The 2026 State of Structured Data

Structured data is not dying. It's changing jobs.

For most of its history, schema markup existed to earn a rich result, a star rating, an FAQ dropdown, a recipe card. That job is shrinking: Google has retired more rich-result types since 2023 than it's added, and the last of the FAQ eligibility disappeared on May 7, 2026. If you're measuring structured data purely by how many gold stars and expandable panels it earns you in Google's results, 2026 looks like a story of decline.

That's the wrong measurement. The same period that narrowed rich results is the period in which ChatGPT, Perplexity, and Google's own AI Mode became a primary way people get answers, and every one of those systems leans on structured data to understand, verify, and cite what's on a page. Schema's old job (decorate the search result) is shrinking. Its new job (tell AI systems exactly what your content and your business are) is growing faster than the old one is shrinking. This is the actual state of the field: less visual, more foundational.

This is a long piece, deliberately. It's meant to be the single reference you come back to when you need the full, current, sourced picture: where structured data came from, what's actually been retired versus what the panic headlines claimed, what the major AI systems are genuinely doing with schema, how the mainstream WordPress tools are responding, and what all of that means for what you should prioritize. We'll separate verified fact from unverifiable claim throughout, because this is exactly the topic where confident-sounding overclaims travel fastest.

Is structured data dead in 2026?

No, and the confusion about this is itself part of the story worth understanding.

In November 2025, Google published a routine deprecation notice retiring several structured data types starting January 2026. It read, to a fast skim, like Google backing away from schema generally. It triggered enough alarm that Google's John Mueller stepped in on Reddit to correct the record directly, making clear that markup types come and go individually while the core ones, the types that describe products, organizations, articles, and reviews, remain central to how Search understands a page. Retiring a lightly-used type like Practice Problem markup is not the same claim as retiring structured data as a concept, and treating them as equivalent was the actual error behind the panic.

The pattern holds up when you look at what's actually been removed since 2023 versus what's been added. Removals have targeted narrow, single-purpose features: Book Actions, Estimated Salary, Vehicle Listing, Claim Review, Special Announcement, Practice Problem, and, as of May 7, 2026, FAQPage, whose last remaining eligibility (for government and health sites) was withdrawn along with everything else. Additions and expansions, meanwhile, have clustered almost entirely around the types that describe fundamental entities: Product variants and organization-level shipping and return policies, an expanded Organization schema that unlocked more of the knowledge-panel experience, Discussion Forum and Profile Page support. Structured data isn't being deprioritized. It's being pruned of its more decorative uses while its identity-and-entity core gets reinforced.

One more fact worth holding onto, because it hasn't changed through any of this: Google has repeatedly and explicitly stated that structured data is not a ranking factor, and that losing eligibility for a rich result, whether through deprecation or through a markup violation, affects display, not rankings. Retired markup doesn't hurt you. It just stops doing the one thing it used to do.

How did structured data actually get here?

It helps to know the shape of the last fifteen years, because 2026 makes a lot more sense as the latest point on a curve than as an isolated news cycle.

Schema.org launched in June 2011, an unusual moment of cooperation between four rival search engines, Google, Microsoft's Bing, Yahoo, and Yandex, who set aside competition to agree on one shared vocabulary rather than forcing webmasters to write different markup for each of them. The initial release covered roughly 297 types and 187 properties. That collaborative, open-community structure, anyone can propose a new type through GitHub, and types graduate from a "pending" staging area into the core vocabulary through review, is why it's grown steadily every year since, rather than being frozen at whatever Google originally found useful.

The 2010s and early 2020s were the rich-result era. As Google built visual features on top of the vocabulary, recipe cards, review stars, FAQ dropdowns, breadcrumb trails, schema markup became, for most site owners, synonymous with "the thing that makes my search result prettier." That's a fair characterization of how most people actually used it during that period, and it's also why the recent wave of deprecations feels so disruptive to anyone who only ever thought about schema in those terms: if the entire value proposition was the visual payoff, and Google keeps retiring the visual payoffs, it can look like the whole discipline is being wound down.

What that framing misses is that the vocabulary itself never stopped growing, and Google's rich-result gallery was always a small subset of it, not the whole thing. That distinction is the key to understanding where things stand now, so it's worth making explicit.

