What role does structured data play in answer engine optimization?

Structured data gives crawlers a machine-readable description of what a page contains, most often schema.org vocabulary embedded in the page source as JSON-LD [1][5]. Google defines it as "a standardized format for providing information about a page and classifying the page content" [1]. Schema.org, the shared vocabulary behind nearly all of it, was founded by Google, Microsoft, Yahoo, and Yandex, and reports that over 45 million web domains carry more than 450 billion schema.org objects [5]. That scale explains why schema appears in every AEO checklist. What the checklists usually skip is that the major answer engines consume it in very different ways, and some appear not to consume it at all.

The split runs along infrastructure lines. Engines that inherit a mature search index treat schema as a documented input: Google parses it for rich results and entity understanding [1], and Microsoft recommends it in its official guidance for getting cited in AI answers [6]. Engines built directly on LLM retrieval are a different story. OpenAI's crawler documentation for ChatGPT search covers robots.txt access without ever mentioning structured data [8], Perplexity's crawler documentation is just as silent [9], and testing shows that when an assistant fetches a page live during a conversation, it reads the visible text and skips JSON-LD entirely [12].

Even where schema is definitely consumed, it is not an AI visibility switch. Google states there is "no special schema.org structured data that you need to add" to appear in AI Overviews or AI Mode, because a page only needs to be indexed and eligible for a snippet [3]. The largest controlled test to date, covering 1,885 pages that added JSON-LD, measured citation changes small enough to classify as noise [10]. So the honest role of structured data in AEO is infrastructure: it supports the search layer most answer engines are built on and keeps entity facts unambiguous, but it has not been shown to directly earn AI citations.

Google consumes structured data more thoroughly than any other platform, and it is also the platform most direct about schema being optional for its AI surfaces. On the search side, Google's gallery currently supports roughly 30 structured data features, including Article, Product, Organization, and Breadcrumb, each of which can trigger an enhanced result appearance [2]. The documented payoff is real: Google's own case studies report that Rotten Tomatoes measured a 25% higher click-through rate on pages with structured data and Food Network saw a 35% increase in visits after marking up 80% of its pages [1]. Google also uses the markup to gather information about entities on the web [1].

For AI features, the guidance is unambiguous. Google's documentation states: "You don't need to create new machine readable files, AI text files, or markup to appear in these features," and adds that there is no special schema.org structured data to add [3]. Eligibility for AI Overviews and AI Mode requires only that a page is indexed and eligible to appear in Search with a snippet, with "no additional technical requirements" [3]. Limits on AI use of content run through snippet controls and the Google-Extended token, not through markup [3].

Google has also been narrowing rich results rather than expanding them. In August 2023 it restricted FAQ rich results to "well-known, authoritative government and health websites" and began winding down HowTo rich results [4]. The FAQPage markup remains valid vocabulary, but for most sites it no longer produces any visible search feature [4].

Microsoft is the only major platform whose engineers have said on record that schema markup feeds their LLMs. At SMX Munich in March 2025, Fabrice Canel, a principal product manager at Bing, confirmed that schema markup helps Microsoft's LLMs, including Copilot, understand site content [7]. He also connected freshness to AI behavior: "Gen AIs value fresh content in particular, partly as a reference check of their LLM training data," and recommended pushing updates through the IndexNow API [7].

Microsoft followed the conference statement with written guidance. An October 2025 post by Krishna Madhavan, a principal product manager on the Bing team, explains that "schema is a type of code that helps search engines and AI systems understand your content" and that it can label content as a product, review, FAQ, or event, "turning plain text into structured data that machines can interpret with confidence" [6]. Markup that Bing parses at indexing time can reach Copilot's answer layer, which makes Bing the strongest documented case for schema in AEO [6][7]. How much weight that deserves depends on how much of your audience uses Copilot and Bing-powered experiences.

