State of AI agent readiness 2026

Can AI agents use the web's sites?

AI agent readiness is whether an agent acting for a person can find what a site offers, read it without a browser and act on it: an llms.txt that says when to use the site, Markdown when an agent asks for it, labelled forms and, for products with an API, a catalog an agent can discover. It is scored across up to 25 pages per site, not the homepage alone.

Headline numbers

Score distributionAgents score, 0 to 100, one per site.

p10
52
p25
57
Median
66.5
p75
71
p90
76
95 or more
1%
0 to under 5: 0 sites5 to under 10: 0 sites10 to under 15: 0 sites15 to under 20: 0 sites20 to under 25: 0 sites25 to under 30: 0 sites30 to under 35: 0 sites35 to under 40: 0 sites40 to under 45: 4 sites45 to under 50: 9 sites50 to under 55: 23 sites55 to under 60: 27 sites60 to under 65: 27 sites65 to under 70: 54 sites70 to under 75: 38 sites75 to under 80: 21 sites80 to under 85: 6 sites85 to under 90: 1 site90 to under 95: 1 site95 to 100: 3 sites54 sites
Sites per 5-point bandMedian 66.5
View values
Agents score distribution
ScoreSites
40 to 44.94
45 to 49.99
50 to 54.923
55 to 59.927
60 to 64.927
65 to 69.954
70 to 74.938
75 to 79.921
80 to 84.96
85 to 89.91
90 to 94.91
95 to 1003

By kind of siteMedian and spread per category and region.

Category

  • SaaS and docs · 59SaaS and docs: median 62 · middle half 56 to 74 · p10 52, p90 80 · 59 sites62
  • Ecommerce · 37Ecommerce: median 61 · middle half 52 to 67 · p10 47.6, p90 69.4 · 37 sites61
  • Media and blogs · 50Media and blogs: median 66.5 · middle half 62.3 to 70 · p10 56, p90 72 · 50 sites66.5
  • Education · 41Education: median 68 · middle half 59 to 73 · p10 51, p90 75 · 41 sites68
  • Local business · 27Local business: median 70 · middle half 66 to 75 · p10 64, p90 77.4 · 27 sites70
MedianMiddle half (p25 to p75)p10 to p90
View values
AI agent readiness scores by category
GroupSitesp10p25Medianp75p90
SaaS and docs595256627480
Ecommerce3747.652616769.4
Media and blogs505662.366.57072
Education415159687375
Local business276466707577.4

Region

  • Global · 162Global: median 66 · middle half 56 to 71 · p10 51.1, p90 76.9 · 162 sites66
  • Brazil · 28Brazil: median 67 · middle half 60.8 to 70 · p10 55.7, p90 72.3 · 28 sites67
  • Germany · 24Germany: median 68.5 · middle half 59 to 71.5 · p10 54, p90 74 · 24 sites68.5
MedianMiddle half (p25 to p75)p10 to p90
View values
AI agent readiness scores by region
GroupSitesp10p25Medianp75p90
Global16251.156667176.9
Brazil2855.760.8677072.3
Germany24545968.571.574

Most common failuresShare of sites where the check applied and failed or partly failed.

  1. Missing pages answer 404 with a Markdown bodyEssential · agents/act/agent-friendly-404193 of 201 sites fail or partly fail96% · 193 of 201
  2. API responses tell agents their rate limitRecommended · agents/integrate/rate-limit-headers36 of 38 sites fail or partly fail95% · 36 of 38
  3. Agents are told when to use the siteRecommended · agents/discover/agent-instruction179 of 189 sites fail or partly fail95% · 179 of 189
  4. The API has a versioning and deprecation policyRecommended · agents/integrate/api-versioning-policy34 of 36 sites fail or partly fail94% · 34 of 36
  5. The homepage negotiates Markdown with Vary: AcceptEssential · agents/read/markdown-negotiation-vary196 of 212 sites fail or partly fail92% · 196 of 212
  6. Every API operation has an id and a descriptionRecommended · agents/integrate/api-schema-analysis57 of 63 sites fail or partly fail90% · 57 of 63
  7. API errors follow a typed error modelRecommended · agents/integrate/api-error-model29 of 33 sites fail or partly fail88% · 29 of 33
  8. API operations map cleanly to agent toolsRecommended · agents/integrate/function-calling-compat55 of 63 sites fail or partly fail87% · 55 of 63

What to fix first

  • Publish an llms.txt that says what the site is for and when an agent should use it.
  • Serve Markdown when a request sends Accept: text/markdown.
  • Label every form field, so an agent knows what to type where.
  • List public APIs in /.well-known/api-catalog.

Method

Corpus
214 public sites scanned, of 299 picked
Source list
Tranco top 1M, list N26XW (14 Oct 2026), hand-sorted into categories
Dates
16 Oct 2026
Scanner
ParadigmaBot, robots.txt respected, at most 2 requests at a time per site
Pages
Up to 25 per site, homepage included, 21.9 read on average
Checked
The 76 scored checks of the AI agent readiness checker, same engine and version for every site
Not checked
Signed-in API or MCP calls, JavaScript rendering, real agent sessions, logged-in pages
Not scored
22 checks, shown in the tool only
Left out
Refused or challenged (401, 403, 429, 503): 53 · robots.txt unreadable to our crawler (5xx, 429, timeout or a redirect we did not follow): 24 · Other HTTP error on the homepage: 3 · No answer (timeout, DNS or TLS): 2 · robots.txt disallows our crawler: 2 · Redirects to another domain: 1
Published
Counts and distributions only, no site named

Where does your site land?The same checks, free, with evidence for every result.

Run the AI agent readiness checker