Definition
Structured data is machine-readable information in a web page’s source code that describes its content with a fixed vocabulary, usually Schema.org in JSON-LD format. Google uses it to understand pages and show enhanced search results called rich results. According to Google, AI features in Search don’t require it, and no effect on citations in AI answers has been shown.
Key takeaways
- Google uses structured data to understand content and show rich results. Even error-free markup doesn’t guarantee that they appear.
- Google requires neither structured data nor any special markup for AI Overviews and AI Mode.
- No effect on citations in AI answers has been shown. When 1,885 pages added schema markup, their citations barely changed.
- Mark up only what’s visible on the page. Violations can cost a page its rich results, but Google says its ranking isn’t affected.
- Google ended FAQ rich results on May 7, 2026, and how-to results in 2023. Old markup does no harm but no longer helps in Google Search.
Does schema markup help with AI search and ChatGPT?
There’s no evidence that schema markup helps you get mentioned in AI answers. Google requires neither structured data nor any special markup for AI Overviews and AI Mode. Bing says structured data may support grounding, which ties answers to sources, but doesn’t guarantee visibility. When 1,885 pages added schema markup, their citations barely changed.
| Who | What’s publicly known | Sources |
|---|---|---|
| Google Search with AI Overviews and AI Mode | not required, and no special markup; still useful for rich results; the data should match the visible text | 21, 22 |
| Google, for products and local businesses | for visibility in AI answers, Google points to Merchant Center and Google Business Profile | 21 |
| Microsoft Bing and Copilot | structured data may support grounding, which ties answers to sources, but doesn’t guarantee visibility; markup must accurately reflect the visible content; Microsoft Advertising recommends schema for AI search | 23, 24 |
| OpenAI (ChatGPT) | product results draw in part on structured metadata from providers and merchants, and merchants can submit product data; the help pages on search and shopping say nothing about Schema.org markup on websites | 25, 26, 27 |
| Anthropic (Claude) and Perplexity | nothing about structured data in their crawler documentation | 28, 29 |
The numbers in the “Sources” column refer to the source list at the end of this article. So far, studies on structured data for AI search come from vendors of SEO and analytics tools. Ahrefs compared 1,885 pages that added JSON-LD between August 2025 and March 2026 with 4,000 similar pages. Afterward, the pages with markup were cited 2.4% more often in AI Mode and 2.2% more often in ChatGPT. Neither difference was statistically significant. In AI Overviews, they were cited 4.6% less often, but both groups had already been losing citations there. Ahrefs only studied pages that AI systems were already citing heavily. Whether markup helps pages that AI systems don’t cite at all yet remains an open question.
Source: Ahrefs (external site), · vendor study · 1,885 pages, 4,000 control pages, 30 days before and after
When an AI assistant fetches a page directly, it apparently picks up little from the JSON-LD. In a searchVIU test on October 30, 2025, none of five systems named a price that appeared only in the JSON-LD. A test that OtterlyAI published in March 2026 painted a similar picture: when asked, only Gemini out of seven platforms reproduced the JSON-LD correctly. FAQ markup didn’t show an advantage in an SE Ranking analysis either. Pages with FAQ markup averaged 3.6 citations in ChatGPT, while pages without it averaged 4.2.
Source: searchVIU (external site), · vendor study · test of Oct 30, 2025, eight price sources, five AI systems
Source: OtterlyAI (external site), · vendor study · its own website, seven platforms
Source: SE Ranking (external site), · vendor study · 129,000 domains, correlation without proof of cause
For the current state of evidence on generative engine optimization (GEO), meaning visibility in AI answers, see What’s proven about GEO, and what isn’t? in our GEO guide. Our explainer on AI Overviews covers which pages are eligible as sources.
What does structured data look like?
Structured data usually sits in a page’s source code as a JSON-LD block. It uses Schema.org terms to name what the page describes, such as a company or an event. Visitors don’t see the block. Google also reads the Microdata and RDFa formats but recommends JSON-LD because it’s the easiest to maintain.
