Schema Markup That Wins AI Citations in 2026

Schema markup spent most of 2025 in awkward limbo. Authors kept pasting FAQPage blocks into every post even though Google had quietly capped FAQ rich results to government and health sites back in August 2023. Then May 7, 2026 made the situation honest. Google fully deprecated FAQ rich results, and the entire reason most people were shipping JSON-LD evaporated overnight. The good news is that schema in 2026 is more useful than ever for a different reason. AI Overviews, Perplexity, and ChatGPT lean on structured data to figure out who you are, what your page actually claims, and whether you are worth quoting.

Quick Answer: Schema markup in 2026 is no longer about FAQ rich results, which Google deprecated on May 7, 2026. The schema types that still win SERP features are Article, HowTo, Product, Review, Recipe, BreadcrumbList, and Organization. The schema types that win AI citations from Google AI Overviews and ChatGPT are Article (especially with author and citation properties), Organization, Person, HowTo, and Dataset. JSON-LD remains the recommended format. The biggest 2026 shift is treating schema as an entity-disambiguation tool for AI engines, not a SERP feature lottery ticket.

What Changed in May 2026 With the FAQ Rich Result Deprecation

On May 7, 2026 the FAQ rich result officially stopped appearing in Google Search across every site, including the small list of government and health domains that had retained it since August 2023. June 2026 dropped FAQ from the Rich Results Test interface, and August 2026 removes the Search Console API surface entirely. That sequence matters because plenty of agencies are still selling FAQPage schema as a hot tactic. It is not. It has not been for any commercial site since 2023 and now it is officially gone for everyone. Google has been clear that FAQPage markup itself is not harmful and the engine still parses it for understanding, but it will not produce a visible search feature.

The shift forces a healthier mental model. Schema is not a coupon you redeem for a rich result. It is a machine-readable assertion about what your page is and who said it. SERP features are one downstream benefit. AI Overview citations, Knowledge Graph entries, and entity disambiguation in LLM training corpora are equally real downstream benefits, and they now matter more.

If you have FAQ blocks live today, here is the practical move:

• Leave them in place if the questions and answers genuinely match user intent. They still help Google understand the page.
• Remove them if they were stuffed in for the rich result and the answers feel forced. Forced FAQ blocks can hurt readability without any upside.
• Stop adding FAQ blocks to net-new posts unless the format is the right format for the content itself.

Schema Types That Still Win SERP Features in 2026

Despite the FAQ deprecation, Google still surfaces six schema-driven rich results consistently. These are the ones worth implementing first because they create visible CTR lift, not just understanding gains.

Article remains the workhorse. It triggers the headline-card and image treatment for news and blog content, and it feeds the publishedTime and author signals that AI Overviews now use to weigh citation candidates. HowTo continues to render step previews for procedural content. Product earns price, availability, and review-stars displays for any sellable item. Review and AggregateRating render the star ratings in services-and-software listings. Recipe still gets the cooking-time and rating chip. BreadcrumbList replaces the URL in the result with a friendlier hierarchy.

Here is the priority order for a typical content site:

1. BreadcrumbList on every page. Universal, zero-risk, improves the visible URL.
2. Article on every blog post. Carries author and dates, which AI engines use.
3. Organization on the homepage. Defines who you are once for the whole site.
4. Person for each author. Connects bylines to verifiable people.
5. HowTo or Recipe on the small subset of pages where the format matches.
6. Product or Review only when you actually sell or rate a thing.

Resist the temptation to add Product schema to non-commercial content. Google has gotten strict about this since 2024 and the page can be flagged for misleading structured data. The same applies to LocalBusiness on a generic blog. Schema should describe what the page actually is.

Schema Types That Win AI Citations Are a Different List

The schema that earns citations from Google AI Overviews, Perplexity, ChatGPT, and Claude is not always the schema that earns SERP features. AI engines are reading your JSON-LD to answer two different questions. Who wrote this and how trustworthy is the source. What concrete entities, claims, and procedures are on the page.

For the first question, Article with full author, datePublished, dateModified, publisher, and citation properties is dominant. Organization with sameAs links to your verified social profiles tells the engine that your site is connected to a real entity. Person schema for each byline, again with sameAs to LinkedIn, GitHub, or a verified academic profile, turns an author name into a node in the LLM's entity graph.

For the second question, HowTo with explicit step objects, Dataset with documented variables, and SoftwareApplication with version and feature lists are the formats AI engines parse most cleanly. The pattern is that LLMs prefer schema where the content of the page is also represented as structured data, so the model can quote you with high confidence that the claim is exactly what you said.

