No. Schema markup does not currently have a demonstrated direct effect on whether an AI tool cites your page.

Schema.org markup was built to provide machine-readable information about entities, products, organisations, events, reviews, addresses and other page content. It remains useful for search engines and established search features.

What has not been demonstrated is that an AI assistant gives a page additional citation weight because the information also appears inside valid Schema.org markup.

If an agency says adding schema will directly improve your visibility in ChatGPT, Perplexity, Copilot, AI Overviews or AI Mode, the important question is not whether an AI tool can read JSON-LD.

The question is whether valid schema markup itself caused an increase in citation or retrieval probability after the underlying information, page content and other variables were controlled.

At the moment, there is no strong public evidence showing that causal relationship.

This distinction is the part that often gets lost in discussions about AI SEO.

JSON-LD is text inside the HTML of a page. An AI system may be able to retrieve that HTML, read strings inside a script block and use those strings when creating an answer.

That does not prove the AI system recognised the markup as valid Schema.org vocabulary.

It also does not prove the system trusted the information because it was structured data, ranked the page more highly because of it, or selected the page as a citation because schema was present.

Those are different behaviours. Demonstrating the first does not prove the second.

SEO professional Mark Williams-Cook tested one of the most common arguments used to support schema markup for AI visibility.

He created a fictional business called DUCK YEA T-SHIRTS. The visible page content did not contain the company's address. Instead, the address appeared inside a JSON-LD script.

There was one important twist: the structured data was deliberately nonsense.

{
  "@type": "MallardEnterprise",
  "flockName": "DUCK YEA T-SHIRTS",
  "nestingGrounds": {
    "@type": "LilyPadAddress",
    "reedNumber": "77",
    "puddle": "The Muddy Bank",
    "region": "South Pondshire",
    "featherCode": "DK99 YEA",
    "country": "United Queendom"
  }
}

MallardEnterprise, LilyPadAddress, flockName and nestingGrounds are not Schema.org types or properties.

Williams-Cook then asked ChatGPT and Perplexity for the company's address. Both were able to return the address contained inside the fake JSON-LD.

“Schema, good. Repackaging the basics as some magical new GEO formula, bad.”

That result matters because valid Schema.org semantics are not required to explain what happened.

If an AI tool can retrieve the same information from completely invented structured-data labels, the simpler explanation is that the system can read text contained in the page HTML.

The experiment does not show that schema is useless. It shows why an AI system returning information from JSON-LD is not enough evidence to say Schema.org markup caused the result.

HyperSpace's view: The DUCK YEA test attacks the causal claim, not structured data itself. It shows that text being retrieved from JSON-LD does not prove Schema.org semantics were responsible.

A common schema demonstration follows a simple process.

1. Place an address, fact or product attribute only inside valid schema markup.
2. Ask an AI assistant for that information.
3. Receive the correct answer.
4. Claim the AI system is using Schema.org markup.

The problem is that this test has no meaningful control.

To establish whether schema caused the result, the experiment needs to separate the information itself from the Schema.org vocabulary surrounding it.

Williams-Cook's test does that in a simple way. The information remains inside JSON-LD, but the Schema.org vocabulary is replaced with nonsense. The AI system can still retrieve the answer.

That means Schema.org semantics are not necessary to produce the observed result.

Citation behaviour is a much harder test

AI visibility also involves more than reading information from a page.

A page may need to be accessible to the system, discovered or retrieved for a question, selected as useful supporting material and then chosen for attribution or citation.

A model finding an address inside one HTML document tells us very little about that wider process.

A credible schema-for-AI experiment therefore needs to measure citation behaviour, not merely whether an AI assistant can repeat a fact that exists somewhere in the page source.

Google's current guidance is unusually clear.

Google Search Central states that structured data is not required for generative AI search and that there is no special Schema.org markup websites need to add for Google's generative search features.

Google still recommends structured data as part of normal SEO because valid markup can support established search features such as rich results.

That distinction matters.

Structured data can have a documented role in Google Search without being a direct AI citation lever.

Google's guidance for AI Overviews and AI Mode continues to point website owners towards familiar search fundamentals: crawlability, indexability, useful textual content, internal linking, page experience and accurate information.

Google also advises that structured data should match the visible text on the page.

There is no special schema fast lane into AI Overviews or AI Mode.

OpenAI

OpenAI's publisher guidance explains how websites can become eligible to appear in ChatGPT search and how publishers can allow OAI-SearchBot to access their content.