Schema.org is a shared vocabulary that Google, Microsoft, Yahoo and Yandex founded and that an open community continues to develop (Schema.org home page (external site)). That’s why structured data is also known as schema markup. The vocabulary comprises 826 types and 1,540 properties. JSON-LD is a JSON-based format for linked data. Version 1.1 has been a W3C Recommendation since July 16, 2020 (JSON-LD 1.1 (external site)). The W3C is the organization where web standards are developed (W3C (external site)).
Source: Schema.org, · version 30.1
Here’s how a small business could mark up its company details on its home page:
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Organization",
"@id": "https://www.example.com/#organization",
"name": "Example GmbH",
"url": "https://www.example.com/",
"logo": "https://www.example.com/logo.png",
"email": "info@example.com",
"address": {
"@type": "PostalAddress",
"streetAddress": "Example Street 1",
"postalCode": "12345",
"addressLocality": "Anytown",
"addressCountry": "DE"
}
}
</script>@context names the vocabulary and @type the type; the remaining lines are properties. @id gives an entry a fixed identifier that other entries can point to, for example an article to its author. Google recommends such references for entries that belong together (Google’s general guidelines (external site)).
About half of the web pages studied carry structured data. Web Data Commons, a project of the University of Mannheim, found it on 1.3 billion of 2.4 billion pages in the October 2024 Common Crawl. Common Crawl is a free archive of web pages that anyone can use. In its July 2025 crawl, the HTTP Archive found structured data on 50% of home pages and JSON-LD on 43%.
Source: Web Data Commons (University of Mannheim) (external site), · Common Crawl of October 2024, 51.25% of pages
Source: HTTP Archive, Web Almanac 2025 (external site), , updated · home pages, July 2025 crawl
Our technical SEO guide explains where structured data fits alongside crawling, indexing and page speed.
What does structured data do in Google Search?
Google uses structured data to understand content, and it can show rich results based on it: enhanced listings with details such as event dates or prices. There’s no guarantee, though. Even error-free markup only makes a page eligible. Whether a rich result appears is up to Google and depends on the search, location and device.
For the home page of a company website, Google recommends Organization markup. It helps Google understand a company’s administrative details and tell it apart from other companies (Google on Organization markup (external site)). Experts call such a clearly identifiable thing, like a company or a person, an entity. Some properties work only behind the scenes. Others help determine which logo Google shows in the knowledge panel, the information box about a company. Why consistent company facts matter to AI assistants as well is covered in the step Are your company facts clear and consistent everywhere? of our ChatGPT how-to. Our article Entity SEO explains why AI systems mix up companies.
With WebSite markup on your home page, you tell Google which site name to show above your pages in search results. Google also looks at titles and headings for this but gives the WebSite data the most weight (Google on site names (external site)). According to Google, Article markup helps it show better titles, images and dates for articles (Google on Article markup (external site)).
To show that rich results can bring clicks, Google points to case studies from major brands. Rotten Tomatoes, for example, added structured data to 100,000 pages and measured a 25% higher click-through rate on them than on pages without it. The four examples on Google’s introduction page aren’t dated.
Source: Google Search Central (external site), · case study without its own date
Which rich results does Google still show?
Google’s overview of supported features lists 25 entries, including articles, breadcrumbs, events, job postings, products and organizations. FAQ and how-to results are no longer among them. Google stopped showing FAQ rich results on May 7, 2026, and how-to rich results on September 13, 2023.
