TL;DR
Structured data is useful infrastructure. It is not a proven way to get cited by AI engines.
Google says structured data is not required for its generative AI features and no special schema is needed.
Two studies found no clear citation gain from schema. Both have limits.
No verified ranking of schema types exists for AI. Start with the types that describe what your page really is.
Mark up only what is visible, use JSON-LD, connect your entities and test the markup.
Track a fixed prompt set before and after a change, and log every other edit.
Structured data helps machines read your pages, but the evidence that it makes AI engines cite you is weak. Google says it is not required for its generative AI features, and two studies of AI citations found no clear gain from adding it. Keep it accurate and consistent, and do not expect it to fix thin content.
1What is structured data, and what role does it play in AI search?
Definition: structured data
Structured data is machine-readable markup, usually JSON-LD using the schema.org vocabulary, that states facts about a page in a fixed format, such as who published it, what a product costs or which organization it describes.
Its proven job is in classic search, where valid markup can make a page eligible for rich results. Google recommends JSON-LD and also supports Microdata and RDFa, and says valid markup does not guarantee a rich result (Google Search Central).
For AI search, the role is less clear. Google’s guide says structured data is not required for generative AI features in Search and that you do not need special schema.org markup for them. It still recommends structured data as part of overall SEO (Google Search Central). A Microsoft Bing representative said on stage that schema helps Microsoft’s LLMs understand content, but the report says it is unclear whether Gemini, OpenAI, Perplexity or others use it (Search Engine Land).
2What does the evidence say about schema and AI citations?
| Source | What it found | Limits |
|---|---|---|
| Google guide | Structured data is not required for generative AI features; no special schema needed | Covers Google only |
| Microsoft Bing remark | Schema helps Microsoft’s LLMs understand content | A spoken remark, not a published test. Other engines have not confirmed use |
| Citation coverage study | Domains with more schema coverage were not cited more often by OpenAI, Gemini or Perplexity | Observational, measured presence only, not type or quality |
| Before and after study | Pages that added JSON-LD showed no clear citation gain against matched controls | Only heavily cited pages, other edits overlapped, 30 day window |
Quotable stat
In a before and after test of 1,885 pages that added JSON-LD, citations in Google AI Overviews, AI Mode and ChatGPT showed no clear gain against matched controls (TechWyse summary).
Read both studies with care. The first compared domains and shows correlation at most. The second tracked pages that already had at least 100 AI Overview citations, so it says nothing about pages with no AI visibility, and schema changes often came with other edits. One experiment it mentions found chatbots ignored JSON-LD when fetching a page directly, which does not rule out a role in indexing (Search Engine Land summarizes the wider debate and calls schema “infrastructure, not a magic bullet”).
The fair reading: schema has not been shown to raise AI citations, and it has not been shown to hurt. There are no controlled studies that settle it.
3Which schema markup is most important for AI search visibility?
No verified evidence ranks schema types for AI search. A practical starting set is the types that describe what the page is, which also support classic SEO. Treat the list as working practice, not a proven order.
| Type | Use it for | Source |
|---|---|---|
| Organization | Your company name, logo and official profiles on key pages | Practice |
| Article or BlogPosting | Guides and posts, with author and publisher | Practice |
| Person | Real authors and reviewers with links to background | Practice |
| Product, Offer, Review | Product pages, only for facts visible on the page | |
| Service | Service pages that state what you offer | Practice |
Source tags: Google comes from Google documentation, Research from a published study, Practice is working practice.
4What are the best practices for schema markup in AI search?
| Practice | What to do | Source |
|---|---|---|
| Match the page | Mark up only what users can see, and make sure the page really says it | |
| Use JSON-LD | It is Google’s recommended format. Microdata and RDFa also work | |
| Include required fields | Items missing required properties are not eligible for rich results | |
| Stay accessible | Do not block marked-up pages with robots.txt or noindex | |
| Never mislead | Do not invent reviews or mark up unrelated content. Misuse can lead to a manual action | |
| Connect entities | Link your Organization, authors and pages with stable IDs so they describe one entity graph | Practice |
| Keep it current | Update markup when prices, authors or facts change | Practice |
| Fix the page first | Clear headings, direct answers and original detail matter more than markup | Practice |
Do not expect schema to offset weak content or low authority. It labels what a page already says.
5How should ecommerce brands use structured data for AI visibility?
For online stores, Product markup should carry the same price, availability and review facts a shopper sees on the page. Keep them in sync with your catalog, because wrong or stale markup is a trust risk and can break eligibility.
We found no verified study showing that Product schema gets products cited by AI engines. Treat it as accuracy and eligibility work, and measure the effect on your own prompts rather than assuming one.
6How do you test schema markup?
- Validate. Run the page through Google’s Rich Results Test and the Schema Markup Validator, and fix errors.
- Monitor. Watch the rich results reports and the Manual Actions report in Search Console.
- Fix a prompt set. Write the questions buyers ask and keep them unchanged.
- Measure before and after. Record whether your pages are cited, repeat each prompt several times and log every other change made to the page.
- Judge slowly. A short window and a single run can mislead. Look for a repeated pattern.
What only Wellows adds
Testing schema means seeing which engines cite you before and after a change. LLM Citations tracks ChatGPT, Gemini, Perplexity, Google AI Overviews and Google AI Mode, refreshes every 24 hours and shows the pages behind each answer. Our generative engine optimization platform brings that tracking together.
Before editing markup, find the pages worth the effort. The free Content Decay Tool connects to Search Console and flags pages whose performance is falling, so you can refresh the ones that matter first.
Conclusion
Structured data is worth doing for accuracy and classic search eligibility, but it has not been shown to lift AI citations. Mark up only what is visible, use JSON-LD, connect your entities, test it, and measure the effect on a fixed prompt set before you claim a result.
It has not been shown to. Google says structured data is not required for its generative AI features, and two studies of AI citations found no clear gain from adding it. It is still worth using for accuracy and classic search eligibility.
No verified evidence ranks schema types for AI search. A sensible start is the types that describe the page, such as Organization, Article or BlogPosting, Person and, for stores, Product. Mark up only what is visible on the page.
Neither has confirmed it. A Microsoft Bing representative said schema helps Microsoft’s LLMs understand content, but other engines have not said whether they use it.
Use JSON-LD, mark up only visible content, include required fields and keep the markup accurate and current. Do not block marked-up pages or invent reviews. Clear, original content matters more than markup.
Validate it with Google’s Rich Results Test and the Schema Markup Validator, then track a fixed set of prompts before and after the change. Repeat each prompt several times and log any other edits.
We found no verified study that shows it. Keep Product markup identical to the visible price, availability and reviews, and measure the effect on your own prompts.