in November 2024. The markup does no harm, but it no longer provides a display.

Making entities unique: @id and sameAs

Imagine a large address book. If you enter the name Miller twice, no one knows if it is the same person until a unique customer number creates clarity. This is exactly where the bridge between markup and AI visibility lies.

A language model must recognize that the company in the legal notice and the brand in the product markup are the same thing. You need two tools for this.

First, @id gives each entity a stable identifier, such as a URL. All other blocks reference this identifier instead of repeating the name, which is why you should never change these identifiers without reason.

Second, sameAs links the entity to external profiles like LinkedIn, Wikidata, or industry directories. How you build this strategically is in our article on Entity SEO.

Validation: the mandatory step

Faulty markup costs you the eligibility for rich results. If a policy violation is added, Google can penalize the entire page. Therefore, three testing stages belong before every go-live.

1. Schema Markup Validator: Checks the syntax independently of Google.
2. Google Rich Results Test: Shows you which results Google can specifically derive.
3. Search Console: Continuously monitors the served pages and reports errors after changes.

For the start, we have two free tools. The Schema Markup Generator builds the code for you from form fields, the JSON-LD Entity Validator checks if your entities fit together. Our overview of Structured Data Tools lists further tools.

This page is itself an example. It serves TechArticle, plus FAQPage, BreadcrumbList, and Organization. If you call up the Rich Results Test for this URL, it reports three valid elements: articles, breadcrumbs, and businesses.

The FAQ markup no longer appears there because Google has turned off FAQ rich results and the test only checks types that can generate a rich result. The markup remains valid anyway, the Schema Markup Validator continues to show it.

[[IMAGE: Rich Results Test on this knowledge page, passed test with recognized types]]

The most common errors

Five errors make your markup ineffective. All five are easily avoidable.

1. Markup without visible content: A marked-up FAQ that does not exist on the page violates Google's guidelines.
2. Invented ratings: An AggregateRating without real and visible ratings can trigger manual actions.
3. Jumping dates: A dateModified that jumps to today on every visit is manipulation. Only set it for real changes.
4. Missing mandatory fields: Some types need specific properties. If a mandatory field is missing, Google discards the entire block.
5. Contradictions to the text: The visible page shows 129 euros, the markup claims 99 euros. This creates exactly the ambiguity we want to avoid.

1{
2  "@context": "https://schema.org",
3  "@type": "Product",
4  "name": "Beispielprodukt",
5  "offers": {
6    "@type": "Offer",
7    "price": "99.00",
8    "priceCurrency": "EUR",
9    "availability": "https://schema.org/InStock"
10  },
11  "aggregateRating": {
12    "@type": "AggregateRating",
13    "ratingValue": "4.9",
14    "reviewCount": "412"
15  }
16}
Copy

The sixth case is markup that the crawler never sees. This problem gets its own section.

[[IMAGE: Rich Results Test with an intentionally faulty block and the error message]]

Markup that the crawler never sees

Sometimes the code is only inserted in the browser via a Tag Manager, so everything depends on whether the crawler executes JavaScript. Google usually reads this without problems.

However, this is not documented for the crawlers behind ChatGPT and Perplexity, even though these are exactly the ones collecting the data for AI answers.

The consequence is simple. Always deliver your code server-side in the HTML so that it is reliably recognized. Our article on Rendering and Accessibility for AI goes deeper.

Rolling out templates cleanly

Markup is rarely created per page, but rather in the template. Because an error there affects hundreds of URLs overnight, you should roll out changes in four steps.

First change a single page and check it in the Rich Results Test. Then take a sample of five to ten pages of different characteristics before the full rollout follows.

Finally, observe the Search Console for two to three weeks, because only there will you see the result across the entire inventory.

What applies differently to ChatGPT, Perplexity, and AI Mode

The same rules apply to Google's AI Mode as to AI Overviews, while the situation is different for ChatGPT and Perplexity. Both access pages via their own crawlers and process the text more strongly than the markup, so structured data only helps indirectly there.

In practice, this means that server-side delivered HTML counts double. In addition, every statement from the markup must also be in the visible text, because otherwise it will not reach these systems.

How you recognize impact

Validation only tells you if the code is correct. Four measuring points show you whether it really makes a difference.

• Error rate in the Search Console: The base metric. The goal is a permanently low number of errors.
• Served rich results: Visible in the performance report. Shows whether a display results from the markup.
• Mentions in AI answers: How often your brand appears in AI Overviews. The procedure is described in our article on Measuring AI Visibility.
• Schema diff per deployment: An automatic comparison of the code blocks before and after a release. This way you find silent losses immediately.

Interplay with content and entity

Perfect markup on thin text brings no inclusion. Conversely, strong content benefits measurably because Google recognizes its quality more clearly.

At least as important as the markup is the answer structure of the text. You need clear headings and a direct answer far up front so that the search engine grasps the core immediately.

The article Answer Structure and Answer-First Principle covers this. The post Optimizing for Google AI Overviews deepens how you appear specifically as a source.

Conclusion

Structured data makes your content unambiguous for machines, nothing more. It does not trigger a display in AI Overviews, but it takes the interpretation work away from Google. This increases the chance of being understood and named as a source.

Four types are enough for most websites. These include Article with the real modification date, FAQPage for questions that are also visible on the page, Product with price and availability, and an Organization with a stable identifier that all other types refer to.

The implementation is decisive. The markup belongs server-side in the HTML and must match what visitors actually see.

Before going live, you check it in three stages. You control the syntax in the Schema Markup Validator, the eligibility in the Rich Results Test, and the impact in the Search Console. After that, you observe over a few weeks whether impressions, clicks, and mentions in AI Overviews change, because only these numbers show whether the effort was worth it.