A useful internal exercise for any post is to ask, if Google AI Overviews wanted to quote one specific sentence from this page, which schema property would make that quote unambiguous. If the answer is none, you have a schema gap.

For the broader playbook on getting cited by AI engines, the Answer Engine Optimization guide covers the formatting layer that pairs with these schema types.

Article Schema Patterns Google Parses Cleanly

Article is the schema type to get right because it appears on the majority of pages most sites publish, and the AI Overview citation weight runs through it. The full pattern below is the version that Google's documentation parses without warnings and that AI engines extract reliably.

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "Schema Markup That Wins AI Citations in 2026",
  "description": "Which schema types still drive results post-FAQ deprecation.",
  "image": [
    "https://example.com/hero.png"
  ],
  "datePublished": "2026-06-19T12:00:00+00:00",
  "dateModified": "2026-06-19T12:00:00+00:00",
  "author": {
    "@type": "Person",
    "name": "Kevin Gabeci",
    "url": "https://www.kevingabeci.com/",
    "sameAs": [
      "https://www.linkedin.com/in/kevin-gabeci/",
      "https://github.com/TheDovakiin"
    ]
  },
  "publisher": {
    "@type": "Organization",
    "name": "Astro SEO Blog",
    "url": "https://astroseoblog.com/",
    "logo": {
      "@type": "ImageObject",
      "url": "https://astroseoblog.com/logo.png"
    }
  },
  "mainEntityOfPage": {
    "@type": "WebPage",
    "@id": "https://astroseoblog.com/blog/schema-markup-that-wins-ai-citations-2026"
  }
}

The fields that often get cut for brevity are the fields that AI Overviews need most. Keep the full author object with sameAs links. Keep both datePublished and dateModified. Keep the image as an array even with one image, because the array form survives template changes better.

HowTo Schema for Procedural Content

HowTo still earns visible step previews for the right kind of content, and AI engines treat it as a strong signal for extraction. The rule is that the page must actually be a procedure. A how-to article with five real steps in the body deserves HowTo schema. A listicle of unrelated tips dressed up as steps does not, and Google has been demoting that pattern since the 2024 helpful content update.

A clean HowTo block looks like:

{
  "@context": "https://schema.org",
  "@type": "HowTo",
  "name": "How to Add Article Schema to a Blog Post",
  "totalTime": "PT10M",
  "step": [
    {
      "@type": "HowToStep",
      "name": "Generate the JSON-LD",
      "text": "Use a generator or write the block by hand with the Article type."
    },
    {
      "@type": "HowToStep",
      "name": "Add to the head",
      "text": "Place the script tag with type application/ld+json inside the head element."
    },
    {
      "@type": "HowToStep",
      "name": "Validate",
      "text": "Run the Rich Results Test and the Schema.org Validator."
    }
  ]
}

If your platform makes generating this by hand tedious, the best schema markup generators roundup covers tools that produce valid blocks from a form.

Product Schema for Non-Ecommerce Use Cases

Product schema is one of the more flexible types, and there are legitimate non-ecommerce uses that pay off. SaaS apps qualify as Product if they have a price, availability, and brand. Free tools can use SoftwareApplication, which is a subtype of Product. Digital downloads with prices qualify. The thing to avoid is wrapping editorial reviews of products you do not sell, because that is what Review schema is for.

The minimum valid Product block for a SaaS landing page is brand, name, description, offers, and an aggregateRating if you have real reviews. Resist inventing review counts. AI engines compare the aggregateRating count to mentions across the web, and a 4.9 with 2,000 reviews on a tool nobody talks about is a credibility ding rather than a credibility win.

Organization and Person Schema for the E-E-A-T Signal

Since the March 2026 core update, Google has made E-E-A-T enforcement more concrete, and the Experience signal is where most demotions happened. Schema gives you a way to back the byline up with verifiable identity. The pattern is one Organization block on the homepage describing the brand, and one Person block per author that connects to that Organization through the publisher relationship.

The Organization block at minimum needs name, url, logo, and sameAs links to verified social profiles. The Person block needs name, url, jobTitle, worksFor pointing at the Organization, and sameAs links to professional profiles where the person's real work history is documented. LinkedIn, GitHub for engineers, ORCID for academics, and a verified professional site all qualify. This is the structured-data layer that pairs with the human-readable bio and credentials work that the E-E-A-T 2026 guide covers in detail.

JSON-LD Copy-Paste Templates for Each Type

For practical use, here are the minimal valid blocks for the seven types worth implementing. Paste these into your template, swap the placeholder values, and validate.

