- Schema markup labels meaning, not appearance. It tells a machine that a number is a price and a name is the author, so nothing has to be inferred from the layout.
- Schema.org 30.0 defines 823 types. Google renders about 25 of them. The gap between those numbers is where most schema effort gets wasted.
- FAQ rich results ended on May 7, 2026, and Google removed the documentation on June 15. HowTo went the same way in 2023, and seven more types were retired in June 2025.
- Schema is not a ranking factor. It changes how your listing looks and how reliably machines resolve your facts, which is a different lever from position.
- Google says no special schema is required for AI Overviews or AI Mode. Microsoft says schema helps its models understand content. Both are true: it is a clarity multiplier, not an entry ticket.
- The highest-value move is not more types, it is linking the ones you have with
@idso your pages describe one connected entity instead of a pile of disconnected snippets.
Most schema advice was written for a search results page that no longer looks like that. The promise was straightforward: add the code, get stars under your listing, watch the click-through rate climb. Then Google spent three years quietly removing the features that promise depended on, while at the same time the machines reading your markup stopped being only search crawlers.
Schema markup is code you add to a page, almost always as JSON-LD, that labels what your content means rather than how it looks. It tells a machine that this number is a price, this person is the author, this business is in Manchester. Search engines use those labels to build rich results, and AI systems use them to resolve facts about you with less guessing.
What is schema markup?
Schema markup is a shared vocabulary for describing things on a web page in a way software can read without interpretation. A human reading £49 next to a product name knows it is the price. A machine sees a string. Schema removes the guess by naming it: this is a Product, that string is its price, this other one is its availability.
The vocabulary lives at Schema.org, a project run jointly by Google, Microsoft, Yahoo, and Yandex since 2011. Version 30.0, published on March 19, 2026, defines 823 types and 1,529 properties covering everything from recipes to medical trials to ski resorts. You will use perhaps five of them.
The mental model that helps: your page has visible content for people and an invisible data layer for machines. Schema is that second layer. It does not change what a visitor sees, which is exactly why it is easy to get wrong and easy to leave broken for years without noticing.
Did Google just kill schema markup?
No, but it did kill a large part of the reason most small sites added it. FAQ rich results stopped appearing in Google Search on May 7, 2026, a deprecation notice went up on May 8, and the documentation was removed entirely on June 15. Teams pulling FAQ data from the Search Console API were given until August 2026 to adjust their calls.
This was not a one-off. It is the fourth round of the same decision. HowTo rich results were cut back on mobile in August 2023 and removed from desktop that September. In June 2025 Google retired seven more types at once: Book Actions, Course Info, Claim Review, Estimated Salary, Learning Video, Special Announcement, and Vehicle Listing. The stated reasoning each time was consistent, that the features saw little use or no longer added enough value to searchers.
Read the pattern rather than the individual losses. Google keeps the types where structured data genuinely improves a result for a searcher, which means facts with a shape: a price, a rating, a date, an address, a cooking time, a job salary. It retires the types that mostly let publishers take up more vertical space with text they already had on the page. FAQ markup was the clearest example of the second category, and it was abused hard enough that its removal surprised very few people who had been watching.
Practically, if you have FAQPage markup on your site right now, leave it. Google confirms it causes no problem, the type is still valid Schema.org vocabulary that other consumers can parse, and stripping it out is effort spent for zero return. We still ship FAQPage on this article for exactly that reason. Just stop counting on it to change what your listing looks like.
What does schema markup actually do in 2026?
Schema does three separate jobs, and the confusion in most guides comes from treating them as one. Only the first is shrinking. The second and third have grown, which is why the correct response to the FAQ deprecation is to redirect schema effort rather than abandon it.
It earns rich results, on a shrinking list of types
This is the job everyone knows. Valid Product markup can put a price and star rating under your listing. Recipe markup adds cook time and calories. Breadcrumb markup replaces the raw URL with a readable path. The listing gets physically bigger and more informative, so it tends to earn a higher click-through rate at the same position. The catch is the shortlist: about 25 features in Google's gallery as of its June 15, 2026 update, and the list has only gone one direction for three years.
