agentsiteaudit · Research

The Agent Readiness Report · October 2026

Three in four businesses selling to AI agents don't tell agents who they are.

We checked the websites of the businesses listed in the Coinbase x402 Bazaar, the main catalog of services that AI agents can pay for. These are the businesses most likely to be ready for AI agents. We wanted to know how ready their own websites are.

71%of all catalog listings come from a single seller, whose domain does not point to any website at all.
43%of sellers run on free hosting or a bare IP address, with no website of their own.
41%of seller websites have an llms.txt guide for agents.
25%of seller websites say clearly, in a machine-readable way, who the business is.

The catalog

On 30 October 2026 the Bazaar listed 100,028 paid resources. They come from just 2,300 sellers. The ten largest sellers account for 78% of all listings. 978 sellers (43%) run on free hosting platforms or bare IP addresses, which leaves 1,322 sellers with a website of their own. Those are the websites we audited.

Solid websites

On classic quality, these sites do well. Across 1,168 websites that could be scored, the median score was 86 out of 100. The score covers security, SEO, the certificate, speed, the connection, accessibility, links, mobile and domain set-up.

A401
B502
C257
D8
F0

The most common classic problems were:

Agent readiness, check by check

Agent readiness is a separate set of eight checks. Each one is shown on its own. "Not applicable" means the check does not apply to that site, for example a site that publishes no API description. Those sites are not counted as failing.

PassWarningFailNot applicableNot tested

llms.txt guide41% pass

A plain-text guide at /llms.txt telling agents what the site offers. Pass rate counts only the 1181 sites where the check applies.

479 pass · 702 warning · 0 fail · 0 not applicable · 9 not tested

llms.txt links work87% pass

The links inside llms.txt open, so agents are not sent to dead pages. Pass rate counts only the 477 sites where the check applies.

413 pass · 64 warning · 0 fail · 698 not applicable · 15 not tested

AI crawlers allowed91% pass

robots.txt does not shut out the main AI agents and crawlers. Pass rate counts only the 1139 sites where the check applies.

1040 pass · 99 warning · 0 fail · 0 not applicable · 51 not tested

Says who the business is25% pass

The homepage gives a name plus a way to contact or locate the business, readable by machines. Pass rate counts only the 1168 sites where the check applies.

289 pass · 879 warning · 0 fail · 0 not applicable · 22 not tested

Prices readable74% pass

Where prices are shown, they are written as text an agent can read. Pass rate counts only the 362 sites where the check applies.

268 pass · 94 warning · 0 fail · 806 not applicable · 22 not tested

API description valid94% pass

Any published OpenAPI description is valid and readable. Pass rate counts only the 546 sites where the check applies.

514 pass · 32 warning · 0 fail · 550 not applicable · 94 not tested

x402 discovery file87% pass

A /.well-known/x402 file lists the paid resources in a readable form. Pass rate counts only the 381 sites where the check applies.

331 pass · 50 warning · 0 fail · 668 not applicable · 141 not tested

x402 price tags work69% pass

The listed paid resources answer with a valid HTTP 402 price tag. Pass rate counts only the 290 sites where the check applies.

200 pass · 80 warning · 10 fail · 668 not applicable · 232 not tested

What to fix first

  1. Say who you are. Put the business name and a way to contact or locate it on the homepage, and add it as Organization structured data.
  2. Publish an llms.txt guide at /llms.txt, and check that every link in it opens.
  3. Check robots.txt. Make sure it does not block the AI agents you want to reach you.
  4. If you sell to agents, describe it. Publish an OpenAPI description and a /.well-known/x402 file, and make sure every listed resource returns a valid price tag.
  5. Write prices as text, not only in images or scripts, so an agent can read them.

How we did it

Published 30 October 2026 by AgentSiteAudit. Questions or corrections: derek@agentsiteaudit.com.