Afrikaans
Shqip
العربية
Asturianu
Azərbaycanca
Български
閩南語 / Bân-lâm-gú
বাংলা
Беларуская
Català
Čeština
Cymraeg
Dansk
Deutsch
Eesti
Ελληνικά
English
Español
Esperanto
Euskara
فارسی
Français
Galego
한국어
Hausa
Հայերեն
हिन्दी
Hrvatski
Bahasa Indonesia
Italiano
עברית
ქართული
Ladin
Latina
Latviešu
Lietuvių
Magyar
```
Shown with 11 long blocks (scripts, styles, long tags) folded into a marker that says how many characters they hide; the pane stops after the first 200 words; the page is 120 KB; 20 spans of the page text shown as \[…\] because the words or characters in them are rejected by this site's own style rules.
**Reading view** the text a plain-fetch client can extract
title: Wikipedia
Wikipedia The Free Encyclopedia
Wikipedia The Free Encyclopedia English 7,254,000+ articles 日本語 1,523,000+ 記事 Deutsch 3.159.000+ Artikel Français 2 785 000+ articles Русский 2 123 000+ статей Español 2.144.000+ artículos Italiano 1.991.000+ voci 中文 1,561,000+ 条目 / 條目 Português 1.185.000+ artigos Polski 1 711 000+ haseł Search Wikipedia Afrikaans Shqip العربية Asturianu Azərbaycanca Български 閩南語 / Bân-lâm-gú বাংলা Беларуская Català Čeština Cymraeg Dansk Deutsch Eesti Ελληνικά English Español Esperanto Euskara فارسی Français Galego 한국어 Hausa Հայերեն हिन्दी Hrvatski Bahasa Indonesia Italiano עברית ქართული Ladin Latina Latviešu Lietuvių Magyar Македонски Malagasy मराठी مصرى Bahasa Melayu Bahaso Minangkabau မြန်မာဘာသာ Nederlands 日本語 Norsk (bokmål) Norsk (nynorsk) Нохчийн Oʻzbekcha / Ўзбекча Polski Português Қазақша / Qazaqşa / قازاقشا Română Simple English Sinugboanong Binisaya Slovenčina Slovenščina Српски / Srpski Srpskohrvatski / Српскохрватски Suomi Svenska Kiswahili தமிழ் Татарча / Tatarça తెలుగు ภาษาไทย Тоҷикӣ تۆرکجه Türkçe Українська اردو Tiếng Việt Winaray 中文 Русский 粵語 Search Read Wikipedia in your language 1,000,000+ articles العربية Deutsch English Español فارسی Français Italiano مصرى Nederlands 日本語 Polski Português Sinugboanong Binisaya Svenska Українська Tiếng Việt Winaray 中文 Русский 100,000+ articles Afrikaans Shqip Asturianu Azərbaycanca Български 閩南語 / Bân-lâm-gú বাংলা Беларуская Català Čeština Cymraeg Dansk Eesti Ελληνικά Esperanto Euskara Galego 한국어 Hausa Հայերեն हिन्दी Hrvatski Bahasa Indonesia
*1100 more words follow in the initial HTML.*
**3 found**, 3 missing, 1 note, from 7 checks run
- H1Found
- 200 wordsFound
- llms.txtMissing
- MarkdownMissing
- Missing pathFound
- JSON-LDMissing
- OpenAPINote
| Check | Result | Detail |
| --- | --- | --- |
| H1 in the initial HTML | Found one H1 | First H1: Wikipedia The Free Encyclopedia |
| First 200 words in the initial HTML | Found 200 of 200 words | The initial HTML carries 1300 words of visible text; the first 200 are marked in the left pane. |
| llms.txt present and linked | Missing absent (404) | /llms.txt returned 404. Absence is not a failure on its own: Ora's lab found it changed nothing for answers already in the HTML. |
| Markdown on request | Missing 200 text/html | Accept: text/markdown returned 200 with text/html. The server ignores the header. |
| Status of a made-up path | Found 404 | /agent-ready-html-missing-path-check returned 404: a real status code for a missing page. |
| JSON-LD types | Missing none | No JSON-LD block on the home page. Ora's lab found it made no difference when the answer was already in the HTML; Google says it is not needed for AI features. |
| OpenAPI link | Note none on the home page | No OpenAPI link on the home page. It matters only for a site that ships an API. |
## Run the full check
The viewer read one page with four requests. The full check we recommend next is Ora's scan. Its Access layer runs 42 checks worth 30 of its 100 points, among them content without JavaScript, bot detection, llms.txt, markdown on request, JSON-LD and the sitemap (Ora's check catalogue, contract 1.38.0, read 5 November 2026). [Open Ora's report for wikipedia.org](https://ora.ai/score/wikipedia.org?utm_source=agentreadyhtml.com&utm_medium=referral&utm_campaign=viewer).
