Which schema types earn AI citations? Data from 5,000 sites and the Ahrefs study reveal what works.
Schema markup for AEO determines whether AI engines trust your content enough to cite it. Most B2B teams treat structured data as a checkbox — add some JSON-LD, move on. A 5,000-site audit from Digital Applied (April 2026) found that 71% of sites deploy at least one schema type, but only 22% pass Google's Rich Results Test across every @type they emit. That 49-point validation gap correlates directly with AI citation rates at +0.34 Pearson.
Below: which schema types earn AI citations, how to implement them, and where most B2B teams waste effort. If you've already read our AEO strategy guide, think of this as the technical implementation layer underneath that framework.
Schema correlates with AI citations. The causal link is weaker than most guides claim.
Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026, matching them against 4,000 control pages. Google AI Overviews showed a −4.6% change, AI Mode showed +2.4%, and ChatGPT showed +2.2%. None of the positive numbers cleared statistical significance (Ahrefs, March 2026).
Every page in that study already had 100+ AI Overview citations before the test began. Ahrefs tested pages that AI systems already crawled, surfaced, and cited regularly.
Contrast that with Stackmatix's analysis: 65% of pages cited by Google AI Mode and 71% of pages cited by ChatGPT include structured data (Stackmatix, 2026). Digital Applied measured +0.34 Pearson correlation between Rich Results Test pass rates and AI citation frequency across 5,000 sites (Digital Applied, April 2026).
Schema doesn't flip a citation switch. It reduces ambiguity in how AI systems parse your content. For pages not yet in the consideration set, clean structured data can be the difference between AI retrieval and silence.
Five schema types cover 80% of B2B use cases: Organization, Article (or BlogPosting), FAQPage, BreadcrumbList, and HowTo.
Digital Applied's audit found that sites shipping Organization + WebSite + BreadcrumbList as a universal baseline outperformed sites deploying more schemas in random combinations (Digital Applied, April 2026). Three correctly combined schemas beat five weakly validated ones in every segment tested.
Build in layers. Your entity layer goes on every page: Organization tells AI engines who you are, WebSite and BreadcrumbList provide navigation context. Then add content-type schemas where they apply — Article for blog posts, FAQPage for Q&A content, HowTo for step-by-step tutorials. This layered approach gives AI retrieval systems a clean, parseable picture of your site.
FAQPage schema had a 67% citation rate in AI responses for relevant queries during Frase's 2026 testing. Pages with QAPage schema get cited by ChatGPT 58% more often than pages using basic Article schema alone (Stackmatix, 2026).
Three rules separate cited FAQPage implementations from ignored ones.
First, match your questions to AI query patterns. Pull questions from ChatGPT, Perplexity, and Google's "People Also Ask" for your topic. If your FAQ questions don't match how people phrase queries to AI assistants, the schema won't trigger retrieval.
Second, keep each answer between 40 and 75 words. This range matches the passage length AI engines prefer to extract — our chunk-first framework guide covers why passage length matters for citation rates.
Third, nest your FAQPage inside an Article schema instead of deploying it standalone. A compound schema gives AI systems both the content type and the specific Q&A pairs in one structured signal. Standalone FAQPage schema without an Article wrapper lacks the authorship and publication context that builds retrieval trust.
HowTo schema maps directly to how ChatGPT and Perplexity answer procedural queries. When someone asks an AI assistant "how do I" anything, the retrieval system looks for content with explicit step structures. HowTo schema makes those steps machine-readable.
Keep each step under 50 words. Name every step with an action verb. Include time estimates and tool requirements where they apply — AI engines surface these details in their responses.
Google deprecated the HowTo rich result in September 2023, which led many teams to strip HowTo schema from their pages. That was a mistake. The rich result display is gone, but AI retrieval systems — including Google's own AI Mode and AI Overviews — still read structured data during their extraction process. Removing HowTo schema eliminated a signal that helped AI engines parse your procedural content.
For B2B content, HowTo works best on implementation guides, onboarding docs, and troubleshooting pages. Pair it with Article schema to provide the authorship and publication context that AI systems use for trust evaluation.
Speakable schema marks which sections of your page are best suited for text-to-speech playback and AI extraction. Google introduced it for news content, but in 2026 it functions as a direct pointer for AI retrieval: "this is the passage worth citing."