| Type | How Google shows it | Status | Sources |
|---|---|---|---|
| Organization | logo and company details, for example in the knowledge panel | supported; on the home page or a page about the company | 7, 11 |
| WebSite | preferred site name above the result | supported; home page only | 8 |
| BreadcrumbList | the page’s path in the result’s URL line | desktop only; on mobile, Google has shown only the domain since Jan 23, 2025 | 12, 13 |
| Article | better titles, images and dates | supported | 9 |
| Event | event search | available in Germany in German; only events the public can book that take place at a physical location | 14 |
| Product, Review | price, availability and review stars | supported; no stars for reviews of your own company | 11, 15 |
| LocalBusiness | details in the knowledge panel, such as business hours | supported; the business’s name and address are required | 11, 16 |
| FAQPage | questions and answers below the result | discontinued as of May 7, 2026; from August 2023 shown only for well-known government and health websites | 17, 18 |
| HowTo | step-by-step instructions | discontinued as of Sep 13, 2023 | 18 |
| WebSite with SearchAction | search box below the result | discontinued as of Nov 21, 2024 | 19 |
| Course info, estimated salary, learning video, special announcement, vehicle listing, fact check | special displays for these kinds of content | phased out from June 2025 | 20 |
According to Google, markup that it no longer uses doesn’t cause problems in Search (Google, Oct 21, 2024 (external site)). It has no visible effect either, and Google says the removals didn’t affect rankings. Many websites still carry it: in July 2025, the HTTP Archive found the SearchAction type for the discontinued search box on 28% of home pages.
Source: HTTP Archive, Web Almanac 2025 (external site), , updated · home pages, July 2025 crawl
How is structured data different from meta tags and llms.txt?
Structured data describes what’s on a page in a fixed vocabulary. The title and meta description suggest the title and text for the search result, and Open Graph tags do the same for the preview when someone shares the page. An llms.txt file offers AI agents a selection of important pages. Google Search doesn’t use it.
| Feature | Structured data | Title and description | Open Graph | llms.txt |
|---|---|---|---|---|
| Purpose | describes the page’s content and how things relate, in a fixed vocabulary | the page’s title and short summary | preview when the page is shared, such as title and image | a selection of key content, with notes, for AI agents |
| Format | JSON-LD, Microdata or RDFa with Schema.org | title element and description meta tag in the page head | meta tags with og: in the page head | Markdown file, usually in the root directory |
| Who reads it | Google, Bing and other search engines | search engines, as one source among several | social networks; Google uses og:title and og:image as one source among several | some agents and tools, not Google Search |
| Visible effect | rich results possible, not guaranteed | title and snippet in the search result, which Google also generates on its own | preview card with title and image | none in search |
| Sources | 4, 10, 23 | 34, 35, 58 | 35, 36, 37 | 21, 38 |
For a result’s preview image, Google considers the primaryImageOfPage property in schema markup as well as the og:image tag (Google on images (external site)). Our explainer llms.txt: what it does and what it doesn’t covers the file in detail.
What does “mark up only what’s visible” mean?
Whatever you mark up must be visible to visitors on the same page, and the markup must be a true representation of the page’s content. Violations can lead to a manual action from Google. The page then loses its rich results, but according to Google, its ranking in web search isn’t affected.
Google gives an example: if the JSON-LD describes a performer, the visible text must describe the same performer. You may not mark up content that visitors can’t see, fake reviews or content unrelated to the page (Google’s general guidelines (external site)). Search Console reports such cases in the Manual actions report, for example as “Structured data found on hidden content” (Search Console Help on manual actions (external site)). Bing requires the same and may ignore misleading markup (Bing Webmaster Guidelines (external site)).
For AI assistants, there’s also a practical reason. In tests where systems fetched a page directly, details that appeared only in the markup almost never made it into an answer. So prices, business hours, and questions and answers belong in the visible text first. The markup then describes them.
What are common structured data mistakes?
Typical problems are star ratings for reviews of your own company, data that contradicts the visible text and markup that only JavaScript adds. Unsuitable types and incomplete or invalid JSON-LD also cause trouble. Outdated markup such as FAQPage does no harm, though. It just no longer has any effect in Google Search.