Neither OpenAI nor Perplexity mentions structured data anywhere in its publisher-facing documentation, which is a meaningful absence given how specific both are about crawler access. OpenAI's stated requirement for ChatGPT search visibility is access, not markup: "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers" [8]. Perplexity's documentation likewise covers PerplexityBot, which indexes pages for search and is "not used to crawl content for AI foundation models," plus the on-demand Perplexity-User fetcher, and offers no schema guidance of any kind [9].

Direct testing goes further than the silence. In October 2025, the analytics firm searchVIU planted product prices in eight formats on a test page, including visible HTML, JavaScript-rendered text, JSON-LD, Microdata, and RDFa, then asked each assistant to retrieve them [12]. No system extracted a fact that existed only in JSON-LD during a live fetch. Claude found none of the hidden formats, ChatGPT recovered three of eight prices, and Gemini recovered four of eight as the only assistant that rendered JavaScript [12].

The honest caveat is that these tests probe one phase of a multi-phase pipeline. The same researchers note that schema "could very well be used in earlier phases," such as index construction or model training, where no outside tester can observe it [12]. The claim the evidence supports is narrower than "schema is useless": LLM-based engines do not document consuming it, and they demonstrably do not read it at answer time.

The largest controlled study found no meaningful citation effect from adding schema. In May 2026, Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026, matched them against 4,000 control pages with similar prior citation levels, and measured citations 30 days before and after [10]. Google AI Overviews moved by negative 4.6%, Google AI Mode by positive 2.4%, and ChatGPT by positive 2.2%, with the two positive figures small enough that the authors classify them as noise [10]. The same study also found that AI-cited pages were almost three times more likely to carry JSON-LD than non-cited pages, a gap the authors attribute to overall site quality rather than to the markup: "Schema could be doing real work, but it could also just be riding the wave of every other signal" [10].

A smaller self-experiment points the same direction with one exception. OtterlyAI added five schema types to its own site in December 2025 and tracked brand visibility across 319 prompts for three months [11]. Google surfaces rose, with AI Mode up 42% and a large AI Overviews jump the researchers partly attributed to unrelated algorithm shifts, while ChatGPT, Copilot, and Gemini visibility fell and Perplexity did not move [11]. Their summary: "Schema was invented for algorithms that couldn't handle unstructured text. AI Search is built to do exactly that" [11]. Read together, the correlation data flatters schema because well-run sites tend to implement it, and controlled tests keep failing to find a citation lift outside Google's search-backed surfaces.

Organization markup identifies the real-world entity behind a site, with properties for the legal name, logo, and sameAs links to official profiles. Google supports it as a structured data feature and uses markup like it to gather information about entities across the web [1][2], which matters because answer engines assemble claims about brands from many sources and benefit from an unambiguous statement of who you are. Implement one Organization object sitewide, reference it from the publisher field of your Article markup, and keep every value identical to your visible about and contact pages [1].

Article markup declares a page's headline, author, publication date, and modification date, and it remains a supported Google rich result feature [2]. Its relevance goes beyond the rich result: Bing's Fabrice Canel notes that generative AI systems value fresh content "partly as a reference check of their LLM training data" [7], and a machine-readable dateModified is the cleanest way to state freshness. Use real dates rather than scripts that bump the date on every deploy, and keep the visible byline and date matching the markup [1].

Product and Review markup carries prices, availability, and ratings, and it is the category with the most concrete platform support. Google lists Product and Review snippet features in its supported gallery [2], and Microsoft names product and review among the four labels that turn "plain text into structured data that machines can interpret with confidence" for AI answers [6]. Commerce is also where structured data has documented dependency: shopping surfaces need exact prices and availability, which prose cannot deliver reliably. Keep on-page markup consistent with any merchant feeds you submit, and only mark up reviews that actually appear on the page [1].

FAQPage markup lost most of its search payoff in August 2023, when Google limited FAQ rich results to "well-known, authoritative government and health websites" [4]. For a typical business site the markup now produces no visible Google feature, so it belongs lower on the priority list than the marketing around it suggests. It has not become worthless: FAQ is one of the content labels Microsoft names in its guidance for AI answer inclusion [6]. The larger AEO value is the visible question-and-answer structure itself, because assistants that fetch your page read the rendered text, not the markup [12]. Write the Q&A for readers first and treat the FAQPage block as an inexpensive add-on.

