Structured data is a layer of machine-readable labelling added to a web page so that software can identify what the page describes instead of inferring it from prose. A product page contains a price, a currency, an availability status and a rating. A reader takes those from the layout; a parser needs them declared. The markup makes the declaration explicit, in a shared vocabulary that search engines, shopping systems and AI assistants agree to read the same way. Its purpose has always been eligibility rather than persuasion: pages carrying valid markup can qualify for display treatments and data extractions that plain HTML cannot reach.

How the markup is written

Three syntaxes carry the same vocabulary. JSON-LD, short for JavaScript Object Notation for Linked Data, places the data in a standalone script block with the type attribute application/ld+json, usually in the document head. A minimal product block opens with @context set to https://schema.org and @type set to Product, then lists name, image, description, sku, brand and an offers object holding price, priceCurrency and availability. Microdata weaves the same facts into visible HTML through itemscope, itemtype and itemprop attributes on existing elements. RDFa, the oldest of the three, uses vocab, typeof and property attributes to the same end.

Google treats all three as valid when correctly implemented and recommends whichever is easiest to maintain, which for template-driven sites is normally JSON-LD, since the data sits apart from the presentation and can be generated server-side without touching the layout. A single page may carry several blocks, commonly an Article or Product block alongside a BreadcrumbList and an Organization declaration.

The vocabulary itself is schema.org, a set of types and properties maintained as an open project. Version 30.0 was published on 19 March 2026. Consuming platforms then impose their own layer of requirements on top: for any given feature, Google's documentation splits properties into required and recommended, and markup that omits a required property is ineligible for the feature even though it remains valid schema.org.

Where it sits in the processing chain

Markup only works if the crawler reaches it. Google cut Googlebot's fetch ceiling from 15MB to 2MB per URL in February 2026, and the documented behaviour at the threshold is truncation rather than rejection, so markup placed deep inside a very large HTML file can be invisible to the system that is meant to read it. Data injected by client-side JavaScript is seen only after rendering, which happens on a separate schedule from the initial fetch.

Extraction is literal. On 21 August 2026 Google changed Googlebot's JSON-LD handling to perform a single pass of HTML unescaping, a parsing detail that breaks markup which double-encodes entities, and which no amount of vocabulary correctness compensates for. After extraction comes eligibility, which is not the same as appearance: Google states that valid markup makes a page a candidate for a feature and nothing more. Search Console rich result reports and the Rich Results Test expose what was parsed, per feature, with errors and warnings separated.

Origin and evolution

The starting point was 12 May 2009, when Google announced rich snippets covering reviews and people, reusing existing microformats and RDFa vocabularies rather than inventing one. Google's own vocabulary, data-vocabulary.org, followed the same year. Two years later, on 2 June 2011, Google, Microsoft and Yahoo launched schema.org as a shared vocabulary; Yandex joined in November 2011.

The encoding question took longer to settle. JSON-LD 1.0 reached W3C Recommendation status on 16 January 2014, after roughly four years of work by a community group, and JSON-LD 1.1 followed on 16 July 2020. Google sunset data-vocabulary.org markup for rich results on 6 April 2020, having issued Search Console warnings from 20 January that year, which left schema.org as the single vocabulary in practice.

Feature expansion peaked around 2019, when FAQ, HowTo and Q&A markup arrived together. Contraction followed. Seven types went in one action on 12 June 2025: Book Actions, Course Info, Claim Review, Estimated Salary, Learning Video, Special Announcement and Vehicle Listing. Product manager Henry Hsu wrote that analysis showed the affected displays were not commonly used and no longer added significant value. Practice problem support ended from January 2026, alongside a clarification that Dataset markup feeds Dataset Search only and never ordinary results. FAQ rich results stopped appearing on 7 May 2026, with Search Console reporting dropped in June and API support in August.

What is actually deployed

Until mid-2026 nobody outside the search engines could measure adoption. Schema.org and Google published the first public usage dataset in June 2026, releasing domain-count buckets monthly in CSV and JSON. The May 2026 file holds 5,545 entries across 958 types and 4,587 properties, and the distribution is steep rather than gradual. Twelve types, 1.3% of the vocabulary, sit above ten million domains: BreadcrumbList, EntryPoint, ImageObject, ListItem, Organization, Person, PropertyValueSpecification, ReadAction, SearchAction, Thing, WebPage and WebSite. At the other end, 485 types, 50.6% of the total, fall below one thousand domains, and 82.4% of properties sit under the same floor. Half the vocabulary is effectively unused.

Crawl-based surveys give the format split. The 2024 Web Almanac, drawn from the HTTP Archive crawl of 16.9 million sites, found JSON-LD on 41% of mobile pages, up from 34% in 2022, with Microdata at 26% and RDFa at 66%, the last figure inflated because Open Graph tags are counted as RDFa. Schema.org's own summary, dated 2024, describes more than 45 million domains carrying over 450 billion objects.

Why it matters for marketing teams

The commercial types survived the contraction, and the maintenance burden falls on retail and publishing teams. Product variant support arrived in February 2024 through hasVariant and variesBy. Return policy markup gained a mandatory returnPolicyCountry field in March 2025, a two-letter ISO 3166-1 code supporting up to 50 countries. Loyalty programmes became expressible in June 2025 through MemberProgram, requiring description, hasTiers and name, with benefits declared as TierBenefitLoyaltyPoints, TierBenefitLoyaltyPrice, TierBenefitLoyaltyReturns or TierBenefitLoyaltyShipping. Shipping policies opened to merchants without Merchant Center accounts in November 2025. On 7 July 2026 the Product category property was extended to accept a CategoryCode object alongside plain text, aligning markup with Merchant Center feed attributes.