OpenAI also explains that ChatGPT search results may contain citations and that search results and citations can sometimes be incomplete, outdated or incorrect.

What its public publisher guidance does not identify is Schema.org markup as a direct ChatGPT citation signal.

That does not prove OpenAI never processes structured data anywhere in its systems. It means there is no published OpenAI guidance showing that adding schema alone increases the probability of a citation.

Microsoft and Bing

Microsoft has gone further in discussing structured content for AI-powered search.

Bing has said that clearly structured content, including schema-marked product information, can help AI systems interpret and summarise information. That is a reasonable reason to maintain accurate structured information.

Microsoft has also introduced AI Performance reporting in Bing Webmaster Tools. Website owners can see citation counts, cited URLs, citation trends and sampled grounding queries across supported Microsoft AI experiences.

This gives SEOs something far more useful than screenshots of one prompt: actual citation measurement.

But Microsoft's public material still does not provide a controlled experiment showing that adding Schema.org markup to otherwise identical content produces a measurable increase in citations.

Saying structured content can assist interpretation is not the same claim as saying schema directly increases AI visibility.

No. Removing schema would solve the wrong problem.

Schema markup still has documented uses in search.

Depending on the page and supported search feature, structured data can help describe products, organisations, local businesses, events, reviews, recipes and other entities in a standardised format.

Google uses supported structured data to determine eligibility for a range of rich search-result features.

The correct response to weak AI claims is not to abandon structured data. It is to stop assigning structured data a job that has not been demonstrated.

• Keep valid schema where it supports documented search features.
• Keep schema accurate and aligned with visible page information.
• Maintain entity and product data where it helps search platforms understand your site.
• Do not sell schema as an AI citation switch without evidence showing a causal citation lift.

HyperSpace's view: Schema remains part of good technical SEO. The problem starts when an established SEO technique is renamed as an AI optimisation tactic and sold with claims the evidence does not support.

If an SEO agency, GEO consultant or AI visibility platform tells you schema will increase AI citations, ask them to prove the markup itself caused the result.

• Where is the control? Ask for a comparison between pages containing the same information where the meaningful difference is the presence or absence of valid schema.
• Was citation rate measured? An AI assistant returning a fact is not the same as selecting and citing the source.
• Was the test repeated? Results should be tested across multiple prompts, sessions, dates and relevant AI systems rather than a single successful response.
• Was valid schema compared with fake JSON-LD? If invented properties produce the same behaviour, Schema.org semantics are not required to explain the result.
• Were the underlying words controlled? The information available to the AI system should remain consistent between the test and control.
• Can they show a causal citation lift? Screenshots, crawler visits and successful information extraction do not establish that schema increased AI visibility.

If those controls are missing, the test may demonstrate that an AI system can access information inside JSON-LD. It does not demonstrate that schema markup improved AI visibility.

If the goal is to increase visibility in AI-assisted search, focus first on the parts of the process that platforms themselves say matter.

Your important pages need to be accessible to search and AI crawlers where relevant. The content needs to be indexable, clear, useful and closely matched to the questions people are asking.

Important claims should be supported with evidence. Entities and relationships should be explained clearly in the visible content rather than hidden only inside markup.

Your website should also give search systems a good reason to retrieve a particular page for a particular question.

• Crawlability: make sure important content can actually be accessed by the search and AI systems you want visibility in.
• Indexation: maintain technically sound pages that search engines can find and index.
• Clear visible content: state important facts, entities, relationships and answers directly on the page.
• Topical relevance: create pages that satisfy specific questions and search intents rather than writing vague content around broad topics.
• Evidence: support important claims with first-party data, credible external sources, examples and expert input where appropriate.
• Authority: build genuine third-party recognition, mentions, links and references around the organisation and its subject matter.
• Measurement: measure actual AI citations, referral traffic and cited pages where platforms and analytics tools provide that data.

Schema can sit underneath that work as part of sound technical SEO.

It should not be mistaken for the strategy itself.

• Mark Williams-Cook: DUCK YEA T-SHIRTS structured-data experiment and analysis of schema claims in GEO.
• Google Search Central: guidance for generative AI features in Google Search and structured data requirements.
• Google Search Central: structured data documentation covering its established role in Google Search and rich results.
• OpenAI: publisher and ChatGPT search guidance covering OAI-SearchBot, search eligibility and citations.
• Microsoft Bing: AI Performance reporting in Bing Webmaster Tools and guidance for content appearing in AI-generated answers.