Google's rich results and schema.org's vocabulary are two different things

This is the single most common point of confusion in 2026 coverage of structured data, including a lot of coverage that otherwise gets its facts right, so it's worth separating clearly.

Google's structured data gallery, the list of types that can trigger a specific visual treatment in Google Search, currently covers roughly 25 supported features, several of which (Product especially) unpack into multiple sub-features. This is the list that's been narrowing since 2023, and it's the list every deprecation headline this piece has discussed refers to.

The schema.org vocabulary itself is a completely different, much larger thing. As of its most recent release (version 30.0, March 2026), it covers around 823 to 827 types and roughly 1,500 properties, spanning far more ground than anything Google displays visually, medical entities, datasets, creative works, actions, and hundreds of other types most sites will never use, alongside the everyday ones like Article, Product, and LocalBusiness that most sites will. That vocabulary hasn't shrunk. It's grown every year since 2011, through an open community process that isn't controlled by any single search engine's rich-result roadmap.

Confusing these two is what produces the most common current mistake: treating a Google rich-result retirement as a retirement of the underlying schema.org type. FAQPage the rich result is gone. FAQPage the schema.org type is not, it's still a perfectly valid, documented type in the vocabulary, and other consumers of structured data (including AI systems that don't share Google's specific display rules) can still make use of it even without a Google visual payoff attached. The same logic applies to HowTo, Claim Review, and everything else on the retired list. Knowing which of these two lists you're actually talking about, Google's roughly 25-feature display gallery, or schema.org's 800-plus-type vocabulary, resolves most of the "is X dead" confusion on sight.

Why is Google retiring rich results if structured data still matters?

Because the two things being narrowed and reinforced aren't in tension, they're the same strategic move, seen from two angles.

What's been cut from Google's gallery is consistently the narrow, vertical-specific, easily-abused stuff: FAQ dropdowns that sites stuffed with keyword-bait Q&As purely to take up more space in results; HowTo cards that cluttered mobile results without reliably answering the query; niche types like Estimated Salary and Claim Review that served a small use case and saw limited genuine adoption. What's been protected and expanded, without exception, is the stuff that describes what something is, organizations, products, articles, local businesses, events, reviews. That's not a coincidence. It's Google reallocating its rich-result real estate away from decorative SERP features and toward the identity layer that both its own AI systems and outside ones like ChatGPT and Perplexity increasingly depend on to understand a page before they'll cite it.

The framing in Google's own messaging has shifted accordingly, less about earning a badge in the results, more about helping systems comprehend content correctly. A narrower, cleaner set of rich results and a more important entity layer are two symptoms of the same underlying shift, not a contradiction to reconcile.

What has Google actually retired, and what does that mean for your site?

Here's the accurate, dated shortlist, because a lot of the guidance circulating in 2026 gets this wrong in one direction or the other.

Fully retired rich results since 2023: HowTo (removed from mobile in August 2023, from desktop in September 2023), Sitelinks Search Box (November 2024), and, in the June 2025 wave, Book Actions, Course Info, Claim Review, Estimated Salary, Learning Video, Special Announcement, and Vehicle Listing. Practice Problem tooling followed in January 2026. FAQPage was fully retired on May 7, 2026, including the government-and-health carve-out that had kept it alive since 2023.

A nuance that trips up a lot of guides: Course Info was retired in that June 2025 wave. Course list was not, and still produces a rich result today. "Course schema is dead" is a common but inaccurate simplification of what actually happened.

What "retired" means in practice, every time: the schema.org type usually stays completely valid, see the vocabulary-versus-gallery distinction above. Your markup doesn't break, doesn't get penalized, and doesn't need to be torn out. What disappears is the visual treatment in Google's results, the panel, the carousel, the stars. If you're running HowTo or FAQPage markup today, there's no urgency to remove it; there's just no reason to expect the rich result it used to earn.

