Last updated August 2026
What the correlation data says
AirOps analyzed more than 12,000 URLs across 900 ChatGPT queries in 15 industries. One finding stands out.
According to that study, 61% of pages cited by ChatGPT carried three or more distinct schema types. Only 25% of Google’s top-ranking pages for the same queries did.
That gap is significant. It means the typical ChatGPT-cited page looks structurally different from the typical top-10 Google result: it is more likely to carry layered, overlapping schema markup that signals content type, entity relationships, and answer intent in combination.
The contradiction you need to know
Ahrefs ran the controlled experiment AirOps did not.
In a 2026 study, Ahrefs tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026, matched against 4,000 control pages. The finding: adding schema produced no meaningful uplift in AI citations across Google AI Overviews, Google AI Mode, or ChatGPT. Google AI Overviews actually showed a small decline (-4.6%). Google AI Mode came in at +2.4% and ChatGPT at +2.2%, both statistically indistinguishable from zero.
The Ahrefs study had one notable constraint: every page in the dataset already had 100 or more AI Overview citations before the experiment started. These were not invisible pages. They were already in the citation pool.
So the evidence, taken together, says something more nuanced than “schema helps” or “schema does not matter”:
- Pages that AI engines cite tend to carry richer, more layered schema than pages that merely rank on Google.
- Adding schema to pages that are already being cited does not appear to push citation rates higher.
- What schema does for pages with zero AI citations is not yet settled by controlled research.
Hold both findings. Neither one is wrong.
The JSON-LD types that appear most on cited pages
The AirOps data and the broader pattern of which schema types AI engines parse most actively point to a consistent shortlist. These are the types you will find most often on pages that earn AI citations.
| Schema type | What it signals | Where to add it | Priority for AEO |
|---|---|---|---|
| FAQPage | A page contains multiple question-and-answer pairs | Topic guides, support pages, FAQ blocks | Highest |
| HowTo | A page contains a step-by-step process | Tutorials, setup guides, process explainers | Highest |
| Article | A page is a distinct editorial piece with an author and publish date | Blog posts, editorial content, deep-dive pieces | High |
| QAPage | A page is built around a single question with a primary answer | Single-question explainers, Q&A pages | High |
| Organization | The page belongs to a named entity with a defined identity | Homepage, About page | High |
| Product | A page describes a specific purchasable item | Product detail pages, SaaS pricing pages | Medium |
| AggregateRating / Review | A page carries verified ratings or structured reviews | Review pages, comparison pages | Medium |
| BreadcrumbList | The page has a defined navigational path | All pages with multi-level navigation | Supporting |
| DefinedTerm | The page defines a specific term or concept | Glossary entries, definition pages | Supporting |
Why FAQPage leads: AI engines consume question-and-answer pairs directly. FAQPage schema maps exactly onto how ChatGPT, Google AI Overviews, and Perplexity retrieve structured responses. When a page carries FAQPage markup with clear questions and complete answers, the engine can parse it as a list of discrete answer candidates rather than having to extract those answers from flowing prose.
Why layering matters: The AirOps finding is about combinations, not single types. A page with FAQPage schema plus Article schema plus Organization schema is signaling three things at once: this content answers specific questions, it is editorial and attributed, and it comes from a named entity. That combination is more machine-readable than any single type alone.
A brief note on the arXiv preprint
A 2025 arXiv preprint by Kumar and Palkhouski (arXiv:2509.10762) audited 1,100 URLs across 1,702 AI citations from Brave Search, Google AI Overviews, and Perplexity. It ranked structured data as the third-strongest predictor of AI citation likelihood, associated with a +39% lift. The two factors ranked higher were metadata and freshness (+47%) and semantic HTML (+42%).
This preprint has not been peer-reviewed in the same way the Ahrefs study has been scrutinized. It also uses a cross-sectional correlation method, not a controlled experiment. The direction of the finding aligns with the AirOps correlation data. The magnitude is not supported by the Ahrefs controlled study.
Take it as a signal worth watching, not a settled conclusion.