It resolves your brand as an entity
Search and AI systems do not think in pages, they think in entities: people, organisations, products, places, and the relationships between them. Organization and Person markup, with sameAs links pointing at your LinkedIn, Crunchbase, and Wikidata profiles, is how you state plainly which entity a page belongs to instead of hoping it gets inferred. This matters most for brands with ambiguous or common names, where an engine has to decide whether you are the software company or the restaurant. Our entity SEO chapter covers the full model, and the entity checker tests whether engines currently resolve your brand at all.
It hands machines facts they do not have to extract
An AI system answering a question about your product can either parse your prose and hope, or read a labelled field. Structured data is the low-error path, and it is the one job that got more valuable as answer engines multiplied. Prices, availability, opening hours, author credentials, and publish dates are all facts that models otherwise reconstruct from context and sometimes get wrong. This is also the job with the least hype attached, which is roughly why it is underinvested.
Notice that jobs two and three do not depend on Google rendering anything. They pay off inside AI Overviews, ChatGPT answers, Bing Copilot, and any agent parsing your site, none of which need a rich result to exist. That is the reframe: schema stopped being a SERP decoration tactic and became part of how you are represented.
Which schema types are worth adding?
Four types cover most sites, and a fifth depends on what you sell. Adding more than you need is the most common form of schema busywork, because each type you ship is one more thing that can drift out of sync with the page and start throwing errors nobody reads.
| Type | What it does in 2026 | Add it if |
|---|---|---|
| Organization | Defines your brand entity, logo, and sameAs profiles | Always. This is the one everyone skips |
| WebSite | Names the site and enables the sitelinks search box | Always. One block, sitewide |
| BreadcrumbList | Still renders a readable path in place of the URL | Always, on any site more than one level deep |
| Article / BlogPosting | Author, dates, and publisher for news and blog surfaces | You publish content of any kind |
| Person | Author identity, credentials, and profile links | Named authors write your content |
| Product | Price, availability, and rating in the result | You sell things. Highest payoff type left |
| LocalBusiness | Address, hours, phone, and geo for local surfaces | You have a location or serve an area |
| Review / AggregateRating | Star ratings, within Google's genuine-review rules | You hold real, verifiable reviews |
| Event | Dates, venue, and ticket status | You run events, classes, or webinars |
| VideoObject | Thumbnails, key moments, and video surfaces | You host original video |
| JobPosting | Salary and location in Google job surfaces | You hire directly through your site |
| FAQPage | No rich result since May 7, 2026. Still valid vocabulary | Keep what you have. Do not build new for SERP gain |
| HowTo | Retired from all Google surfaces since 2023 | Never, for search purposes |
The ordering carries the argument. The three types at the top of that table produce no dramatic listing change and are the ones most sites never bother with, while the retired types at the bottom are the ones agencies spent 2019 to 2023 selling. Organization markup earns you nothing visible today and quietly does more for how you are understood than any rich result on the list.
Does schema markup get you cited by AI?
It helps and it is not the gate, and anyone selling you either extreme is overstating their case. The two companies that would know have both said so publicly, in terms that sound contradictory until you read them carefully.
Google's own documentation on AI features is unambiguous: "There's also no special schema.org structured data that you need to add." The stated requirement to appear as a supporting link in AI Overviews or AI Mode is that the page is indexed and eligible to show with a snippet. Microsoft has taken a warmer line, with Bing principal product manager Fabrice Canel telling SMX Munich in March 2025 that schema markup helps Microsoft's language models understand content.
Both statements can hold at once. Schema is not an eligibility requirement, so a page with no markup can absolutely be cited. It is a disambiguation aid, so a page with clean markup gives the model less to infer and fewer opportunities to attribute your facts to somebody else. Marginal advantages compound when thousands of pages compete for a handful of citation slots, which is the honest version of the claim.
