README.md

metadata

license: cc-by-4.0
pretty_name: AI-search readiness of 9,973 top sites (October 2026)
size_categories:
  - 1K<n<10K
language:
  - en
tags:
  - seo
  - structured-data
  - schema-org
  - llms-txt
  - sitemap
  - robots-txt
  - web-crawling
configs:
  - config_name: domains
    data_files: domains.csv
    default: true
  - config_name: figures
    data_files: figures.csv

AI-search readiness of 9,973 top sites, October 2026

What the home pages and public files of 9,973 top sites say about themselves: whether the organization is described in structured data (Organization and sameAs), what other structured data is there (FAQ and the rest), whether a valid sitemap and an llms.txt exist, and whether the main text is in the HTML as served or needs scripts. Tranco list N26XW, read 2026-10-24 (October 2026 edition). Every figure is a count of files read by fixed rules. There is no model and no score.

Page, method and figures: https://queryrev.com/research/ai-search-readiness

Files

• domains.csv
• figures.csv
• figures.json
• types.csv

Panel

The first 10,000 places of Tranco list N26XW (created 2026-10-14), the same panel as the crawler study, with the places that are not a valid domain skipped and a fixed list of domains of AI model companies left out. 9,973 domains remain. Sites whose robots.txt closes the whole site to QueryRevBot by name are counted as opted out and are not in domains.csv.

Method in short

• Reader: the QueryRev crawler as QueryRevBot, with its network guard (public addresses, ports 80 and 443, every redirect checked). Requests to one host are at least a second apart, and a host that answers 429 or 503 is left alone.
• Per domain, in order: robots.txt; the home page when robots.txt allows it (redirects only within the registrable domain, 2 MiB after decompression); the llms.txt; the sitemap (the first Sitemap line on the same domain, else /sitemap.xml, 5 MiB after decompression). No inner pages, no scripts run, no cookies.
• Main text: the text of main when present, else of the body without header, navigation, footer and aside; script, style, noscript, template and svg are never counted. Text in the HTML means 150 words or more.
• Organization types: Organization and its subtypes in schema.org version 30.1. A sameAs address counts when it is an absolute http(s) address not on the site’s own domain. Profile kinds by host: wikidata (wikidata.org); wikipedia (wikipedia.org); linkedin (linkedin.com); x (x.com, twitter.com); facebook (facebook.com, fb.com); instagram (instagram.com); youtube (youtube.com, youtu.be); tiktok (tiktok.com); github (github.com); crunchbase (crunchbase.com); pinterest (pinterest.com), other.
• A home page is not read when it is a bot check (a 202, or a small page showing an interstitial message) or a placeholder (a default server page or a parking page: under 150 words and one of the marks listed below).
• A valid sitemap is an XML urlset or sitemapindex with at least one address on the site’s domain, or a text sitemap; an HTML page at that address is not a sitemap. The share of valid sitemaps is shown on the sites that were asked for one (robots.txt did not keep QueryRevBot from it).

Placeholder marks

A home page with under 150 words whose HTML starts with one of these patterns (regular expressions, any case, matched against the first 64 KiB) is counted as a placeholder and not read:

• nginx_default: welcome to nginx (any case)
• apache_default: apache2? (?:\w+ )?default page|apache http server test page|test page for the apache (any case)
• apache_it_works: <h1>\s*it works!?\s*<\/h1> (any case)
• iis_default: iis windows server|internet information services|<title>\s*iis7?\b (any case)
• caddy_default: caddy works! (any case)
• panel_default: web server's default page|default web site page|domain default page|plesk default page|cpanel[^<]{0,40}default (any case)
• hosting_default: future home of something quite cool|this site can.t be reached yet|website coming\s+soon.{0,40}(?:namecheap|godaddy) (any case)
• parked_domain: this domain (?:is|may be) (?:for sale|parked|available)|domain (?:name )?(?:is )?for sale|buy this domain|is parked (?:free|courtesy)|parkingcrew|sedoparking|sedo\.com\/search|hugedomains|dan\.com\/buy-domain|afternic\.com\/forsale|bodis\.com|domainsponsor|above\.com\/parked (any case)