How to think about schema in your AEO workflow
Schema is not a magic lever. It is a communication layer between your content and the machines that parse it. Here is how to position it correctly.
Schema does not replace content quality. A page with five schema types and thin prose is not going to earn AI citations. Schema helps machines understand what good content is saying. It does not make weak content stronger.
Schema works alongside structural formatting. The same AirOps analysis found that 68.7% of ChatGPT-cited pages followed a sequential heading structure (H1 to H2 to H3), compared to just 23.9% of Google’s top-ranked pages. Schema and clean heading hierarchy reinforce each other. One tells the machine what type of content you have. The other tells it how the content is organized.
The implementation cost is low. JSON-LD sits in a script tag in your page head. It does not require you to restructure your HTML. Adding FAQPage and Article schema to a well-written page takes minutes. If the causal evidence catches up with the correlation data, you will already be set. If it does not, you will have lost almost nothing.
Validate before you publish. Schema that is syntactically broken does nothing. Run every JSON-LD block through Google’s Rich Results Test and Schema.org’s validator before deploying.
The implementation order
If you are prioritizing schema for AEO and need a sequence, this is where the correlation data and the structural logic converge:
- FAQPage on every page with two or more Q&A pairs. This is the single highest-signal type for AI answer-box content.
- HowTo on step-based guides and tutorials. Process content is heavily cited for instructional queries.
- Article on all editorial and blog content. An author, a publish date, and an organization linkage add entity context.
- Organization on your homepage and About page. Entity disambiguation matters across all AI engines that rely on knowledge graph signals.
- QAPage on single-question pages. Useful for pages built around one specific question.
- Product on product and pricing pages. Important for commercial and comparison queries.
- AggregateRating on comparison or review pages. According to AirOps’ separate research, comparison pages containing three or more HTML tables earn 25.7% more AI citations than those without. Schema on review-rich pages reinforces that signal.
- BreadcrumbList site-wide. Structural context helps AI crawlers understand where a page sits in your information architecture.
- DefinedTerm in glossary and definition content. Useful for queries that are explicitly definitional.
Tools that support schema implementation for AEO
Several tools cover part or all of the schema-to-citation workflow.
Surfer builds structured-data recommendations into its Content Editor alongside NLP content scoring. You optimize and score for AI answer boxes without leaving the editing environment. It is the strongest option if content creation is your primary workflow.
SE Ranking’s Site Audit flags missing and invalid schema types at the crawl level. It catches broken JSON-LD and gaps in implementation across your full site, useful when schema is already partially deployed and you need an audit.
Semrush’s On Page SEO Checker and Site Audit tools surface schema errors alongside broader on-page signals. Practical for teams already in the Semrush ecosystem.
Writesonic supports structured-data output in its AI content generation workflow. If you are producing AEO-optimized drafts at volume, it can generate JSON-LD alongside the content itself.
Temso generates FAQ schema as part of its AEO content-fix workflow. If you are tracking which answer boxes you are losing across eight AI engines and fixing them end to end, the schema generation is built into the fix cycle rather than treated as a separate implementation step. Entry price is $89/mo. Full tool profile: /tools/temso.
None of these tools guarantee citation uplift. What they do is remove implementation friction and reduce errors so that your schema, at minimum, communicates correctly.
What to read next
- The evidence debate on schema and AI citations in full: Schema Markup and AI Citations: What the Data Actually Says
- How to measure whether your schema work is moving citation rates: How to Measure AI Answer Citations
- Full AEO tool ranking: /rankings/aeo-tools
- Temso tool profile: /tools/temso
- Writesonic tool profile: /tools/writesonic
- AEO glossary: /glossary
- How we rank tools: /methodology
Schema markup is worth implementing at low cost and low risk. The honest picture of the evidence is that it correlates with AI citation, but one careful controlled study found no causal lift for already-cited pages. Start with FAQPage and HowTo, layer Article and Organization on top, and then measure whether your citation rates move. If you want to run that loop inside one tool, see the full AEO tool ranking for the options that cover monitoring, content fixes, and schema generation together.