Semantic HTML is the practice of building a web page from HTML elements chosen for what the content is rather than how it should look. A heading goes in a heading element, a list in a list element, the site menu in a nav element and a publication date in a time element, so that software can tell the parts of a page apart without seeing them. It exists because screen readers, search engine crawlers, reader modes and, more recently, AI agents read pages as well as browsers, and depend on markup to understand structure. The HTML Standard states that elements, attributes and attribute values are defined "to have certain meanings (semantics)", letting browsers and search engines "present and use documents and applications in a wide variety of contexts that the author might not have considered."

How meaning travels through markup

Landmark elements divide a page into regions: header, nav, main, aside, footer and, since 2023, search. Structural elements organise the content inside those regions: h1 to h6 for headings, ul, ol and li for lists, table with th for data, figure with figcaption for an illustration, article for a self-contained piece and section for a thematic group. Inline and control elements label smaller items: time, whose datetime attribute holds a machine-readable date, button for an action and label for a form field's name. Two elements, div and span, carry no meaning at all.

A browser parses the HTML into the document object model (DOM), then derives the accessibility tree, a parallel structure that exposes each element's role, name and state to assistive software. Under the HTML Accessibility API Mappings (HTML-AAM) specification from the World Wide Web Consortium (W3C), a nav element becomes a navigation landmark, main becomes main, aside becomes complementary, search becomes search, and a header or footer not nested inside an article or section becomes a banner or contentinfo landmark. A section becomes a region landmark only if it has an accessible name; otherwise it is as generic as a div.

Screen reader users jump between those regions and step through headings with a single key. The HTML Standard now defines a page's outline as "all headings in a document, in tree order". At least one should be level 1, and each heading must sit no more than one level deeper than the one before, so a jump from h2 to h4 is non-conforming.

A product page shows the difference. Built semantically, it has a header and nav, then a main region with an h1 for the product name, a time element for the release date, a table whose th cells label each specification and a button reading "Add to basket", which is focusable, fires on Enter or Space and is announced as a button with no extra code. Built from styled div elements, it looks identical, but each fake control needs an Accessible Rich Internet Applications (ARIA) role, a tabindex value and keyboard handlers to match the native element. The W3C's Nu Html Checker tests conformance, and audit tools such as Google's Lighthouse flag faults like headings out of order.

From font tags to a living standard

Through the 1990s, pages mixed meaning and appearance, with font tags and table layouts. HTML 4, completed in 1997 according to the HTML Standard's own history, deprecated presentational markup and offered a "Transitional" mode to ease authors away from it. In 1998 the W3C turned to XHTML, an XML reformulation completed as XHTML 1.0 in 2000.

After a 2004 W3C workshop rejected their proposal to extend HTML, Apple, Mozilla and Opera formed the Web Hypertext Application Technology Working Group (WHATWG) to continue the work. The W3C rejoined in 2007 through a joint working group, and the first W3C draft of HTML5 followed in January 2008. HTML5 added the sectioning vocabulary in use today: header, nav, article, section, aside, footer, figure, figcaption and time, with main added later. The spec explained the shift bluntly: presentational markup leads to "poorer accessibility", a "higher cost of maintenance" and "larger document sizes", so it "has been removed from HTML in this version." The W3C published HTML5 as a Recommendation on October 28, 2014.

In 2011 the WHATWG chose a continuously maintained Living Standard while the W3C wanted a finished "HTML5". The same year, on June 2, 2011, Google, Microsoft's Bing and Yahoo launched Schema.org, a shared vocabulary that formalised a machine-facing description layer separate from structural tags. In 2019 the WHATWG and W3C agreed to maintain a single version of HTML, the WHATWG's Living Standard.

The search element reached Safari, Firefox and Chrome between September 18 and October 10, 2023, according to Mozilla's MDN Web Docs. The Web Content Accessibility Guidelines (WCAG) 2.2, published on October 5, 2023, removed success criterion 4.1.1, Parsing, as obsolete because "assistive technology no longer has any need to directly parse HTML".

Why marketers and publishers deal with it

Law is the most direct pressure. WCAG success criterion 1.3.1, Info and Relationships, at Level A, requires that structure "conveyed through presentation can be programmatically determined", and native elements are the simplest way to meet it. The European Accessibility Act, Directive (EU) 2019/882, has applied since June 28, 2025 to e-commerce, banking and other consumer services, with compliance generally measured against EN 301 549, which incorporates WCAG 2.1 Level AA. In the US, the Department of Justice adopted WCAG 2.1 AA for state and local government sites in 2024, while private businesses face litigation: plaintiffs filed 3,117 website accessibility lawsuits in US federal courts in 2025, up 27% on 2024, according to law firm Seyfarth Shaw.

Search is the second reason. The HTML Standard notes that search engines can use headings "to more effectively index a page". Google renders JavaScript and indexes the resulting DOM, so script-generated markup counts as well as server HTML. In February 2026 Google cut the amount of HTML Googlebot fetches per file from 15MB to 2MB, making markup weight a crawl concern for the heaviest pages.

AI systems have widened the readership. Google named three ways agents perceive a site: screenshots, HTML source and the accessibility tree, the last built from semantic elements. Shortcuts around HTML have met resistance: Google and Bing said separate markdown pages for AI violate search policies, and llms.txt adoption rose 8.8 times while 97% of files received no AI requests. Conversion tools work from the HTML itself; Cloudflare cut AI token costs by 80% with markdown conversion, a conversion that can only produce markdown headings and lists where the source HTML marks them.