Publishing-side uses are narrower but persistent. Google confirmed on 2 March 2026 that both schema.org markup and the og:image meta tag feed thumbnail selection in Search and Discover. Profile page and discussion forum markup opened in November 2023. Apple takes the schema.org isAccessibleForFree property as a paywall signal, keeping marked pages indexable while excluding them from AI-generated output.

Limitations and contested ground

Markup does not improve rankings, and the evidence that it improves AI visibility is thin. Ahrefs tested schema implementation across a large sample of pages already appearing in AI Overviews and found no measurable ranking difference, a result consultant Marie Haynes cited when arguing that schema deserves little attention on informational content and real attention on commerce, where agents need accurate prices and availability. The counter-argument, that structured question-and-answer content raises citation probability regardless of rich result eligibility, remains unresolved by Google's own statements.

Deprecation risk is structural. Every feature listed above exists at the discretion of one company, and FAQ markup shows the pattern: launch, mass adoption, abuse, restriction, removal. Enforcement compounds it. A structured data manual action leaves a page indexed while instructing Google to ignore its markup, stripping the visible benefit without a traffic collapse to signal the cause. Undisclosed incentivised reviews joined that category on 24 July 2026. SISTRIX analysis of German search losses listed schema markup misuse among the shared characteristics of penalised sites.

The adoption dataset carries its own caveats, disclosed by its publishers: it reflects only the web as Google indexes it, excludes sites blocked in robots.txt, and does not separate JSON-LD from Microdata or RDFa.

Not the same as

Structured Data Files, the DV360 bulk CSV format abbreviated to SDF, share a name and nothing else; that format moves campaign configuration between a spreadsheet and a demand-side platform. Metadata tags such as Open Graph and Twitter cards describe a page for social preview rendering using a different vocabulary and no type hierarchy. Product feeds submitted to Merchant Center carry overlapping fields but travel as a separate file rather than living in the page, and the two can disagree. The Semantic Web, the W3C programme of machine-readable linked data, is the intellectual parent; schema.org markup is the fragment of it that achieved scale.

Recent developments

The direction of travel through 2026 has been commercial expansion and reference contraction running in parallel, with parsing rules tightening. FAQ support finished its withdrawal in August 2026. Merchant listing documentation gained category codes and sale duration guidance in July. The JSON-LD unescaping change landed in August. Proposals now circulate for exposing markup outside the page entirely: the Schemamap convention indexes one JSON-LD endpoint per resource through a schemamap.xml file, one of 128 items in the website specification published by Joost de Valk in May 2026. Only 7.4% of Fortune 500 companies publish an llms.txt file, while JSON-LD remains the mechanism actually deployed at scale for telling machines what a page contains.

Timeline

  • 12 May 2009: Google announces rich snippets for reviews and people, using microformats and RDFa
  • 2009: Google publishes data-vocabulary.org as its own markup vocabulary
  • 2 June 2011: Schema.org launched by Google, Microsoft and Yahoo
  • November 2011: Yandex joins schema.org
  • 16 January 2014: JSON-LD 1.0 published as a W3C Recommendation
  • 2016: Google introduces Rich Cards
  • 2017: Rich Results Test tool released
  • 2019: FAQ, HowTo and Q&A markup types introduced
  • 20 January 2020: Search Console begins warning on data-vocabulary.org markup
  • 6 April 2020: data-vocabulary.org markup becomes ineligible for rich results
  • 16 July 2020: JSON-LD 1.1 published as a W3C Recommendation
  • November 2023: Profile page and discussion forum markup supported
  • February 2024: Product variant support added through hasVariant and variesBy
  • 14 March 2025: returnPolicyCountry becomes required for merchant return policies
  • 10 June 2025: MemberProgram loyalty markup introduced
  • 12 June 2025: Seven structured data types deprecated
  • 5 November 2025: Practice problem markup deprecated; Dataset scope clarified
  • 12 November 2025: Shipping policy markup extended to merchants without Merchant Center
  • 6 February 2026: Googlebot fetch limit cut from 15MB to 2MB per URL
  • 2 March 2026: Google confirms schema.org and og:image both feed thumbnail selection
  • 19 March 2026: Schema.org version 30.0 published
  • 7 May 2026: FAQ rich results stop appearing in Google Search
  • June 2026: Schema.org and Google publish the first public adoption dataset
  • 7 July 2026: CategoryCode added to merchant listing markup
  • 24 July 2026: Undisclosed incentivised reviews become a structured data violation
  • 21 August 2026: Googlebot moves JSON-LD extraction to a single unescaping pass

Summary

Who: Schema.org maintains the vocabulary, governed by Google, Microsoft, Yahoo and Yandex as founding sponsors, with the W3C responsible for the JSON-LD, Microdata and RDFa encodings. Search engines, shopping platforms and AI systems consume it. Site owners, publishers, retailers and technical SEO teams produce it.

What: Machine-readable labelling embedded in a page, declaring types such as Product, Article or Organization and their properties, which makes the page eligible for rich results, merchant listings and structured extraction by automated systems.

When: Traceable to Google's rich snippets announcement of 12 May 2009, with schema.org launched on 2 June 2011, JSON-LD reaching Recommendation status in 2014 and 2020, vocabulary version 30.0 published on 19 March 2026, and the first public adoption dataset released in June 2026.

Where: In the HTML of the page itself, most often as a JSON-LD script block, validated through the Rich Results Test and monitored through Search Console rich result reports.

Why: Automated systems cannot reliably infer a price, a policy or an author from layout alone, and the systems now reading pages include shopping agents as well as search crawlers. Markup is the declared version of that information, which is why commercial types keep expanding while reference types are withdrawn, and why enforcement against false markup runs through a dedicated manual action.