Get now

Document structure

Landmarks and an unbroken heading order are how an agent builds a map of the page before acting on it.

Standard
Recommended
56%Top 100 adoption56 of 100 sites implement it8/9OverviewEstablishedWidely supported and widely deployed.

Who supports it

  • ClaudeSupported (announced)

    Published 1 Aug 2026 · last checked 9 Oct 2026: Claude API docs: Browser use toolread_page 'Returns the page's accessibility tree as text' (role + accessible name per element, e.g. button "Search"); Claude also uses screenshots. Date is approximate, taken from the toolset version id browser_toolset_20260801; page shows no date. The same read_page tool is exposed by the Claude in Chrome extension.

  • CopilotPartial support

    Published 8 Oct 2025 · last checked 9 Oct 2026: Microsoft Advertising blog: Optimizing your content for inclusion in AI search answers (Krishna Madhavan, Bing)Bing: H1 used to interpret the page; H2/H3 headings define 'where one idea ends and another begins' when AI parses content. About search-answer parsing rather than the accessibility tree; landmarks not mentioned.

  • ChatGPT AtlasSupported (announced)

    Published 1 Sept 2026 · last checked 9 Oct 2026: OpenAI Help Center: Publishers and Developers - FAQHelp Center article shows 'Updated: last month' as of 2026-10-09 (month-only date). Text: 'ChatGPT Atlas uses ARIA tags—the same labels and roles that support screen readers—to interpret page structure and interactive elements.' Explicitly covers 'page structure'.

  • Gemini in ChromePartial support

    Published 5 May 2026 · last checked 9 Oct 2026: Chrome for Developers: Accessibility for agents (Lighthouse Agentic browsing)Chrome team guidance for agents in general ('Agents review the accessibility tree to identify interactive elements'; 'Agents rely on the accessibility tree ... as their primary data model', toolkit blog 2026-06-22). It does not name Gemini in Chrome explicitly, hence partial.

23 others: no evidence yet. See every agent and browser

What it checks

The extension reads the rendered page’s outline in your browser and looks for three problems:

  • No <h1>. The page has no top-level heading.
  • Skipped heading levels. A heading jumps down more than one level, such as an <h2> followed directly by an <h4>. One skip is tolerated; two or more count as a problem.
  • No <main> landmark. Neither a <main> element nor role="main" is present.

Headings are read from inside <main> when the page has one, so headings in the navigation, sidebar and footer do not count against the outline. Hidden elements are ignored. Multiple <h1> elements are not flagged.

Results

Status When
Pass There is an <h1>, at most one skipped level, and a <main> landmark
Warn There is an <h1>, but levels are skipped more than once or there is no <main>
Fail The page has no <h1>
Unmeasured The extension could not read the page (open it in the active tab)

When levels are skipped, the result lists each offending pair and the heading it follows.

How to fix

Wrap the primary content in <main>, give the page one <h1>, and step heading levels down one at a time:

<body>
  <header>…</header>
  <main>
    <h1>Blue running shoes</h1>
    <h2>Sizing</h2>
    <h3>Wide fit</h3>
    <h2>Care</h2>
  </main>
  <footer>…</footer>
</body>

An agent reads the outline as a tree. A jump from <h2> to <h4> makes a section look like the child of the wrong parent, and without <main> it cannot skip the navigation and footer to reach the content.

Support by agent and browser

Announced means the vendor’s own docs or announcement say so. Published is the date on the vendor’s own page, where it has one. Checked is the last time our weekly research found that page still saying so.

AgentStatusSource
AI assistants
Claude AnthropicSupported (announced)

Claude API docs: Browser use tool Published · Checked read_page 'Returns the page's accessibility tree as text' (role + accessible name per element, e.g. button "Search"); Claude also uses screenshots. Date is approximate, taken from the toolset version id browser_toolset_20260801; page shows no date. The same read_page tool is exposed by the Claude in Chrome extension.

Copilot MicrosoftPartial support

partial Microsoft Advertising blog: Optimizing your content for inclusion in AI search answers (Krishna Madhavan, Bing) Published · Checked Bing: H1 used to interpret the page; H2/H3 headings define 'where one idea ends and another begins' when AI parses content. About search-answer parsing rather than the accessibility tree; landmarks not mentioned.

AI browsers
ChatGPT Atlas OpenAISupported (announced)

OpenAI Help Center: Publishers and Developers - FAQ Published · Checked Help Center article shows 'Updated: last month' as of 2026-10-09 (month-only date). Text: 'ChatGPT Atlas uses ARIA tags—the same labels and roles that support screen readers—to interpret page structure and interactive elements.' Explicitly covers 'page structure'.