What matters more than schema for citations is whether crawlers can reach and read you in the first place. Structured data injected by a tag manager after page load is invisible to most AI crawlers, because they do not execute JavaScript. Blocked user agents remove you entirely. Both of those failures outrank any markup decision, and we cover them in which AI crawlers to block and which to allow and the crawler access chapter. The mechanics of getting quoted once you are readable are in how to rank in Google AI Overviews and ChatGPT SEO.
How do I add schema markup to my website?
Five steps, and the fourth and fifth are the ones that separate markup that works from markup that merely validates. Use JSON-LD throughout: Google recommends it, it sits in a single script tag instead of being woven through your HTML attributes, and it is the only format you can edit without touching your templates.
Pick the one type that describes the page
One primary type per page, matching what the page actually is. A blog post is an Article or BlogPosting. A product page is a Product. A location page is a LocalBusiness. Then add the sitewide supporting cast: Organization, WebSite, and BreadcrumbList. If you find yourself stacking four types on one page to squeeze out more features, you have left the useful part of this exercise.
Generate the JSON-LD
Free generators from TechnicalSEO.com and Merkle produce valid blocks for the common types, and Google's own documentation carries a complete example for every supported feature. On WordPress, check first: Rank Math, Yoast, and Schema Pro already emit Article, Organization, and Breadcrumb markup, and layering a second source on top is how sites end up with two conflicting Organization nodes. On a custom stack, build the object in your template so it is generated from the same data that renders the page.
Put it in the server-rendered HTML
The block belongs in a <script type="application/ld+json"> tag that is present when the page is fetched, not one injected later by Google Tag Manager. Googlebot renders JavaScript and will usually find the late-injected version eventually. The AI crawlers that decide whether you get cited mostly do not, so tag-manager schema is invisible exactly where it now matters most. View source, search for ld+json, and confirm it is there before rendering. The rendering chapter covers the wider version of this failure.
Make it match what the visitor sees
Every fact in your markup must be visible on the page. Marking up a price of £49 while the page shows £79, or claiming an aggregate rating with no reviews on display, is a structured data violation and it is enforced with manual actions, not just lost features. This trips up honest sites too: prices change, stock sells out, and static schema drifts out of sync with a dynamic page. Generate the values from the same source that renders them and the problem cannot occur.
Connect the nodes with @id
This is the step almost every guide omits, and it is where the entity payoff lives. Give each node a stable identifier, then reference it instead of repeating it. Your Organization gets @id: "https://example.com/#organization", and every article on the site points its publisher at that same identifier rather than restating your company details. Your author pages, articles, and breadcrumbs resolve to one linked graph instead of thousands of disconnected snippets. That is the difference between a machine learning that your site published 400 articles and learning that one identifiable organisation, with named authors who have credentials, published all of them. The schema chapter in our playbook works through the full graph.
How do I check my schema markup is working?
Three tools, each answering a different question, and you want all three because passing one proves very little about the others. Run them in this order.
- Google's Rich Results Test. Answers "would Google show a feature for this". It only reports on types Google supports, so a valid Organization block returns nothing eligible, which is not an error. Test the live URL rather than pasted code so you also catch rendering problems.
- The Schema Markup Validator at validator.schema.org. Answers "is this valid Schema.org at all", including the many types Google does not render. This is the one that tells you whether your
@idreferences actually resolve. - Search Console's Enhancements reports. Answers "what does Google see across the whole site, at scale, over time". The other two check one page in a lab. This one catches the template that started emitting broken markup three weeks ago on 4,000 URLs.
One test the tools will not run for you: fetch your page with JavaScript disabled and confirm the JSON-LD is still in the source. That single check catches the most expensive schema failure there is, and it takes about ten seconds. It is check nine in our 12-point SEO audit.
What breaks schema markup?