Limits: Ora reads public URLs only, says two runs of one scan land within a couple of points without publishing repeat-scan data, and keeps the product that fixes what it finds behind a waitlist. Cloudflare's scanner at [isitagentready.com](https://isitagentready.com/) is a quick free alternative, and [Mintlify's score](https://www.mintlify.com/score) covers documentation.
## What the viewer does
It sends four plain GET requests with the user agent `Mozilla/5.0 (compatible; agent-readiness-crawl/1.0)`: the URL itself, the same URL with `Accept: text/markdown`, `/llms.txt` on the same host, and a made-up path at the root. It runs no JavaScript. The left pane is the first bytes of the HTML as text, with long scripts and styles folded into a marker that says how many characters they hide. The H1 and the first 200 words of visible text are marked in both panes. The flags under the panes say what the four responses added up to.
## Limits
- Ten seconds for all four requests together, 1.5 MB read from the home page, at most five redirects, each one checked again.
- http and https only, default ports only. Private, local and internal addresses and names are refused.
- Ten runs per address per minute. The counter lives in memory on the edge function, so it is per instance rather than global. Past ten you get a 429 with a Retry-After header.
- Results are cached for 24 hours by the response headers. Nothing is written to a database: the viewer keeps the first 200 words and the extracted fields for the length of the request and keeps no page body beyond that.
## What it does not tell you
A plain fetch from this server is not a real agent. Bot protection weighs the source address more than the user agent, so a page that answers 200 here can still refuse an agent, and the reverse. Ora's lab found (2026-10-19) that returning 403 to agent user agents and hiding the answer in JavaScript were the only two of 19 configurations that stopped retrieval; the viewer can see the second, and can only hint at the first. The [initial HTML reference](https://agentreadyhtml.com/initial-html/) covers both. The server logs and the published crawler address ranges settle the first.
## The seven flags
H1 in the initial HTML
Whether the HTML the server sends already contains an H1. It is the cheapest sign that the page does not depend on scripts to say what it is.
First 200 words in the initial HTML
Whether at least 200 words of visible text arrive in the initial HTML. Under 40 words with an app root element is flagged as a JavaScript shell.
llms.txt present and linked
Whether /llms.txt answers with a plain-text file, whether the home page links it (an anchor, a link element or a Link header), and the structure grade the index uses.
Markdown on request
Whether a request with Accept: text/markdown comes back as text/markdown. Cloudflare Radar found 3.9 percent of scanned top sites passing this at its April 2026 launch.
Status of a made-up path
Whether a path that should not exist returns 404 or 410. A 200 is a soft 404 and reads as a live page.
JSON-LD types
The schema.org types found in JSON-LD blocks on the page. Ora's lab found JSON-LD made no difference when the answer was already in the HTML.
OpenAPI link
Whether the page links an OpenAPI file, by an anchor or a service-desc link relation. It matters only for a site that ships an API.
