What structured data is and why it matters across paradigms
Structured data is code added to a page that declares its meaning in a fixed vocabulary, so machines no longer have to infer it from the text. A homepage with Organization markup does not hope a search engine recognises the business name, address and social profiles. It states them.
The vocabulary almost everyone uses is schema.org, which was started by Google, Microsoft, Yahoo and Yandex, according to the schema.org project page, and formalised academically by Guha et al. (2016). It defines types such as Organization, LocalBusiness, Person, Article, Product and Event, each with its own properties. Google's introduction to structured data recommends the JSON-LD format where a site's setup allows it, because it sits in a separate script block and is easier to maintain than inline Microdata or RDFa.
That shift from inference to declaration is why the same markup can serve three goals at once. Classic SEO uses it for rich results. Answer engines use it to pick out facts. Generative AI systems benefit when the entities behind a page (company, author, location) are named consistently. The returns differ per layer, and some of the most quoted ones have shrunk since 2023, so it pays to look at each separately.
Nickel et al. (2016), in their review of knowledge graph embeddings, explain the deeper mechanism: AI systems that represent knowledge as entity-relation-entity triples evaluate content for its contribution to machine-comprehensible knowledge structures. Structured data makes that contribution explicit, converting unstructured text into the entity-relational format that knowledge graphs and AI retrieval systems prefer.
What the research says, and what it doesn't
Recent research links structured data to AI citations more consistently than almost any other technical factor. The most detailed source is a comparative study by Santhosh Kumar Iyappan (2026) in the GOYBO International Journal of Marketing Intelligence, based on a corpus of 162 analysis units and 100 prompt-response pairs across five AI platforms.
Its main findings on structured data:
| Finding in Iyappan (2026) | Result |
|---|---|
| Structured data implementation vs AI citation frequency (Table 6) | Strong positive correlation (AEO, GEO) |
| FAQ schema vs featured snippet inclusion (Table 6) | Strong positive correlation (AEO) |
| AI citation rate, context-rich long-form content (Table 4) | 92% |
| AI citation rate, entity-optimised content (Table 4) | 89% |
| AI citation rate, structured data-heavy pages (Table 4) | 85% |
| AI citation rate, FAQ-formatted pages | 67% |
| AI citation rate, keyword-focused articles | 41% |
| Structured data sensitivity by platform (Table 7) | Very high: Gemini. High: Claude, Perplexity, Copilot. Moderate: ChatGPT |
Structured data is therefore correlated with citation frequency in two paradigms at once, and four of the five major AI platforms rate their sensitivity to it as high or very high. Structured data-heavy pages reach an 85% citation rate, behind only entity-optimised (89%) and context-rich long-form content (92%). The correlations are observational, and AI platforms change their behaviour faster than studies are published, so the platform profiles are worth rechecking over time.
Kargaev (2026), whose SSRN paper quantifies how ranking factors diverge between traditional and generative search, reaches a compatible conclusion from a different angle. The paper identifies schema markup as "the technical bridge between SEO accessibility and AI readability": the signal that lets content optimised for traditional search also perform in AI-generated responses.
The same study also found a very strong correlation between topical authority and visibility across SEO, AEO and GEO. Structured data is one cross-paradigm lever, not the only one.
Google's own position adds a second check. Its guide to generative AI features in Search says AI Overviews and AI Mode run on the same ranking and quality systems as normal Search, and that structured data is not required to appear in them, while it remains useful for rich results. Markup supports good content; it does not replace it.
The Guha et al. (2016) paper on schema.org explains why the vocabulary carries weight with answer engines: the vocabulary was developed in direct collaboration with the search engine developers who built the extraction systems, which makes it the native language of those systems rather than a compatibility layer.
How structured data works in classic SEO
In classic search, structured data makes a page eligible for rich results and helps Google understand the entity behind the site. Eligibility is not a guarantee: Google decides per query whether to show an enhanced result, and incomplete markup is ignored.