What genuinely does need action, and this is the distinction most deprecation coverage blurs: a handful of these changes weren't display changes at all. When Google made returnPolicyCountry a required field in return-policy markup, that's a markup-completeness requirement, implementations from before the change may now be incomplete and need an update. That's a fundamentally different kind of change from a rich result disappearing, and it's worth knowing the difference so you spend urgency where it's actually warranted. We keep a dated, sourced log of every individual change, retirements, additions, and requirement changes, in our running 2026 structured data changelog, if you want the change-by-change detail behind this summary.

What are AI systems actually doing with structured data?

This is the part most 2026 guidance either overstates or ignores entirely, so it's worth being precise, and worth breaking down by platform, since "AI search" is not one thing with one behavior.

The honest baseline, stated by Google itself: there is no special schema you need for AI Overviews or Google's AI Mode. Optimizing for AI search is, in Google's own framing, optimizing for search, the same fundamentals of quality, relevance, and clarity apply, and no markup unlocks a guaranteed AI citation. Anyone promising otherwise is overselling.

Google's AI Mode works by decomposing a query into a set of sub-questions, often referred to as query fan-out, researching each, and synthesizing an answer. Structured data helps at the research stage: an Organization or Article entity with clear, machine-readable facts is easier for that synthesis process to verify and attribute correctly than the same facts embedded only in prose. When John Mueller was asked directly on Reddit in 2026 whether schema helps with LLMs, his answer was genuinely nuanced rather than a flat yes or no: it depends on the specific feature and how that system uses it, some features, like Shopping results, lean heavily on structured data for the specific facts (price, availability, shipping); others use it more lightly, to enrich rather than to determine. That "it depends, by feature" answer is a more honest summary of where things stand than either "schema is essential for AI" or "schema does nothing for AI," both of which you'll see stated confidently and neither of which is quite right.

ChatGPT Search retrieves live pages through a combination of its own crawler and a Bing-index layer, then re-ranks for how cleanly a page answers the query. OpenAI has confirmed that ChatGPT uses structured data to help determine what it surfaces from a page, meaning a page with clear Product, Organization, or Article markup gives the retrieval-and-ranking step less to infer and more to simply read.

Perplexity functions similarly to ChatGPT Search in that it retrieves and cites sources directly and visibly, which makes it a useful, low-friction way to see whether your entity clarity is landing at all, because unlike some AI answers, Perplexity shows its sources in the open, you can check in real time whether structured, well-labeled pages are the ones it's pulling from.

The concrete mechanism common to all three: AI systems reason in entities, not keywords. Whichever platform is doing the retrieving, it's ultimately asking the same question about any given page, what is this, who made it, and can I trust it enough to cite it. Structured data is the most direct, unambiguous way to answer that question in a form a machine doesn't have to guess at. Clear Organization and Person markup, accurate Article or Product data, and a consistent entity identity across your site remove exactly the ambiguity that causes a system to hedge, cite a competitor it understands better, or quietly leave you out. We go deeper on the specific, practical version of this for one platform in our ChatGPT citation playbook.

The one piece of independent product news from 2026 that illustrates this shift concretely is worth knowing regardless of what tools you use. In March 2026, Yoast, arguably the most widely deployed schema implementation in WordPress, shipped Schema Aggregation, built with Microsoft's NLWeb protocol and Schema.org co-creator R.V. Guha. It exposes a site's entire connected structured-data graph through a single endpoint (what Yoast calls a "schemamap"), so an AI system can retrieve a complete, deduplicated picture of a site's organization, authors, and content in one request instead of crawling page by page to reconstruct it. Yoast has continued extending it since launch, an August 2026 update added a schemamap.xml file exposed at the site root by default. That's a serious, well-credentialed team betting real engineering effort specifically on making the entity graph, not any individual rich result, legible to AI systems at scale. When one of the most mainstream tools in the space ships a feature built entirely around that idea and keeps investing in it months later, it's a strong signal about where the field's center of gravity has actually moved, independent of any single vendor's positioning.

Where do the mainstream WordPress SEO tools stand on this?

Because most of the web still runs on WordPress, and most WordPress schema comes from a handful of SEO plugins, it's worth a fair, specific look at where the major players actually stand in 2026, not as a competitive knock, but as a snapshot of how the field is responding to the same shift this piece has been describing.