## Sources
1. [Ora: AgentReady launch post (Ora's lab findings)](https://ora.ai/blog/agentready-launch?utm_source=agentreadyhtml.com&utm_medium=referral&utm_campaign=tool) https://ora.ai/blog/agentready-launch, read 4 November 2026
2. [Ora check catalogue, contract 1.38.0 (132 checks, 42 in the Access layer)](https://ora.ai/api/checks?utm_source=agentreadyhtml.com&utm_medium=referral&utm_campaign=tool) https://ora.ai/api/checks, read 5 November 2026
3. [Ora home page (the fixing product is behind a waitlist)](https://ora.ai/?utm_source=agentreadyhtml.com&utm_medium=referral&utm_campaign=tool) https://ora.ai/, read 5 November 2026
4. [Ora: Introducing Deep Scan (repeat scans)](https://ora.ai/blog/introducing-deep-scan?utm_source=agentreadyhtml.com&utm_medium=referral&utm_campaign=tool) https://ora.ai/blog/introducing-deep-scan, read 5 November 2026
5. [Ora llms.txt (public URLs only)](https://ora.ai/llms.txt?utm_source=agentreadyhtml.com&utm_medium=referral&utm_campaign=tool) https://ora.ai/llms.txt, read 5 November 2026
6. [Vercel: Make your site readable by AI agents](https://vercel.com/kb/guide/make-your-site-readable-by-ai-agents) https://vercel.com/kb/guide/make-your-site-readable-by-ai-agents, read 4 November 2026
7. [llms.txt proposal (llmstxt.org)](https://llmstxt.org/) https://llmstxt.org/, read 4 November 2026
Last reviewed 5 November 2026
Edition
Crawl
2026-11-04
Ora reads
2026-11-05
Page updated
2026-11-05
Method
How the viewer reads a page, how a file is graded and how a platform is scored: [the method page](https://agentreadyhtml.com/method/).
---
---
title: "What agents reach on a site: two Ora studies: Agent Ready HTML"
description: "Homepage and docs first, llms.txt and openapi.json later and mostly by link: Ora's studies of June and October 2026 side by side, with their limits."
canonical: https://agentreadyhtml.com/what-agents-reach/
site: "Agent Ready HTML"
last_updated: 2026-11-05
date: 2026-11-05
section: "Evidence"
---
# What agents reach on a site: two Ora studies: Agent Ready HTML
> Homepage and docs first, llms.txt and openapi.json later and mostly by link: Ora's studies of June and October 2026 side by side, with their limits.
Source: https://agentreadyhtml.com/what-agents-reach/
Site: Agent Ready HTML
Last reviewed: 2026-11-05
# What agents reach on a site: two Ora studies side by side
In short
Ora's June study found agents used docs pages 88 percent of the time and the homepage 84 percent when they existed, against 46 percent for llms.txt and 5 percent for llms-full.txt. Its October study found 86 percent of llms.txt fetches arrived through a link. Both are vendor results with different units.
The check we recommend
Ora's scan. Its Access layer runs 42 checks worth 30 of its 100 points, from content without JavaScript to markdown on request. Every check is listed at [ora.ai/api/checks](https://ora.ai/api/checks?utm_source=agentreadyhtml.com&utm_medium=referral&utm_campaign=reference) with its tier, and 25 of the 132 with a spec link. Reads need no key and no sign-up (Ora's pricing page, read 5 November 2026).
Ora reads public URLs only, says two runs of one scan land within a couple of points without publishing repeat-scan data, and keeps the product that fixes what it finds behind a waitlist.
Two of Ora's published studies describe where agents go on a site. They measure different things and should not be added together, but they agree on the order: the pages a person would open first come before the files made for machines, and the machine files are reached mostly through links. Every figure here is Ora's, taken from the posts named under each chart and labelled as a vendor result.
## When a resource exists, how often agents use it
Ora's research post of 16 June 2026 drew on episodes of agents landing on websites and trying to get something done. For each resource it reports the usage rate once the resource existed on a site.
Share of sites with the resource where agents used it, Ora's traced agent episodes, published 2026-06-16.