Gemini in Chrome GooglePartial support

partial Chrome for Developers: Accessibility for agents (Lighthouse Agentic browsing) Published · Checked Chrome team guidance for agents in general ('Agents review the accessibility tree to identify interactive elements'; 'Agents rely on the accessibility tree ... as their primary data model', toolkit blog 2026-06-22). It does not name Gemini in Chrome explicitly, hence partial.

No evidence yet23
  • ChatGPT
  • Gemini
  • Perplexity
  • Meta AI
  • Comet
  • Dia
  • Edge Copilot Mode
  • Claude Code
  • Cursor
  • Codex
  • GitHub Copilot
  • Windsurf
  • CCBot
  • Applebot
  • Amazonbot
  • Bytespider
  • Cohere
  • Mistral
  • DuckAssist
  • Chrome
  • Edge
  • Firefox
  • Safari

Names and marks belong to their owners and are used only to identify products. None of these vendors is affiliated with Sitebulb. Spotted something out of date? Send a correction.

Adoption on the top 100 sites

56 of 100 measured sites implement it

Scanned 23 Oct 2026 with the Sitebulb Labs extension, against the Tranco list (tranco-56Z8N).

9 Oct 2026: 55%12 Oct 2026: 54%23 Oct 2026: 56%
9 Oct 2026 55% → 23 Oct 2026 56%
Sites implementing it56
  1. #1 google.com (with warnings)
  2. #2 cloudflare.com
  3. #3 aws.amazon.com (via amazonaws.com)
  4. #9 microsoft.com
  5. #16 x.com (via twitter.com) (with warnings)
  6. #18 linkedin.com
  7. #24 azure.microsoft.com (via azure.com)
  8. #26 m365.cloud.microsoft (via office.com)
  9. #32 github.com
  10. #33 wikipedia.org
  11. #37 fastly.com (via fastly.net)
  12. #39 whatsapp.com (via whatsapp.net) (with warnings)
  13. #43 netflix.com (with warnings)
  14. #44 triplinkintl.com (with warnings)
  15. #50 marketingplatform.google.com (via doubleclick.net)
  16. #55 wordpress.org
  17. #56 teams.live.com (via skype.com)
  18. #57 cloudflare.net (with warnings)
  19. #60 digicert.com
  20. #68 zoom.com (via zoom.us)
  21. #70 sso.myfritz.net (via myfritz.net) (with warnings)
  22. #71 endpoints.news (with warnings)
  23. #72 api.whatsapp.com (via wa.me) (with warnings)
  24. #75 roblox.com (with warnings)
  25. #76 opera.com
  26. #81 tiktokv.com (with warnings)
  27. #84 nginx.org (with warnings)
  28. #86 msn.com (with warnings)
  29. #90 vimeo.com (with warnings)
  30. #95 mozilla.org
  31. #96 f5.com (with warnings)
  32. #99 store.epicgames.com (via epicgames.com) (with warnings)
  33. #101 wordpress.com
  34. #106 sentry.io
  35. #107 letsencrypt.org (via lencr.org)
  36. #109 ui.com
  37. #114 st.okcdn.ru (via okcdn.ru) (with warnings)
  38. #116 bitly.com (via bit.ly)
  39. #117 ipify.org (with warnings)
  40. #120 samsung.com
  41. #121 office365.com (with warnings)
  42. #123 aliyuncs.com (with warnings)
  43. #128 one.one.one.one (via cloudflare-dns.com)
  44. #130 registrar-servers.com
  45. #131 pages.github.com (via github.io) (with warnings)
  46. #132 dropbox.com
  47. #135 discord.com (via discord.gg)
  48. #136 developer.adobe.com (via adobe.io)
  49. #139 apache.org
  50. #140 android.com
  51. #141 nih.gov
  52. #153 webex.com (with warnings)
  53. #154 snapchat.com
  54. #156 godaddy.com
  55. #158 kaspersky.co.uk (via kaspersky.com)
  56. #159 firebase.google.com (via crashlytics.com)

Overview

8/9

Established

Widely supported and widely deployed.

How this rating is worked out8 of 9 points
FactorPointsWhy
Vendor support2 / 32 vendors supporting, 2 partly or planned
Top-100 adoption3 / 356% of measured sites
Spec maturity3 / 3Published standard

8 or more points is Established, 5 to 7 is Gaining traction, 2 to 4 is Speculative and under 2 is Stalled. A feature is also Stalled when more vendors have said no than yes. Read the spec.

Community insights

No community insights for this feature yet.

Quotes are from public posts, checked against the original before they appear here. Suggest an insight, or ask for yours to be removed.