Yoast has the most entity-forward story of the mainstream tools. Its schema output has long been a genuinely connected @graph, distinct Organization, WebSite, WebPage, and content nodes cross-referenced by @id, rather than isolated blocks, and its 2026 Schema Aggregation feature, described above, extends that same connected-graph thinking outward to AI retrieval specifically. It's a suite-wide SEO tool with schema as one strong feature within it, not a dedicated schema product.

Rank Math has taken a different route: broad type coverage, a visual custom schema builder, and, as of 2026, its own AI-search-facing additions, including llms.txt support and an AI-traffic tracker that monitors references from AI-powered search engines. Its schema generator is widely regarded, including by direct competitors, as one of the most comprehensive in the WordPress ecosystem.

AIOSEO markets its schema and broader toolkit explicitly around "GEO" (generative engine optimization) positioning, with a next-generation schema generator and an emphasis on AI-search readiness alongside its more traditional SEO automation.

None of these are the "shrinking rich-result template" tools that were fair criticism a few years ago. All three have visibly moved toward AI-search framing in 2026, in their own ways, which is itself evidence for the thesis running through this whole piece: the market's center of gravity has genuinely shifted toward entity clarity and AI legibility, not just one vendor's marketing claiming it has. The honest distinction worth drawing, if you're evaluating tools specifically for how deep that entity work goes, is between an on-site connected graph (which several of these do well) and the external-validation and scored topical-authority work, sameAs networks, Wikidata connections, disambiguated knowsAbout mapping, that goes beyond your own site's graph toward recognition across the wider web. We lay that comparison out specifically and fairly for one of these tools in our Yoast schema comparison.

Is schema markup still worth doing if there's no guaranteed rich result or AI citation?

Yes, and the honest case for it is actually more durable than the old one.

Chasing a specific rich result was always chasing something Google could retire on a documentation update with no warning, as FAQPage sites just found out. Building a clear, accurate entity, the kind every AI system and search engine needs to understand who you are, what you make, and what you know, is a different kind of investment. It doesn't depend on one vendor keeping one visual feature alive. Every system that reasons about entities needs the same underlying clarity, and that need isn't going anywhere regardless of which specific rich results Google keeps or cuts next quarter.

That reframes the practical priority for 2026 into three tiers, roughly in order of durability:

Tier one: entity fundamentals. Organization, Person, and the core content types (Article, Product, LocalBusiness) that describe who you are and what you make. These have survived every retirement round since 2023 without exception, and they're what both Google's AI features and outside systems draw on most. If you do nothing else, this is the floor.

Tier two: connection. A single Organization entity, consistently referenced by @id across your Articles, your author Person entities, your Products and Services, rather than scattered, disconnected schema blocks. This is what Yoast's @graph approach gets right, and it's the same principle a dedicated entity profile builds on further, connected data reads as one coherent picture to a machine; disconnected data reads as a pile of unrelated facts about the same page.

Tier three: rich-result-specific types. Recipe, Event, Review, genuinely valuable where they apply, and worth doing well, but understand you're chasing a feature Google can retire on a Tuesday, the way it did to FAQPage. Treat these as a real bonus on top of a solid entity foundation, not the foundation itself.

Here's a simplified illustration of what tier-one-and-two thinking looks like in practice, the connected shape that both AI systems and Google's own remaining rich results reward, regardless of which specific types come and go around it:

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "Organization",
      "@id": "https://example.com/#organization",
      "name": "Bright Coffee Roasters",
      "url": "https://example.com",
      "logo": "https://example.com/logo.png",
      "sameAs": [
        "https://www.linkedin.com/company/bright-coffee-roasters",
        "https://www.wikidata.org/wiki/Q000000"
      ]
    },
    {
      "@type": "Person",
      "@id": "https://example.com/authors/maria-santos#person",
      "name": "Maria Santos",
      "worksFor": { "@id": "https://example.com/#organization" }
    },
    {
      "@type": "Article",
      "@id": "https://example.com/blog/single-origin-guide#article",
      "headline": "A Guide to Single-Origin Roasting",
      "author": { "@id": "https://example.com/authors/maria-santos#person" },
      "publisher": { "@id": "https://example.com/#organization" },
      "dateModified": "2026-07-14"
    }
  ]
}

Notice what this example doesn't chase: no FAQPage block, no HowTo steps, nothing built for a rich result that's already gone or could go next. Every node exists to establish identity and connect it to the next node by @id. That's the shape both search engines and AI systems reward right now, and it's the shape that survives whatever Google retires next.