- Stars for yourself: Reviews of your own company don’t make pages with
OrganizationorLocalBusinessdata eligible for review stars, even if the reviews come from a Google or Facebook widget. Since July 24, 2026, Google has also explicitly ruled out fake reviews and undisclosed incentivized reviews (Google on review snippets (external site)). - Contradictions: A price or a date in the markup differs from the one on the page, often after an update that only reached one of the two places. Google requires, for example, that visible dates and marked-up dates match (Google on byline dates (external site)). The fix: generate the text and the markup from the same source.
- Markup only via JavaScript: A plugin or tag manager adds the JSON-LD only in the browser. Google reads it after rendering, but for products it warns that this can make Shopping crawls less frequent and less reliable (Google on JavaScript markup (external site)). In a Vercel measurement from late 2024, the crawlers of OpenAI, Anthropic and Perplexity didn’t execute JavaScript (Vercel (external site)). There’s more on this in the step Are your services visible in the HTML? of our ChatGPT how-to.
- Unsuitable type: For
LocalBusiness, Google requires the address of the business (Google on LocalBusiness (external site)). Schema.org lists theProfessionalServicetype as deprecated (ProfessionalService on Schema.org (external site)). Google advises using the most specific type that actually fits. - Incomplete or invalid markup: If a type is missing required properties, the page isn’t eligible for that rich result. The Rich Results Test ignores comments in JSON-LD. The JSON-LD standard doesn’t support them, though, so in actual use they can cause errors (Search Console Help on the Rich Results Test (external site)).
- Wrong expectations: Many see markup as a lever for citations in AI answers, but there’s no evidence for that. You don’t need to remove outdated markup such as
FAQPageorHowTo; according to Google, it doesn’t cause problems (Google on HowTo and FAQ (external site)).
How do you test structured data?
Start with the page source, because that’s what crawlers get without JavaScript. Google’s Rich Results Test then shows whether the page is eligible for rich results, and the Schema Markup Validator checks the vocabulary. Once the page is live, Search Console reports errors and manual actions.
- View the source code: In Chrome, press Ctrl+U on Windows or Linux to open a page’s HTML source, or Command+Option+U on a Mac (Chrome Help (external site)). Search it for
application/ld+json. If the block is missing even though a plugin adds structured data, the plugin probably creates it with JavaScript. - Rich Results Test: Google’s tool tests a URL or a code snippet and shows the types it detected, with errors and warnings. It fetches the page as Google-InspectionTool, by default with the smartphone user agent. Its code explorer shows the rendered source code, meaning the page after JavaScript has run (Search Console Help on the Rich Results Test (external site)).
- Schema Markup Validator: The Schema Markup Validator (external site) checks syntax and vocabulary, including types that Google doesn’t use. It doesn’t check whether a page is eligible for rich results (Google, Dec 15, 2020 (external site)).
- Compare with the page: Check by hand that every value in the markup appears on the page and reads the same. According to Google, its quality guidelines for structured data can’t easily be tested with automated tools.
- Monitor Search Console: Under Enhancements, Search Console has a report for each rich result type with a sample of the detected items. The URL Inspection tool shows individual pages (Search Console Help on rich result reports (external site)). To set up Search Console, follow the step Have you set up Google Search Console? in our how-to on websites that don’t show up on Google.
- Measure the effect: Google recommends marking up a few frequently visited pages and comparing their performance over a few months with the period before. For citations in AI answers, Ahrefs also recommends control pages that don’t get markup during the same period.
What’s specific to Germany and the EU?
In the European Economic Area (EEA), Google is testing carousels that require structured data on a summary page, for example for hotels, local businesses, events or products. Google’s event search is available in Germany in German. For manufacturers and retailers, Schema.org added vocabulary for the EU Digital Product Passport in 2026.
Carousels: They show several entries from a single website side by side, each tile with details such as price, rating and image. To be eligible, a summary page needs ItemList markup combined with LocalBusiness, Product or Event. The feature is in beta and available only in EEA countries, Türkiye and South Africa (Google on carousels (external site)). Searchers in other countries don’t see it. The EEA comprises the 27 EU member states plus Iceland, Liechtenstein and Norway (EUR-Lex glossary (external site)). Carousels matter for retailers, event organizers and directories, not for a company website without such listings.