The HTML Standard defines no advertising element, so ad slots are usually empty div containers filled by scripts, as consent banners often are. Where those containers sit, inside main or in an aside, decides how assistive technology and extraction tools treat them. The link to Core Web Vitals is loose: semantic tags do not make a page faster, though fewer nested wrappers mean fewer elements to lay out.

Where it falls short

Adoption is partial. In the 2024 HTTP Archive Web Almanac, div made up 28.7% of all elements on mobile pages, and the authors concluded that "'divitis' is still a thing". The 2025 edition counted main landmarks on 47% of pages, against 71% for navigation, 68% for footers and 67% for headers. The ARIA button role appeared on about 53% of pages, which the authors said often indicates div or span elements dressed up as buttons, or redundant roles on native ones.

Misuse is common: headings chosen for font size, landmark roles duplicated on elements that already carry them, and role="presentation" on a list, which strips meaning from every item. Custom elements carry no built-in meaning, and pages using them loaded a median 1,286KB of JavaScript against 522KB for pages without, according to the 2024 Almanac. The 2025 Almanac reported that Stimulus, Remix and Qwik historically scored several points above React, Svelte and Ember on accessibility audits, "likely because they prioritize progressive enhancement and semantic HTML."

The search benefit is disputed. The MDN glossary says search engines "will consider its contents as important keywords to influence the page's search rankings". Search Engine Journal has reported Google's position that semantic HTML is not a ranking factor, though still worth using. Automated testing has limits too. "Heading hierarchy can be audited, but semantic meaningfulness requires human judgment," the 2025 Almanac authors wrote. Overlays promising automatic fixes are contested, and a standards-based website spec from Joost de Valk lists them as something to avoid.

Not the same as

  • Structured data. Structured data, usually Schema.org vocabulary in the JSON-LD format, makes explicit statements about entities such as products, prices and reviews, often for rich results. Semantic HTML describes the document's own structure; a page can have either without the other.
  • The DOM. The live tree a browser builds from markup. Semantic HTML decides what kinds of nodes it holds, and so what the accessibility tree exposes.
  • ARIA. The W3C's ARIA specification adds or overrides roles and states through attributes. It fills gaps where no native element exists, such as tabs; the Almanac authors describe favouring native elements first as "ARIA's primary guideline".
  • Alt text. Alt text is one attribute supplying a text equivalent for an image. Semantic HTML is the wider discipline it belongs to.

Recent developments

As of October 2026, the HTML Standard includes a headingoffset attribute that raises the computed level of headings inside a component; MDN marks it experimental. Google's crawler list now includes Google-Agent, a user-triggered fetcher for AI browsing. On October 1, 2026, Google extended its call for human review to AI-written image descriptions.

Timeline

  • 1997: HTML 3.2 completed, followed later that year by HTML 4, which deprecates presentational markup
  • 1998: The W3C stops evolving HTML in favour of XHTML
  • 2000: XHTML 1.0 completed
  • 2004: Apple, Mozilla and Opera form the WHATWG after a W3C workshop rejects their proposal
  • 2007: The W3C charters a working group to develop HTML5 with the WHATWG
  • January 2008: First W3C draft of HTML5 published
  • 2011: The WHATWG moves to a Living Standard as the W3C pursues a finished HTML5
  • June 2, 2011: Google, Bing and Yahoo launch Schema.org
  • October 28, 2014: HTML5 becomes a W3C Recommendation
  • April 17, 2019: The European Accessibility Act is adopted
  • 2019: The WHATWG and W3C agree to maintain a single HTML standard
  • September 18 to October 10, 2023: Safari, Firefox and Chrome ship the search element
  • October 5, 2023: WCAG 2.2 published, removing the Parsing criterion
  • April 2024: The US Department of Justice adopts WCAG 2.1 AA for state and local government sites
  • June 28, 2025: European Accessibility Act requirements begin to apply
  • February 2026: Google cuts Googlebot's HTML fetch limit from 15MB to 2MB
  • April 20, 2026: The Department of Justice extends its Title II compliance dates by one year
  • October 1, 2026: Google extends human review guidance to AI-written alt text

Summary

Who. Web developers, content teams, SEO specialists and publisher ad operations staff write the markup; the WHATWG maintains the HTML Standard and the W3C maintains the accessibility mappings and WCAG; screen reader users, search engines, reader modes and AI crawlers and agents consume it; the European Union and US courts and regulators set the legal floor.

What. The use of HTML elements for their defined meaning rather than their appearance: headings for headings, lists for lists, nav for navigation, main for the primary content and button for actions. Browsers translate those elements into roles in the accessibility tree, which assistive technology and, increasingly, software agents read.

When. Presentational markup was deprecated in HTML 4 in 1997. The HTML5 sectioning elements arrived with the first W3C draft in January 2008, and HTML5 became a W3C Recommendation on October 28, 2014. The search element shipped in 2023, WCAG 2.2 followed on October 5, 2023, and the European Accessibility Act has applied since June 28, 2025.

Where. In the source code of every web page, in the DOM browsers build from it, and in the accessibility tree derived from that DOM, which is where screen readers, crawlers, markdown converters and browser agents meet the same structure.

Why. A page is read by more machines than browsers. Semantic HTML gives all of them the same map of what each part of a page is, which supports accessibility compliance, clearer indexing and more reliable extraction by AI systems, without depending on any one company's ranking factors.