Schema markup AI Overviews actually use

Quick spoiler before the long read: schema doesn’t rank anything, it identifies you - so wire Organization, Article/author and Product properly, connect them into one graph, and stop expecting FAQPage to earn a rich result it lost in 2023.

I read a lot of JSON-LD, and the same pattern turns up more often than it should: FAQPage on every page, HowTo where no procedure exists, and no Organization anywhere! Every time I ask the same thing: who is this markup for? Not the reader, who never sees it. Not Google’s rich results, which stopped rewarding most of it in 2023. For a ranking boost, then? One that, as far as I can tell, never existed.

So: schema markup doesn’t rank you - it tells a machine who you are, what a page is about, and who to attribute a claim to. In 2026 the useful question isn’t “which schema gets me a rich result” but “which Schema.org types do AI engines actually read.” Some still earn visible SERP features, some are pure entity signals, and a couple that people still obsess over - FAQPage and HowTo - quietly lost their Google rewards in 2023. Here’s what to keep, what to drop, and how to ship JSON-LD that validates and gets read.

Schema is an entity signal, not a ranking trick

The oldest myth in structured data is that adding schema “boosts rankings.” It never did, and it arguably matters less than ever - yet it still turns up in proposals. What schema actually does is disambiguate. When an answer engine reads Organization with a sameAs pointing to Wikidata, LinkedIn, and Crunchbase profiles, it can connect the page to a known entity instead of guessing. When it reads Article with a named author, it knows who to attribute a claim to - which is exactly the check a model runs before it cites you in an AI Overview.

So treat schema like a passport, not a megaphone. It doesn’t make a site louder; it makes it identifiable and quotable. That reframing changes which types are worth the time - and, honestly, retires probably half of the schema to-do lists I’ve seen.

Which types AI engines read vs ignore (2026)

Not all markup is equal. Some types drive visible features, some feed the entity graph silently, and some are effectively decorative now. My read of the field, type by type:

Schema type	What AI engines do with it	Worth it in 2026
Organization + sameAs	Resolve you to a known entity (Knowledge Graph, Wikidata)	Yes - foundational
Article + author	Attribute claims to a named, verifiable author	Yes - the E-E-A-T anchor
Product + offers/review	Read price, availability, ratings for shopping answers	Yes - commerce
BreadcrumbList	Understand site hierarchy and a page’s place in it	Yes - cheap, structural
FAQPage	Parse Q&A pairs (no Google rich result since 2023)	Partial - parsing only
HowTo	Parse ordered steps (no Google rich result since 2023)	Partial - parsing only
WebSite + SearchAction	Enable the sitelinks search box	Minor

The pattern: the durable value is identity and attribution (Organization, author), then commerce facts (Product), then structure (BreadcrumbList). FAQPage and HowTo are still legal, still parsed, but no longer a SERP win - so mark them up if the content is genuinely a Q&A or a procedure, not as a growth hack. Is a pricing page a procedure? It is not.

JSON-LD: placement and shape

Google recommends JSON-LD over Microdata and RDFa, and so does everyone else. The mechanics are simpler than the folklore suggests (and the folklore is considerable):

• One block, either location. A single <script type="application/ld+json"> works in <head> or <body> - Google renders and reads both. It doesn’t need to be above the fold, whatever the checklist said.
• One connected graph per page. Use a top-level @graph array and reference entities by @id so the Article, its author, and the Organization link to each other instead of floating as orphans.
• Match the visible content. Every value in the JSON must correspond to something a user can see on the page. A price in schema that isn’t on the page is the fastest way to get markup ignored or flagged.

The mistakes that void your markup

Validate before you ship

Two free checks, both worth running on every template change (yes, every one!):

1. Schema.org’s Schema Markup Validator (validator.schema.org) - checks your markup against the vocabulary itself, independent of any search engine’s feature rules.
2. Google’s Rich Results Test - tells you whether Google can parse the page and which, if any, rich features it’s eligible for.

They answer different questions: the first is “is this valid Schema.org,” the second is “will Google do anything with it.” A page can pass the first and show no eligible features in the second - which, post-2023, is exactly the normal state for FAQPage and HowTo. Does that mean the markup is broken? No - it means the reward is gone and the signal isn’t. To automate the whole pass, our schema-fix tool audits every page’s JSON-LD, flags mismatches with visible content, and patches the graph so Article, author, and Organization actually connect.

Checklist

• Organization on the site, with a real sameAs array (Wikidata, LinkedIn, socials).
• Article (or BlogPosting) with a named author as a Person node.
• Product carries offers + genuine review/aggregateRating where applicable.
• BreadcrumbList on every deep page.
• One @graph per page; entities linked by @id, no orphans.
• Every schema value matches visible on-page content.
• Passes both validator.schema.org and Google’s Rich Results Test.
• FAQPage/HowTo only where the content is genuinely a Q&A or a procedure.

Visible Google rich result 0%

Still parsed as entity/content data 100%

source: internal audits, 2026

The chart is the point: the same FAQPage markup that earns zero visible SERP features is still fully parseable content for the crawlers and AI engines that read structured data. Deprecation of a feature is not deletion of the signal - a distinction that seems to get lost every time the word “deprecated” appears in a changelog.

In short

• Schema is an entity and attribution signal, not a ranking factor - build for identity, not for a boost.
• The durable types are Organization+sameAs, Article+author, and Product+offers - in that order.
• FAQPage and HowTo lost their Google rich results in 2023 but are still parsed content.
• JSON-LD goes in one block, head or body, as one connected @graph; every value must match the page.
• Validate with both Schema.org’s validator and Google’s Rich Results Test before every deploy.

Schema won’t make a page louder. It just makes sure that when the engine asks “who said this?”, the answer isn’t a shrug.

Sources

• Schema.org - the vocabulary itself and its getting-started guide; JSON-LD is the recommended syntax.
• Google Search Central, “Structured data markup that Google Search supports” - the authoritative list of which types produce features.
• Google Search Central Blog, “Changes to FAQ and HowTo rich results” (August 2023) - the deprecation of FAQPage and HowTo rich results for most sites.
• Google Search Central, “Article (Article, NewsArticle, BlogPosting) structured data” - author, datePublished, and attribution fields.