When rich results do appear, they can change click behaviour. Google's introduction to structured data cites Rotten Tomatoes, where pages with structured data had a 25% higher click-through rate than pages without, and Nestlé, where pages shown as rich results had an 82% higher click-through rate. These are publisher case studies, not averages, but they show the mechanism.
Organization markup works differently. Google's Organization documentation says it helps Google understand your administrative details and disambiguate your organisation in search results, and that the logo property can influence visual elements such as the knowledge panel. There are no required properties; you add those that apply, such as name, url, logo, address, contactPoint, sameAs and foundingDate.
What changed: FAQ, HowTo and Speakable
Three schema types that older guides present as quick wins have lost most of their visible effect in Google Search. This is the part most structured data advice still gets wrong.
Google announced in August 2023 that FAQ rich results would only be shown for well-known, authoritative government and health websites. HowTo rich results were first limited to desktop and then removed entirely in September 2023. Speakable markup is still in beta and, according to Google's documentation, works only for users in the US with Google Home devices set to English.
| Schema type | Rich result in Google today | Value beyond rich results |
|---|---|---|
| Organization | Logo and knowledge panel details can be influenced | High: core entity declaration for every system |
| LocalBusiness | Business details for local queries | High: machine-readable address, hours, area |
| Article / BlogPosting | Article details such as headline, image, dates | High: author and date attribution |
| Product / Offer | Price, availability, reviews | High for shops |
| BreadcrumbList | Breadcrumb trail in results | Medium: clarifies site structure |
| FAQPage | Only for authoritative government and health sites | Low to medium: clear Q&A pairs, no snippet for most sites |
| HowTo | Deprecated | Low |
| Speakable | Beta, US English only | Low outside US news publishers |
FAQ markup is not useless. Clearly separated question-answer pairs are still easy for any system to parse, and Iyappan's correlation finding concerns FAQ structure as much as markup. But for most businesses it no longer earns an accordion in the results, so it belongs lower on the list than older guides suggest.
How structured data helps in GEO and AI search
In generative search, structured data works mainly as an identity layer: it states consistently who the organisation is, where it is based, who wrote a page and when. That reduces the room for an AI system to confuse a business with a namesake or attach outdated details to it.
Kargaev (2026) and Iyappan (2026) both identify brand entity signals as the dominant authority factor in GEO. Organization markup is the technical implementation of those signals: it turns an unverified web presence into machine-readable entity verification.
How much each AI platform reads JSON-LD directly is not publicly documented. Much of the effect runs through the search indexes these systems draw on, which is why the same markup that helps Google also helps answers built on top of search results. For the wider picture of how AI engines choose sources, see our guide to generative engine optimization.
The basics are often missing. In our research across 310 websites of Dutch SMEs, 52% had no Organization or LocalBusiness schema, 89% did not state their address in machine-readable form, and 41% had no machine-readable date. Those are exactly the facts an AI system needs to name and recommend a local business with confidence.
Article or BlogPosting markup closes a second gap. With headline, author, datePublished, dateModified and publisher declared, a page carries its own attribution. Linking the author to a Person entity with a name, job title and profile URL makes that attribution traceable. An AI system drawing on such a page can cite the author, the publication and the date, the attribution chain that Gao et al. (2023) identified as essential for citation-capable language model generation.
Which schema types to implement first
Organization markup on the homepage gives the broadest return for almost every business, because every search and AI system needs to know who the site belongs to. After that, the order depends on your business model.
- Organization (homepage). Name, url, logo, description, address, contactPoint and sameAs links to your real profiles. Keep it identical to your Google Business Profile and other listings.
- LocalBusiness or a subtype (if you serve an area). Address, opening hours, areaServed and geo coordinates. Subtypes such as LegalService or MedicalBusiness add precision. Our guide to local visibility with AI covers the rest of the local picture.
- Article or BlogPosting (all content). Author linked to a Person entity, plus dates filled automatically from the CMS so they stay correct after updates.