What are the biggest structured-data mistakes in 2026?

A few keep recurring, and most of them are downstream of treating 2023-era advice as current.

Following an undated guide that still recommends FAQPage or HowTo for a rich result. Both have been fully retired from Google's results (FAQPage as of May 7, 2026, HowTo since September 2023). The markup itself won't hurt you, but building a content strategy around either expecting a visual payoff you won't get is wasted effort, and if a 2026-dated article confidently promises either result, it wasn't checked against current documentation before publishing.

Conflating "deprecated" with "harmful." The panic that followed Google's November 2025 notice, and the confusion around Course Info versus Course list, both trace back to this. Retired markup doesn't get penalized and doesn't hurt rankings, Google has said this repeatedly and explicitly. There is no reason to urgently strip retired schema out of a site; there's only a reason to stop expecting the rich result it used to earn.

Confusing Google's rich-result gallery with the full schema.org vocabulary. As covered above, these are different lists of different sizes tracking different things. A type disappearing from Google's roughly 25-feature gallery says nothing about whether it's still a valid, usable part of the 800-plus-type vocabulary that other systems, including AI crawlers, can still read.

Treating structured data as a guaranteed AI-citation lever. The honest position, supported directly by Google's own statement that no special schema is required for AI features, is that structured data supports comprehension and trust, which are real inputs into whether a system cites you, without being a lever you can pull for a guaranteed outcome. Anyone selling the guaranteed version is selling something Google's own documentation doesn't support.

Marking up content that isn't genuinely on the page. This one hasn't changed and won't. It was always the fastest route to a manual action long before any 2026 update, and it remains true independent of any specific deprecation or algorithm change. Markup that overstates or invents what's visible is the one mistake in this list that was never about currency, it's just always been wrong.

Ignoring the requirement changes hiding inside the display changes. Most 2026 deprecation news is about a rich result disappearing, which needs no urgent response. A smaller number, like returnPolicyCountry becoming required, are genuine markup requirement changes that can leave existing implementations incomplete. Treating every announcement with the same low urgency, or the same high urgency, means either missing something that matters or panicking over something that doesn't.

Letting markup drift out of sync with a changing page. This is the quiet, unglamorous failure mode behind a large share of "why did my rich result disappear" mysteries, and it has nothing to do with any Google policy change. A price updates, a review count grows, an author leaves, and static markup written once doesn't follow. Google requires markup to match what's genuinely visible on the page, a mismatch is a mismatch whether it was caused by a deprecation or by nobody updating a hardcoded block after a redesign.

How does this connect to entity-first schema?

This is the thread that ties the whole 2026 picture together, so it's worth stating plainly.

Everything this piece has covered, the narrowing of Google's rich-result gallery toward identity-describing types even as the underlying schema.org vocabulary keeps growing, AI systems across Google, ChatGPT, and Perplexity all reasoning in entities rather than keywords, Yoast and the other mainstream WordPress tools all visibly investing in AI-facing schema features in 2026, points at one underlying shift. Structured data's center of gravity has moved from decorating a search result to establishing a legible, connected entity. That's not a marketing framing invented to sell a tool; it's the pattern that falls out of tracking what Google has actually retired versus protected since 2023, what schema.org's own vocabulary growth shows, and what every major AI system has said about how it uses schema.

The practical implication is that "doing schema well in 2026" increasingly means the same thing as "having a strong entity profile," a precise, consistent core identity; genuine external validation connecting your brand to the rest of the web (sameAs links, a Wikidata entry); and a clearly declared, content-backed statement of what you're an authority on. Those three, core identity, external validation, and topical authority, are the dimensions worth actually building against and, ideally, tracking your progress on rather than treating as a vague aspiration. We cover what building that looks like, step by step, in our entity profile guide, it's the natural next read if this piece has convinced you the identity layer is where the real, durable value sits.