{
  "@context": "https://schema.org",
  "@type": "BreadcrumbList",
  "itemListElement": [
    {"@type": "ListItem", "position": 1, "name": "Home", "item": "https://example.com/"},
    {"@type": "ListItem", "position": 2, "name": "Blog", "item": "https://example.com/blog/"},
    {"@type": "ListItem", "position": 3, "name": "Schema Guide"}
  ]
}

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Your Brand",
  "url": "https://example.com/",
  "logo": "https://example.com/logo.png",
  "sameAs": [
    "https://twitter.com/yourbrand",
    "https://linkedin.com/company/yourbrand"
  ]
}

{
  "@context": "https://schema.org",
  "@type": "Person",
  "name": "Your Name",
  "url": "https://example.com/about",
  "jobTitle": "Editor",
  "worksFor": {"@type": "Organization", "name": "Your Brand"},
  "sameAs": ["https://linkedin.com/in/yourname"]
}

{
  "@context": "https://schema.org",
  "@type": "SoftwareApplication",
  "name": "Your Tool",
  "applicationCategory": "BusinessApplication",
  "operatingSystem": "Web",
  "offers": {"@type": "Offer", "price": "29.00", "priceCurrency": "USD"}
}

Keep each block in its own script tag for clarity, or combine them in a single graph using @graph if your template makes that easier. Both forms are valid.

Validating With Rich Results Test Versus Schema.org Validator

The two validators answer different questions and both belong in your pre-publish checklist. The Rich Results Test, available at search.google.com/test/rich-results, tells you whether Google will treat the block as eligible for a search feature. The Schema.org Validator at validator.schema.org tells you whether the block is syntactically valid against the schema.org vocabulary regardless of what Google does with it.

The Rich Results Test will quietly ignore types that no longer trigger features, which is now the case for FAQPage. The Schema.org Validator will tell you that the FAQPage block is still valid markup even though Google will not surface it. Use both. Run the Rich Results Test to confirm your Article and HowTo will get the visible treatment. Run the Schema.org Validator to confirm that your Organization, Person, and other entity-disambiguation blocks are clean even if they do not produce a feature directly.

A common production mistake is shipping schema that validates in isolation but conflicts with another block on the page. The Rich Results Test catches this. If you have two Article blocks pointing to different URLs on the same page, the second one wins or both get ignored, depending on the conflict. One canonical block per type per page.

FAQ

Should I remove FAQ schema from my old posts after the May 2026 deprecation? Only if the FAQ blocks were added purely for the rich result and the questions feel forced. If the questions are real and the answers match user intent, leave them in place. Google still uses FAQPage markup to understand the page even without the visible rich result, and removing valid markup creates work for no upside.

Does AI Overviews actually read my JSON-LD or does it just use the rendered text? Both. The page text is the primary input but the JSON-LD is what disambiguates author, publication date, and entity. Articles with full Article schema get cited at a measurably higher rate in AI Overviews than identical pages without it, which is why Article schema is now table stakes.

What about FAQSection or the new community-proposed types? FAQSection is a community proposal aiming to be an unambiguous replacement for FAQPage that signals an in-page FAQ section without the now-deprecated rich result expectation. Google has not formally adopted it. For now, stick with Article or WebPage as the parent type and let the FAQ live as content inside.

Is microdata or RDFa still acceptable or do I have to use JSON-LD? JSON-LD is what Google recommends and what their parsers handle most reliably. Microdata and RDFa still validate but they are markup-inline approaches that tangle with template changes. New schema should be JSON-LD.

How many schema types is too many on one page? There is no hard cap, but realistic pages need one or two. Article plus Breadcrumb is the common pair for blog posts. Adding Organization and Person from the homepage and author template is fine because those reference the homepage entities. Stacking five types on a single article is usually a sign that something does not belong.

Does schema help with ChatGPT and Perplexity citations or just Google? It helps with both. ChatGPT and Perplexity rely on retrieval indexes that surface structured data alongside page content, and entity disambiguation through Organization and Person schema is one of the strongest signals for being chosen as a cited source over a competitor without it.

Do I need separate schema for translated pages? Yes. Each language version is its own URL and needs its own JSON-LD with the correct inLanguage value. The Article should also reference the canonical English version through translationOfWork if you want to be explicit about provenance.

Sources and Further Reading

Astro SEO Blog has covered the AI search transition extensively, including how to optimize for Google AI Overviews in 2026 and the broader topical authority playbook that schema work feeds into. For the foundational concept, the what is schema markup primer covers the basics.

External references worth bookmarking:

• Google's structured data documentation for the canonical guidance.
• Schema.org for the full vocabulary.
• Google Search Central's FAQ deprecation post for the May 2026 timeline.