Point the Speakable cssSelector at your lead answers and key takeaways — the 40–75 word passages you've already optimized for AI extraction. Bing Copilot uses Speakable for voice responses (Stackmatix, 2026). Google Assistant reads Speakable-tagged sections when answering voice queries.
Implement Speakable as a property within your Article schema, not as a separate JSON-LD block. Reference the CSS classes wrapping your lead answer paragraphs. This tells AI systems exactly which passage on a 2,000-word page to extract, rather than forcing them to guess.
Speakable remains in beta on Google, but early implementation gives your pages an extraction advantage as AI voice interactions grow. B2B teams publishing research, benchmarks, or tactical guides benefit most — these formats produce the citation-ready passages that Speakable is designed to highlight.
Digital Applied's 5,000-site audit answered this question with hard numbers: the +0.34 Pearson correlation between Rich Results Test pass rates and AI citation frequency is stronger than most on-page SEO signals (Digital Applied, April 2026).
Across the 78% of sites in the "deploy-but-broken" segment, five error patterns account for over 90% of validation failures. Missing required properties, incorrect @type nesting, invalid date formats, broken URL references, and mismatched entity IDs. Every one of these errors is catchable with automated testing before deployment.
Invalid schema is worse than no schema. Conflicting structured data signals force AI retrieval systems to resolve contradictions, and resolution usually means skipping the page. Run every page through Google's Rich Results Test and the Schema.org Validator before deployment.
Then set up monitoring. Schema breaks silently when CMS templates change, plugins update, or developers modify page structures. Wire a JSON-LD validator into your CI/CD pipeline so broken schema never reaches production. Google Search Console's Enhancements report catches errors at scale — check it weekly.
Track whether your validation fixes translate into AI visibility changes with Nobori.
The highest-cited B2B pages use three to four schema types together: Article for the page itself, BreadcrumbList for navigation context, FAQPage for Q&A sections, and Speakable to highlight extractable passages.
Nest related schemas rather than deploying separate JSON-LD blocks. Embed FAQPage within your Article schema's mainEntity property. Add Speakable as a property of Article. This compound signal tells AI engines both the content type and the specific passages worth extracting — in one parseable structure.
Deploy Organization schema separately at the site level, present on every page but not nested inside individual article schemas. Organization provides the entity layer. Article provides the content layer. Mixing them inside the same JSON-LD block confuses validators and AI parsers.
The practical structure for a B2B blog post: one JSON-LD block with Organization + WebSite (site-wide template), one block with Article + nested FAQPage + Speakable (page-specific). BreadcrumbList goes in its own block because it describes navigation, not content. This separation keeps each schema type clean and each signal unambiguous.
Five errors account for most schema failures in the Digital Applied audit. B2B teams hit these repeatedly.
Deploying schema without validation. Only 22% of sites pass the Rich Results Test cleanly. Most teams add JSON-LD once and never check it again — then CMS updates or template changes break the markup silently.
Using schema generators without review. Automated tools produce technically valid JSON-LD that often contains generic placeholder data. AI engines can detect when structured data doesn't match the actual page content, and the mismatch hurts trust signals.
Removing HowTo schema after Google deprecated the rich result. The AI retrieval pipeline still reads HowTo structured data. Removing it eliminated a parsing signal for no benefit.
Deploying too many schema types without proper nesting. Five loosely connected JSON-LD blocks create ambiguity. Three properly nested schemas create clarity. Quality beats quantity.
Ignoring Organization schema. Without it, AI engines have no structured way to connect your content to your brand entity. Every page needs it.
Schema markup optimization is technical work. Knowing whether it paid off requires tracking your visibility across every AI engine that matters.
Nobori monitors your brand's presence across ChatGPT, Google AI Overviews, Perplexity, Gemini, and Claude — updated daily. After you implement schema changes, Nobori tracks whether AI engines start citing your pages more frequently, which competitors get cited instead, and which queries trigger or miss your content.
The platform generates actionable optimization tasks based on your visibility data. Rather than guessing which pages need schema work, you get specific recommendations tied to real citation gaps across all five AI platforms.
See if AI engines are citing you → nobori.ai