Schema fails quietly. Nothing on the page changes, no visitor complains, and the loss shows up as features you never got rather than something that visibly broke. These are the five failures worth checking for by name.
- Marking up content that is not on the page. The most serious one, because it is a guidelines violation rather than a mistake. Invisible FAQ blocks and phantom ratings both fall here.
- Two sources emitting the same type. A plugin and a hand-pasted block both declaring Organization gives search engines two answers to one question. Audit what you already output before adding anything.
- Schema that only exists after JavaScript runs. Passes the Rich Results Test, invisible to most AI crawlers. The gap between those two outcomes is now the whole problem.
- Copied examples with the placeholders left in. Validators happily approve markup describing a company in Mountain View that is not yours, because the syntax is fine. Read every value.
- Orphan nodes. Valid blocks with no
@idand no references between them. They parse, they just never assemble into anything a machine can reason about.
If you want the wider context on how machines assess credibility rather than just parse it, the E-E-A-T for machines chapter covers the signals that sit alongside markup, and building an SEO-optimized website puts schema in sequence with everything else.
Frequently Asked Questions
Schema markup is structured data you add to a page, almost always as JSON-LD, that labels what your content means rather than how it looks. It tells a machine that a number is a price, a name is the author, or a block of text is a review. Search engines use those labels for rich results and to understand your entities.
No, not directly. Google has stated repeatedly that structured data does not boost rankings on its own. What it does is change how your result appears and how reliably machines understand the page. The traffic gains people attribute to schema come from higher click-through rates on richer results, not from a higher position.
Structured data is the general concept: any standardised, machine-readable data about a page. Schema markup is the specific vocabulary most of the web uses for it, maintained at Schema.org by Google, Microsoft, Yahoo, and Yandex. In practice the two terms get used interchangeably, and nobody will misunderstand you either way.
Not for a Google rich result. FAQ rich results stopped appearing in Google Search on May 7, 2026, and Google removed the documentation on June 15, 2026. FAQPage remains valid Schema.org vocabulary, so it still describes your content for other consumers, and leaving existing markup in place causes no harm.
Rich snippets, which Google now calls rich results, are search listings enhanced with extra detail pulled from your structured data: star ratings, prices, stock status, event dates, cooking times, or breadcrumbs. They occupy more space and typically earn a higher click-through rate than a plain blue link at the same position.
Most SEO plugins already emit it. Rank Math, Yoast, and Schema Pro all output Article, Organization, and Breadcrumb markup automatically, so check what you already have before adding more. For anything custom, paste a JSON-LD block into the header via your theme or a code-snippet plugin, then confirm it appears in the page source.
It helps, but it is not the gate. Google states plainly that no special structured data is needed to appear in AI Overviews or AI Mode. Microsoft has been more positive, with Bing product lead Fabrice Canel saying at SMX Munich in 2025 that schema helps its language models understand content. Treat it as a clarity multiplier, not an entry ticket.
Schema.org version 30.0, released March 19, 2026, defines 823 types and 1,529 properties. Google supports about 25 of those as rich result features. Most sites need four: Organization, WebSite, BreadcrumbList, and whichever type describes the page, such as Article, Product, or LocalBusiness.
The Bottom Line
Schema markup is worth less than it was for decorating a search listing and worth more than it was for being understood. Google has retired FAQ, HowTo, and seven other types since 2023, and about 25 features remain in its gallery against 823 types in the vocabulary. If your schema strategy is a hunt for the next rich result, that trend line is not going your way.
The version that still pays is unglamorous. Ship Organization, WebSite, and BreadcrumbList sitewide. Add the one type that describes each page. Put it in the server-rendered HTML, keep every value true to what visitors see, and connect the nodes with @id so your site resolves as one entity rather than a thousand fragments. That takes an afternoon, produces no screenshot worth celebrating, and is the part that machines actually read. For where it sits in the wider system, work through the GEO Playbook, or see how we handle it as technical GEO and entity SEO.