Breadcrumb markup describes where a page sits in your site hierarchy, and it stays on the list because it is supported in Google's feature gallery and costs almost nothing [2]. Most modern CMS platforms generate it automatically, along with basic WebSite and WebPage objects. The benefit is contextual: hierarchy markup helps a crawler understand that a pricing page belongs to a product section, which supports accurate snippets in the search layer Google's AI features draw from [2][3]. Verify your CMS output in a validator once, then leave it alone.

Implementation is a half-day project for most sites if you follow the constraints the platforms actually publish.

1. Use JSON-LD. Google recommends it over Microdata and RDFa because it lives in a script tag instead of being woven through your visible HTML [1].
2. Describe only what a reader can see. Google's guidelines prohibit marking up information users cannot see on the page, even when it is accurate [1].
3. Validate before shipping. Google's Rich Results Test previews which supported features your markup is eligible for [2].
4. Keep decisive facts in server-rendered visible HTML. Live-fetch assistants read rendered text only, and in testing just one of them executed JavaScript [12]. A price or definition that exists only in markup is invisible to most assistants.
5. Let the crawlers in. Allow OAI-SearchBot in robots.txt for ChatGPT search visibility [8] and PerplexityBot for Perplexity [9], since access is the gating requirement both platforms document.
6. Push updates through IndexNow, which Bing recommends for signaling the fresh content its engineers say generative AI weighs heavily [7].

The markup itself serves the search-index engines that document reading it, while the visible-content and access rules serve the LLM engines that do not.

The main trade-off is opportunity cost. Schema implementation competes for the same developer and content hours as producing answerable content, and the only controlled tests available found no citation lift from markup outside Google's surfaces [10][11]. A team that spends a sprint on exotic schema types instead of publishing well-structured answers has probably optimized the wrong layer.

Markup drift is the quiet operational risk. Schema that stops matching the visible page as content changes violates Google's guidelines [1], and stale prices or dates in markup hand wrong facts to the one platform family that definitely reads them.

Measurement is unstable enough that single-site results deserve suspicion. OtterlyAI attributed part of its own AI Overviews jump to unrelated algorithm shifts [11], and Ahrefs found a small, statistically significant AI Overviews decline among schema-adding pages that it does not present as causal [10]. Before-and-after citation numbers on one site rarely isolate the markup.

The evidence window is also early. Current tests can only observe the live-fetch phase of AI retrieval, and schema could be consumed invisibly during indexing or training [12]. Microsoft's public commitment suggests at least one major AI pipeline intends to rely on it more over time [6], so the cheap position is to maintain accurate core markup and re-evaluate as testing matures.

Not a requirement for AI visibility

Google states directly that no special markup, machine-readable file, or schema.org addition is needed to appear in AI Overviews or AI Mode [3]. A vendor pitch framing schema as the price of entry to AI search contradicts the platform's own documentation.

Not something chatbots read at answer time

When ChatGPT, Claude, or Perplexity fetches your page during a conversation, testing shows it processes the visible text and ignores JSON-LD, Microdata, and RDFa [12]. Facts you want quoted need to exist in readable prose.

Not a proven citation lever

The correlation between schema and AI citations is real, but the controlled test that isolated markup found effects indistinguishable from noise [10]. Cited pages tend to have schema because well-maintained sites have schema.

Not invisible metadata you can stuff

Google's policies require markup to describe content users can actually see on the page [1], so schema cannot serve as a side channel for claims your page does not make.

Not obsolete

Google still builds roughly 30 search features from it [2], Microsoft actively recommends it for AI answer inclusion [6], and adoption spans tens of millions of domains [5]. The case for schema has weakened as an AI citation tactic, not as web infrastructure.