Events: Only events that the public can book and that take place at a physical location are eligible. Google doesn’t support purely online events (Google on events (external site)).
VAT ID: If your company has a VAT identification number, you can add it to Organization as vatID. Google calls it an important trust signal because users can look up your business in public VAT registries (Google on Organization markup (external site)). Make sure all details match your legal notice and your entries in business registers.
Digital Product Passport: With version 30.0 of March 19, 2026, Schema.org published examples for the EU’s Digital Product Passport. Version 30.1 of September 16, 2026, adds the DigitalProductPassport type, which products and offers can carry (Schema.org releases (external site)). Google’s documentation on products doesn’t mention it yet (Google on merchant listings (external site)). Which products need such a passport is a legal question that this article doesn’t answer.
From our work: how we built this website’s structured data
Every page on datafalke.de carries exactly one JSON-LD block with a single connected graph. On November 9, 2026, that was true of all 98 pages in our sitemaps. The block is also the page’s only script element. Our pages load no JavaScript, and our Content Security Policy blocks scripts. That’s a security rule the server sends with every page. The JSON-LD can still be read, because browsers don’t execute a data block like this (HTML standard (external site)).
- Fixed identifiers: The company, the website and the author each have a fixed
@id, such ashttps://datafalke.de/#organization. The company is described in full on the home page and the About page in both languages, and on our German page about Bottrop. Pages that refer to the company, for example as the publisher of an article, carry a short entry with its name and URL. The website is described in full only onhttps://datafalke.de/, the home page of the domain. The English home page in the example below refers to it with a short entry. - One source: Name, address and email come from the same configuration file as the footer of every page. A page’s name in the graph is its published title, and a term’s definition appears word for word in the text.
- What we don’t publish stays out of the markup: Our service pages carry no offer with a price as long as they don’t show a price. We don’t publish a phone number, so the graph doesn’t contain one. Links to profiles on other platforms (
sameAs) will only be added once those profiles are maintained and linked on the website. - Excluded types: Our code doesn’t allow the types
LocalBusiness(and with it its subtypeProfessionalService),AggregateRating,Review,HowToorFAQPage. Type checking rejects such entries, and our page check reports them as errors. - Checks before publishing: Our page check is a test script. It reports an error if a page carries more than one block, if a reference points nowhere or if the page’s name differs from its title. It does the same for a
BreadcrumbListthat doesn’t match the visible breadcrumb navigation. A page only goes online once the check finds no errors.
{
"@context": "https://schema.org",
"@graph": [
{
"@type": "Organization",
"@id": "https://datafalke.de/#organization",
"name": "Datafalke",
"legalName": "Mohammad Alboush – Alboush IT & Webdesign",
"url": "https://datafalke.de/",
"logo": { "@id": "https://datafalke.de/#logo" },
"founder": { "@id": "https://datafalke.de/team/mohammad-alboush/#person" },
"email": "kontakt@datafalke.de"
},
{
"@type": "WebSite",
"@id": "https://datafalke.de/#website",
"url": "https://datafalke.de/"
},
{
"@type": "WebPage",
"@id": "https://datafalke.de/en/#webpage",
"url": "https://datafalke.de/en/",
"isPartOf": { "@id": "https://datafalke.de/#website" },
"about": { "@id": "https://datafalke.de/#organization" }
}
]
}Only Search Console data will show whether Google turns this into a knowledge panel or rich results. The website has been online since November 7, 2026, so we don’t claim any effect. We’ll report the first numbers here once there’s a measurement log.
To check your own site, run this command in a terminal on Linux or macOS. It counts the JSON-LD blocks in the HTML that the server delivers, before any JavaScript runs. For our home page, the result is 1.
curl -s https://www.example.com/ | grep -o 'application/ld+json' | wc -l