Can an AI agent actually buy from your store?

Meta's Muse does not shop from a connector list. It browses the open web with its own browser, from a cloud machine, and it reads your real pages. If your edge challenges automated traffic, or your price only appears after JavaScript runs, or your checkout demands an account, the agent stops and tells its customer your site is not working.

We run a scored audit of your storefront the way an agent sees it, and hand you a report with the five fixes worth the most. It is a one-hour job for us, your developer can act on it without us, and it is the cheapest way to find out where the money is leaking.

Email us your store URL

What we check

Six areas, 100 points. Every check is scored pass, warn, fail or skipped, and skipped checks come out of the denominator rather than counting against you.

Reachability and bot posture, 25 points

Why it matters to an agent. Muse drives a browser from a datacentre IP range. Blanket bot rules, aggressive rate limits, mandatory challenges and hard geo blocks all read that as an attack. The agent does not see a challenge and get creative. It gives up and moves on. We request your homepage twice, once as a normal desktop browser and once as an obvious bot, and we parse your robots.txt against thirteen agent tokens.

Agent discovery files, 15 points

Why it matters to an agent. These are the files written for the agent rather than for a crawler. /llms.txt is a few hundred bytes that tell an agent what you sell, where the catalogue is and what your policies are, instead of it rendering fifty pages to guess. /.well-known/ucp is the Universal Commerce Protocol business profile, and it is the only confirmed machine-discovery endpoint in agentic commerce today.

The Shopify split. If you are on Shopify, you already publish the discovery profile: it appears to be served platform-wide, and you did nothing to get it. If you are not on Shopify, you almost certainly do not publish it, and adding it is half a day of work. That single difference is the most common reason two similar stores score twelve points apart.

Structured product data, 20 points

Why it matters to an agent. Before it can buy, the agent has to be certain what the thing is, what it costs and whether it is in stock. A price rendered by JavaScript into a div is a guess. JSON-LD Product with an Offer carrying price, currency and availability is a fact. An agent that has to guess picks the wrong variant or recommends a competitor whose data was legible. Missing availability is the single most common gap, and it is the one that makes an agent order something you do not have.

Checkout signals, 20 points

Why it matters to an agent. Stripe is the rail Muse rides. At a merchant where Stripe Link is live the agent pays in one step and the customer approves the total in chat. Everywhere else Link mints a single-use virtual card and the agent has to drive your checkout form by hand, where every extra field is a chance to drop the order. We also check that a cart URL is reachable and that guest checkout exists, because an account wall is a hard stop: the agent has no password, and it cannot go and fetch an email verification code.

Policy pages readable without a login, 12 points

Why it matters to an agent. Returns and shipping are the questions a customer asks before they let an assistant spend their money. An agent that cannot find the policy either says so out loud, which kills the sale, or summarises something wrong, which creates a support ticket. We check returns, shipping, contact, privacy and terms all return a real page without a login and without JavaScript.

Performance basics, 8 points

Why it matters to an agent. A cloud browser pays the latency on every navigation and the agent pays context for every byte. We check time to first byte, homepage weight and the number of external scripts, each of which is another thing that can hang or trip a bot detector.

What you get

• A scored report out of 100, with the grade band and the denominator printed, so the number is reproducible from the table rather than being our opinion.
• The top five fixes, ranked by the points they cost you, written in plain language with the evidence we actually saw attached underneath. Ranking by points means they are automatically ordered by how much they matter.
• The raw machine-readable results, so you can re-run the comparison yourself later.
• A 30-minute walkthrough call on the standard tier and above.

What it costs

Tier	Price	What you get
Single store	$500	One audit, one product page, the report, and an emailed summary.
Standard	$1,000	Up to five product pages across categories, a benchmark against two named rivals, and the walkthrough call.
Multi-brand or enterprise	$2,000	Up to five storefronts or locales, a re-audit after 30 days to show movement, and a written remediation plan with owners and effort estimates.

A quarterly re-audit is half price. The specs move most weeks right now, so the second one is usually worth more than the first.

If you want us to do the fixes

The report is written so your own developer can act on it without us. That is deliberate: it has to be obviously worth the money before we pitch anything else. If you would rather we did the work, we quote it per item.

• JSON-LD Product and Offer across the catalogue, generated from your existing product data.
• An /llms.txt and a /.well-known/ucp profile, generated from your catalogue.
• An agentic commerce catalogue feed, wired to a regular inventory and price cadence.
• Bot-management rule changes that let shopping agents read product pages while keeping the protections on login, checkout and account endpoints.
• Turning Stripe Link on and surfacing it before the sign-in step.

Guest checkout, llms.txt and a readable returns page are usually same-day fixes, quoted as a flat package. Structured data and bot rules are real projects and are quoted separately.

What we will not claim

• Agent traffic is small today. Muse is United States only and weeks old, and agent-driven orders are a tiny share of anyone's revenue. The argument is that these fixes are cheap, they are all things a good storefront should do anyway, and you are buying a position rather than this quarter's revenue.
• We do not run your JavaScript. Everything comes from the server-rendered HTML of a first request. A store that renders product data client-side will score lower, which is usually a real finding, because most shopping agents do not run it either.
• We cannot prove Stripe Link is on. It resolves server side after the payment element mounts. A miss is reported as "not visible in static HTML", never as "Link is off".
• We never defeat a bot challenge. When we see one we record it and move on.
• One page per store on the single-store tier. A store can be excellent on the page we sampled and poor everywhere else.

Get in touch

Email support@noerrands.com

Put your store URL in the subject line, replacing the example. In the body, one product page URL is enough to get started. Tell us your platform if you know it, and whether anyone owns your payments configuration, because the Stripe Link item is usually a conversation rather than a project.

We come back with the score and the worst finding, usually the same day.

Want the agent to be able to do more than buy from you? See get listed in Meta Muse, where we build the connector itself.