- Product and Offer (shops). Price, currency, availability and brand, matching what the page shows.
- BreadcrumbList and sector-specific types. Service, SoftwareApplication or similar where they describe the page accurately.
Use FAQPage where a page has a genuine FAQ, but do not expect rich results from it unless you are a government or health site.
Common structured data mistakes
The most damaging mistake is markup that contradicts the visible page or reality. An old address, a wrong business type or a dead social profile in Organization markup is structured misinformation: it gets declared with the same confidence as correct data.
Other frequent problems:
- Validating once. Theme updates, plugin conflicts and URL changes break markup without anyone noticing. Check the rich result reports in Search Console and retest key templates after every CMS update.
- Markup without matching content. Google's guidelines expect structured data to describe what users can see. Hidden FAQs or reviews that exist only in JSON-LD risk being ignored or flagged.
- Missing the homepage. Many sites have Article markup on blog posts from a plugin but nothing on the homepage, where the organisation is defined.
- One type where several apply. A service page can carry Organization (via the site-wide graph), Service, BreadcrumbList and, where relevant, FAQPage together.
How to measure the effect
Structured data shows its effect in different places for each layer, so no single report covers it. Plan for three checks.
For classic SEO, Search Console's rich result reports show which types Google detects, which pages are valid and where errors sit. The performance report can be filtered by search appearance to compare clicks and click-through rate for pages with rich results. For a quick check during development, use Google's Rich Results Test and the Schema Markup Validator at validator.schema.org.
For answer engines, test the questions your customers actually ask and note which of your pages are quoted. The differences between AEO and GEO explain why the same page can do well in one and not the other.
For AI search, run a fixed set of prompts in ChatGPT, Perplexity and Gemini each month and record whether your business is named, and whether the details (address, services, founding year) are correct. Errors in those answers often point straight to inconsistent or missing entity data. Tools such as Otterly.ai and Peec AI can automate this tracking.
Frequently asked questions
What is structured data in SEO?
Structured data is markup that declares what a page means, using a shared vocabulary that machines can read. In practice that vocabulary is schema.org, written as JSON-LD in a script block. It tells search engines and AI systems which entity a page is about, who wrote it and when, rather than leaving them to infer it.
Does structured data improve Google rankings?
Structured data does not directly improve rankings. Its value in classic search lies in eligibility for rich results, which can raise click-through rate, and in helping Google understand and disambiguate your organisation. A page with perfect markup and weak content will still rank poorly.
Do I need schema markup to appear in AI Overviews?
No, Google states that structured data is not required for its generative AI features. AI Overviews and AI Mode use the same ranking and quality systems as normal Search. Markup still helps by making your organisation, authors and dates explicit, and it remains necessary for rich results.
Is FAQ schema still worth adding?
FAQ schema is worth adding only where a page has a real FAQ, and with low expectations. Since August 2023, Google shows FAQ rich results only for well-known government and health websites. The clear question-answer structure still helps any system parse the content, but most sites will not see an accordion in the results.
Which schema types matter most for AI search?
Organization and, for local businesses, LocalBusiness matter most, because they define who you are and where you operate. Article or BlogPosting with a linked author and correct dates comes next. Together they give AI systems the identity and attribution facts they need to name a business correctly.
How long does it take for structured data to have an effect?
There is no reliable public figure for how long structured data takes to affect AI answers. Google first has to recrawl the page before markup can influence rich results, which depends on how often your site is crawled. Validation tools show errors immediately, so you can at least confirm the markup is correct on the day you publish it.
What is the difference between structured data and schema markup?
Structured data is the general concept of machine-readable meaning on a page; schema markup is structured data written with the schema.org vocabulary. JSON-LD, Microdata and RDFa are three formats for writing it, and Google recommends JSON-LD. In everyday SEO practice the terms are used interchangeably.
To see which entity data your site is missing and where your markup contradicts itself, audit your website for free.