Columns of domains.csv

Column	Meaning
rank	Place in the Tranco list.
domain	Registrable domain of the panel.
host	Host robots.txt was read on: the domain, or www. plus the domain when the domain did not connect.
final_host	Host the home page was read on, after redirects.
robots_txt	file (a robots.txt was read) or missing (a 4xx answer: no restrictions). Blank when not read.
robots_status	HTTP status of the robots.txt answer.
failure	Why robots.txt could not be read (blank when it was). Nothing else is asked of such a site.
sitemap_lines	Number of Sitemap: lines in robots.txt.
home_outcome	read, or the reason the home page was not read: not_allowed, landing_robots_unread (the page landed on another host whose robots.txt could not be read), home_truncated_unmeasured (the HTML went over 2 MiB and was cut, with under 150 words of main text in the part read: the page is left out of every home-page figure, because the cut may have hidden the text), forbidden, challenge, placeholder, not_html, error_status, off_site, redirect_error, network.
home_status	HTTP status of the home page answer.
redirects	Redirects followed to reach the home page.
home_bytes	Size of the home page HTML read, in bytes.
home_truncated	yes when the HTML went over 2 MiB and was read as far as the cut (home_outcome read, or home_truncated_unmeasured).
noindex	yes when the home page says noindex in a robots meta tag or an X-Robots-Tag header (an agent-specific rule does not count).
main_words	Words of main text in the HTML as served (a lower bound when counting stopped at its limit).
content	text_in_html (150 words or more), little_text, or app_shell (an empty application shell).
app_root	yes when an application root element is present.
js_notice	yes when the page asks the reader to enable JavaScript.
json_ld_blocks	ld+json scripts read (at most 50).
json_ld_schema_org	ld+json scripts that parse and whose top-level @context names schema.org.
json_ld_errors	ld+json scripts that do not parse.
types	schema.org types found in JSON-LD and microdata, pipe-separated, at most 20 per page (the first 20 in page order); a name that is not a type of the pinned vocabulary is published as other. Type counts recomputed from this column can therefore be a little lower than the counts in types.csv and in the structured-data-types rows of figures.csv, which count every type on a page.
organization	json_ld, microdata or none: where an Organization (or a subtype) is described.
organization_type	The Organization type of the primary node (a schema.org type, else other).
organization_name_url	yes when the primary node has a non-empty name and an absolute http(s) url.
organization_logo	yes when the primary node has a logo.
same_as_count	Distinct sameAs addresses of the primary node that are not on the site’s own domain. The addresses are not published.
same_as_kinds	Kinds of profile linked in sameAs, pipe-separated, by host.
website	yes when a WebSite node is present.
search_action	yes when a WebSite node has a SearchAction.
faq_page	yes when a FAQPage node is present (JSON-LD or microdata).
sitemap_source	robots (the first Sitemap: line on the site’s domain) or well_known (/sitemap.xml).
sitemap_outcome	valid, not_found, not_a_sitemap, malformed, not_allowed or failed.
sitemap_kind	urlset, index or text for a valid sitemap.
sitemap_locs	Addresses listed in a valid sitemap (up to 100,000 counted).
sitemap_truncated	yes when the sitemap went over 5 MiB after decompression and is valid as far as read.
llms_txt	found, absent, or not_fetched (robots.txt keeps QueryRevBot from /llms.txt).
llms_txt_summary	yes when a found llms.txt has a summary line.
llms_txt_link_sections	Sections of a found llms.txt that list links.
ai_search_open	yes, partly or no: whether robots.txt lets every covered search and user crawler read the root.
rendered_words	Words of main text once scripts ran, for the pages of the render sample; blank elsewhere.

figures.csv has one row per figure, item and segment (the whole panel, two rank bands, country-code domains with at least 50 domains in the panel), with the count, the denominators and the shares. types.csv lists every schema.org type seen and its count of home pages.

Crawlers covered

Crawlers covered: the 11 AI crawlers on this list, from OpenAI, Perplexity, Google, Apple, Common Crawl, Meta, Amazon and ByteDance. Other crawlers are not checked.

Limits

• One page per site, the home page, as served to our crawler from one place at one time; an Organization described on an inner page is not seen.
• Structured data added by scripts is not in the HTML as served and is counted as absent; the render sample shows how often.
• Valid means the rules in the method, not a search engine’s rich-result eligibility or a full schema.org validation.
• A site that refuses our crawler or serves a bot check is not read; it may treat other crawlers differently.
• One sitemap file is read; the pages it lists and the children of an index are not checked. A sitemap named on another host of the site’s domain is asked under the site’s own robots.txt; the robots.txt of that other host is not read.
• Main text includes what a page puts inside main; a page without main is counted without its header, navigation and footer, and some boilerplate may remain.
• The panel includes domains that serve no website (networks, APIs, CDNs); they are counted as not read.
• A home page that went over 2 MiB and was cut, with under 150 words of main text in the part read, is counted apart (home_truncated_unmeasured) and is in no home-page figure: the cut may have hidden its text and its structured data. A page cut with 150 words or more stays in, flagged home_truncated.
• The types column of domains.csv keeps at most 20 types per page, so type counts recomputed from it can be a little lower than the counts in types.csv and in the structured-data-types rows of figures.csv, which count every type on a page.
• An llms.txt that redirects to another registrable domain is not followed (it is counted as absent). The first edition of the crawler study followed such redirects, which explains a part of the gap to its llms.txt figure.
• A row of a rank band or a country-code domain is published only when every denominator its shares are shown on holds at least 30 sites; the quartiles of profiles per site go with it.
• 1,004 sites whose first try ended in a network failure (no answer, a timeout, a refused connection) were asked again from the start on 2026-10-24 from a connection in US, with the same limits and politeness; the better of the two tries is kept for each (294 read more the second time). A name that did not resolve, a refusal and a bot check were not asked again.

Cite

QueryRev Research (2026). AI-search readiness of top sites, October 2026 edition. https://queryrev.com/research/ai-search-readiness. CC BY 4.0