What's worth watching as this continues to develop?

A few open threads are worth keeping an eye on, stated as trends to watch rather than predictions to bank on, this space has already surprised people once in 2026 with how fast an established feature like FAQ disappeared entirely.

Whether NLWeb-style aggregation spreads beyond Yoast. If a single-endpoint, whole-site entity graph turns out to genuinely reduce AI crawling load and improve accuracy, it's a pattern other tools and platforms have obvious reason to adopt. Whether that happens, and how fast, isn't yet settled.

Whether more AI platforms publish specifics on how they weight structured data. Right now, the clearest public statements come from Google (no special schema required, but genuine use in AI Mode) and general confirmations from OpenAI and Microsoft that they use it. More granular disclosure, which types, how much weight, for which query types, would let this kind of analysis get considerably more precise than "it depends by feature."

Whether Google's gallery keeps narrowing at the same pace. Four distinct retirement waves since 2023 is a real pattern, not a one-off. Whether that continues at the same rate, slows, or reverses for any individual type is worth checking against the primary source rather than assuming either direction.

None of these are reasons to wait before acting on what's already well-established. They're reasons to keep checking primary sources rather than freezing your understanding at whatever this piece says today, which is exactly the discipline this piece has been arguing for throughout.

Implementing this on WordPress in 2026

Whatever platform or plugin you're on, the practical priorities that fall out of everything above are consistent: get your Organization and Person entities complete and connected by @id; don't build new content strategy around retired rich results; keep the markup you have accurate to what's genuinely on the page as it changes; and treat any genuine requirement change (not just a display retirement) with real urgency.

The recurring failure mode underneath all of this is drift. Markup gets written once, by hand, by a plugin's default settings, by whoever built the site originally, and the page keeps evolving while the schema doesn't. A price changes, a review count grows, an author leaves, a policy updates, and the structured data quietly stops matching what a visitor (and a crawler) actually sees. That mismatch is exactly what breaks eligibility for the rich results that remain, and it's exactly what gives AI systems an inconsistent picture of who you are.

This is the problem AI Schema Gen is built to solve structurally rather than through discipline. It reads your actual page content and generates schema across 827+ schema.org types from what's genuinely there, rather than a static template that drifts as your content changes, and it's built entity-first, meaning the priority is the connected Organization, Person, and topical-authority layer this piece has been describing, not just chasing whichever rich results Google hasn't retired yet. For teams validating markup as part of a build or deploy process rather than by hand, our guide to validating structured data in CI/CD covers the developer side of keeping schema accurate at scale.

How do I validate my structured data in 2026?

The tools haven't fundamentally changed, but knowing which one answers which question matters more now that Google's own testing tools have narrowed alongside the rich results themselves.

Google's Rich Results Test validates syntax and confirms eligibility for the specific rich results Google still supports. It's the right tool for anything on the current ~25-feature gallery, Product, Event, Recipe, Review, and so on. It's the wrong tool to expect a result from for a retired type (it won't flag HowTo or FAQPage as broken, it simply won't show a rich result, which is expected, not an error) or for a type like Service that never had a rich result to begin with.

The Schema Markup Validator at validator.schema.org checks general schema.org syntax and vocabulary validity across the full 800-plus-type vocabulary, independent of whether Google has a rich result for that type. This is your primary tool for entity-layer markup, Organization, Person, connected @graph structures, where correctness matters even without a visual payoff attached.

Search Console's structured data reports show what Google has actually parsed across your site over time, and its Manual Actions report is where you'd see a penalty for markup that doesn't match visible content. Watch this after any redesign or migration, since that's when drift most commonly creeps in unnoticed.

Direct queries to AI systems are the closest thing to a validation tool for the entity layer specifically. Ask ChatGPT, Perplexity, or Google's AI directly about your brand and your topics. Do they identify you correctly? Is what they say accurate? That's the real-world test of whether your entity clarity is landing, and it's free, immediate, and not gated by any single tool's rich-result support list.

AI Schema Gen generates entity-first structured data from your actual content across 827+ schema.org types, and builds the connected entity profile that keeps working regardless of which individual rich results Google retires next. See where your site stands, free, at aischemagen.com.