ACR · Agentic Commerce Readiness
AI shopping agents already choose and pay for their users. The ACR scanner (Agentic Commerce Readiness) reviews your store like one of them and tells you, with evidence, whether it would let it buy or gets stuck trying.
Your store looks great to people. The new question is how it looks to the machines that buy for them.
An agent does not browse your site: it reads it. If it cannot find the price, the availability or a clear path to pay, it does not ask for help: it picks another store. The ACR scanner shows you exactly where you get left out, before it costs you sales.
What we measure
Eight pillars, from how your catalog reads to whether a machine can follow a path to pay. Five set your score, and every check comes with the evidence so you can verify it yourself.
Whether the site lets in an agent that identifies itself, and whether its robots.txt says the same thing it does in practice.
Whether product, price, availability and variants can be read without running JavaScript, and whether the sitemap leads to the products.
Whether the site declares a purchase protocol an agent can discover, such as UCP or ACP.
Whether the return policy and shipping cost travel inside the offer data, not just on a text page.
The platform, the page weight and how much it costs an agent to crawl the site end to end.
The other three · context only
Payments, identity and post-sale are reported but do not add to the score: from the outside there is no way to measure them well. Pretending otherwise would only inflate the number.
The scanner identifies itself and plays fair. It only looks at what anyone can see from outside, and if you deny it access, it respects that and notes it.
Your score maps to a level, to place you at a glance. The uncomfortable part: today not even the best of Mexican retail reaches the top half of the scale.
Ceiling of the study's Mexican sample: 65.7. No measured site reaches Integrable.
A site is scored only if the measured weight passes a threshold. When much goes unmeasured, the range widens: not measuring never raises the score.
The deliverable
The range widens when much goes unmeasured. Not measuring never raises the grade: the honesty is built in.
From Opaque to Agent-native, to place your stage at a glance before diving into the detail.
Each one with its severity, the estimated effort to fix it, and your position against the 14 commerces in the study.
Delivered by hand and never published. Someone from the Lab reviews it before it goes out, so it arrives with context, not as a data dump.
This matters as much as what it does measure. We would rather say it upfront than inflate a number.
No. It only makes requests to public pages, like any visitor. It does not buy, write data or log in.
The first scan is free. Since someone from the Lab reviews each report before it goes out, delivery takes a couple of business days.
It depends who you block. Blocking abusive crawling is reasonable. Blocking specifically the agent that identifies itself while letting the anonymous one through inverts the incentives.
Never. The report is delivered by hand and in private. No name ever appears next to a defect.
Yes. All the evidence is on the public surface and every check declares the command that produces it.
No. That measures whether you are read and cited. The ACR scanner measures whether an agent can complete a purchase without browsing like a person.
We run the same instrument from the study against your store and send you your score with its range, your level, and the findings with evidence and a fix plan. All on the public surface, touching nothing. Someone from the Lab reviews it, so it arrives within a couple of business days.
Free · Public surface · Delivered by hand · Privacy notice