Buying is moving into agents: a person asks an assistant to buy something, and the assistant
picks from what it can read. We ask your site the sixteen questions an agent asks it, and write the answers
down. The questions are free. The document is 99 dollars.
Free, no card, no account. Your place out of ten and the first three answers appear below.
?
What an agent needs from you
Find you at all: a plain request, without a browser, a cookie or a script that has to run first.
Understand what you sell, in terms a machine can compare instead of a design it has to guess at.
Read your price and your stock, so it can put your product into a basket it is allowed to fill.
Be told who may authorise a purchase, because an agent acting for a person stops when nothing says how it may act.
Where sites actually sit
We have measured 986 sites. The three questions an agent needs answered to buy, the reach, the catalogue and the price, passed together at 0.8 percent of them. So the honest answer for almost every site is: not on the radar yet, and the reasons are small and fixable. The reading puts a site in one of three bands: invisible to agents, listed but not buyable, and buyable. Our own site sits in the middle band, and we publish it that way.
The reasons repeat. Nothing at /agents.txt, so a machine has no description of what it may ask for. No /.well-known/ai-catalog.json, so an agent has no list of what you offer. A missing path answering 503 where a real 404 belongs, which a careful client reads as a broken door.
An agent is not obliged to read your site, and no assistant promises to. llms.txt in particular is read by no system we know of, so the document marks which of the sixteen are declarations and which of them change what a machine can actually do. You pay for the second kind.
What is in the document
Your place on the radar out of ten, and the band it falls in: invisible, listed, or buyable.
Sixteen questions answered one by one: what we asked, the answer that came back, what that answer costs you in sales, the change that fixes it, and the command that shows the fix landed.
The list of what one request cannot see, named instead of guessed: the end of a checkout, a webhook, taxes and shipping, the right to sell a product, and every step a person has to confirm.
The fixer pack built for your site: the findings file and one recipe per question, and each recipe names who does that piece, you or your developer.
If you would rather not touch it: 399 dollars for one site and one round of work, and the 99 comes off that, so the report ends up costing nothing.
We ask your site the way an agent asks it: a plain request, no browser, no cookie.
We read the answer as an answer. An empty 200, a 405 or a challenge header is read as a refusal, not as success.
We put each answer beside what the standard asks for, one question at a time.
You get the numbers with the evidence behind each, and the count they come from: 986 sites measured.
What this is not
This is not a Google ranking, not a security scan, and not a list of keywords. It answers one question: can an agent find your site, understand what you sell, read the price, and be told that it may buy.
What you buy
The radar reading for one site: your place out of ten, the sixteen questions with the answers we got, what one request cannot see, and the fixer pack built for your site. If you want us to do the repairs, they are 399 dollars for one site and one round, and this 99 comes off that.
Price: reading the tillonce, for one site
No account. The document is sent to this address and can be fetched again from the same
address or from the payment receipt. Not a security scan. It says what a machine can read and act on now.