Text equivalent of this chart
| Resource | Value (%) | Counts |
| --- | --- | --- |
| Docs pages | 88 | |
| Homepage | 84 | |
| llms.txt | 46 | |
| .well-known files | 34 | |
| openapi.json | 17 | |
| robots.txt and sitemap.xml | 9 | |
| agents.md | 8 | |
| llms-full.txt | 5 | |
## How many runs reach each surface
The AgentReady launch post of 19 October 2026 reports the share of traced runs that reached each surface at least once. The homepage figure is lower than in June because the unit changed: a share of runs, not a rate among sites that have the resource.
Share of Ora's traced runs that reached each surface at least once, published 2026-10-19.
Text equivalent of this chart
| Surface | Value (%) | Counts |
| --- | --- | --- |
| Docs pages | 83 | |
| Homepage | 69 | |
| .well-known files | 23 | |
| openapi.json (sites that ship one) | 21 | |
| sitemap.xml | 4 | |
## Links beat guessed paths
When agents did open a file, they mostly got there by following a link from a page they had already read. A file that nothing links to is reached late or not at all, which is why this site links its own llms.txt from every page.
Share of fetches of each file that followed a link from a page the agent had already read, published 2026-10-19.
Text equivalent of this chart
| File | Value (%) | Counts |
| --- | --- | --- |
| openapi.json | 97 | |
| .well-known files | 93 | |
| llms.txt | 86 | |
## The order of arrival
In the June episodes the homepage arrived on turn 1.0 on average, docs on 1.7, llms.txt and openapi.json on 3.0, agents.md on 4.6 and llms-full.txt on 6.7. The homepage was reached in about 84 percent of runs, almost always on the first turn (about 95 percent). Whatever the homepage's initial HTML leaves out, the run starts without. The [viewer](https://agentreadyhtml.com/viewer/) shows what it carries and the [guide to checking it like an agent](https://agentreadyhtml.com/guides/check-initial-html-like-an-agent/) measures it.
## The same files, reached or not
Ora reports that on the one site in its set that shipped the full stack and linked it, the same files were used far more: openapi.json rose from 17 to 72 percent, .well-known files from 34 to 63 percent and agents.md from 8 to 42 percent. That site is ora.ai itself, so the result is a single vendor case, but the direction matches the link figures above. Ora also found that a reachable page with stale or wrong content hurts more than no page, because the agent trusts it.
## What this site draws from it
Put the answer in the homepage HTML, link llms.txt and any OpenAPI file from pages agents already read, and serve the conventional paths agents guess. For a whole-site check against those behaviours, the check we recommend is [Ora's scan](https://ora.ai/?utm_source=agentreadyhtml.com&utm_medium=referral&utm_campaign=what-agents-reach), which is also the source of these studies. Ora's limits are printed beside the recommendation.
Cite this: Agent Ready HTML. What agents reach on a site. Edition 2026-11-04. https://agentreadyhtml.com/what-agents-reach/
## Sources
1. [Ora research: What agents actually reach (2026-06-16)](https://ora.ai/blog/what-agents-actually-reach?utm_source=agentreadyhtml.com&utm_medium=referral&utm_campaign=reference) https://ora.ai/blog/what-agents-actually-reach, read 5 November 2026
2. [Ora: AgentReady launch post (Ora's lab findings, 2026-10-19)](https://ora.ai/blog/agentready-launch?utm_source=agentreadyhtml.com&utm_medium=referral&utm_campaign=reference) https://ora.ai/blog/agentready-launch, read 5 November 2026
3. [Ora home page (the studies are listed on it)](https://ora.ai/?utm_source=agentreadyhtml.com&utm_medium=referral&utm_campaign=reference) https://ora.ai/, read 5 November 2026
Last reviewed 5 November 2026
Edition
Crawl
2026-11-04
Ora reads
2026-11-05
Page updated
2026-11-05
Method
How the viewer reads a page, how a file is graded and how a platform is scored: [the method page](https://agentreadyhtml.com/method/).
---
