Discussion
  • AI Search
  • Structured Data

Is there credible evidence that schema markup increases citation in LLM answers, or does retrieval rely on visible text? What would convince you?

Published 1 reply

Discussion starter: an AI search question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.

Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.

1 reply

  • admin
    Admin
    Original poster

    There is no credible public evidence that schema markup increases how often content is cited in LLM answers. The best-supported view is that retrieval and answer generation rely mainly on visible text: systems fetch pages, extract text, and summarise or quote it. Google’s May 2026 guide to optimising for its generative AI features says special schema is not needed for those features. Microsoft’s Fabrice Canel has said structured data helps Bing’s systems understand content, which is plausible for entity understanding and product data, but that is not the same as evidence that markup raises citation rates. What would convince me is a controlled comparison or explicit platform documentation, described below.

    Why visible text is the likely driver

    • Answers quote text. What an assistant cites or paraphrases is a passage of content, not a JSON-LD block.
    • Many pipelines strip markup. Text extraction for retrieval commonly discards scripts, including JSON-LD, unless it is deliberately parsed.
    • Search indexes feed many assistants. Where assistants use search results to choose sources, they inherit ranking systems in which structured data is used for search features and understanding rather than as a direct ranking boost.

    Where structured data plausibly helps anyway

    Use Why it matters
    Product data: price, availability, identifiers Feeds shopping surfaces and merchant listings; helps systems get facts right
    Organization and entity identity Helps connect a brand to its profiles and facts
    Dates, authors and article type Helps systems judge recency and authorship
    Search features that still exist Rich results that affect clicks in ordinary search

    None of these requires believing that markup makes an LLM cite you more. They are reasons to implement accurate structured data on their own merits.

    What would convince me

    1. A controlled experiment: many matched pages with identical visible content, randomly split, with markup added to half; prompts run repeatedly across several assistants before and after; citation rates compared statistically, controlling for crawl timing.
    2. Platform documentation stating that a specific assistant parses a specific schema type and uses it in retrieval or ranking of sources.
    3. Consistent replication across sites and topics, not a single case study where markup was added alongside other changes.

    Correlation studies showing that cited pages often have schema are not persuasive on their own, because well-maintained sites tend to have both structured data and good content.

    Hypothetical example:

    A publisher adds detailed Article, Person and Organization markup across its site and, in the same quarter, improves author pages and updates its most important guides. Its AI citations rise. An agency presents the result as proof that schema drives citations. The publisher runs a follow-up test instead: markup removed from a random half of 400 stable articles, content unchanged. Over two months, citation rates for the two halves are indistinguishable. The markup stays, for search features and entity clarity, but the publisher stops claiming it drives citations.

    Watch for this:

    Some vendors sell “AI schema” packages, including markup types that no platform documents using. Accurate markup that describes visible content is harmless and sometimes useful; markup that describes content not on the page breaks Google’s structured data guidelines and can lead to manual action.

    Making text easy to retrieve and quote

    If visible text is what gets used, a few writing habits matter more than markup. State key facts in complete sentences that make sense on their own, with the subject named rather than implied, so a passage lifted out of context still reads correctly. Put the answer to the question a section addresses in its first sentence. Keep numbers next to their source and date. Use headings that say what the section contains. And make sure the important content is in the server-rendered HTML, because some crawlers used by AI systems do not execute JavaScript. These habits help readers too, which is the best sign that they are not a trick.

    Why the question keeps coming back

    Structured data is appealing because it is concrete: a developer can add it in a sprint and a validator confirms it works. Improving the text that gets cited is slower and harder to measure. That asymmetry makes markup an easy thing to sell and to report as progress. It also explains why correlation studies keep appearing: sites that invest in markup usually invest in content and technical quality too, so cited pages often have both. Until someone separates the two with a controlled test, the honest position is that markup is a sensible part of a well-built site, and that its specific contribution to AI citations is unknown. Say that plainly when stakeholders ask, and budget the work on the benefits that are documented.

    What to prioritise instead

    If the goal is AI citations, spend the effort on the text: clear statements that answer specific questions, accurate and current facts, a visible date and author, and content that others reference. Keep structured data accurate and consistent with the page, because it supports search features and entity understanding, but treat it as hygiene rather than the lever.

    Keep implementing accurate structured data for the benefits it demonstrably has, assume retrieval and citation depend mainly on visible text until a controlled test or platform documentation says otherwise, and be sceptical of anyone who presents correlation as proof.

    Need help with this on your own site?

    Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.

    See SEO services

Add a reply

Replies are open to approved members. Log in or apply to join.