How the checker works

What it reads

Agents start with the HTML your server sends, and so does the checker. It doesn't run JavaScript, log in, fill in forms or add anything to a cart. For each check it asks for, in this order:

• /robots.txt, so it knows what it's allowed to read.
• Your homepage, to see where the store lives. If your domain sends visitors to www, to https or to another address on the same company domain (kowtowclothing.com to nz.kowtowclothing.com, say), it follows. If it sends them to a different domain, it stops and tells you which address to check instead.
• Your sitemaps, to find product pages.
• Up to ten product pages.
• /.well-known/ucp, where a checkout profile lives, and /llms.txt.

On each product page it reads your structured data first (JSON-LD or microdata), because that's what agents rely on. It also notes what the page itself shows, like the main heading and the visible price.

A check is capped at 30 requests and 60 seconds. The crawler page has the full limits and how to block it.

How it picks the ten products

Ten pages can't cover a whole catalogue, so the checker works to make them a fair sample.

• Sitemap first. It reads the sitemaps your robots.txt lists, plus /sitemap.xml, starting with product sitemaps. Sitemaps of brands, categories or collections count as lists of products, not products. Once it has about 100 likely product pages it stops reading, and when a big store splits its products over many sitemaps it reads the first, middle and last.
• Links if there's no sitemap. It follows links from your homepage and up to three collection or category pages.
• Product pages only. Addresses shaped like /products/…, /product/…, /item/… or /p/… come first. Blog posts, guides, policies, cart and account pages are skipped, including copies under country prefixes like /en-au/. If no pattern stands out, it opens one page from each large section and lets the product data decide.
• Spread out, not A to Z. Candidates are put in a fixed shuffled order, so a big store isn't judged on products starting with “a”. Colour or size versions published as separate products share one slot. The same store gets the same sample each time, so a recheck after fixes compares like with like.
• Your own links as a fallback. If it can't find any product pages, you can give it up to three.

What the verdicts mean

Each verdict is Yes, Partly or Not yet. If something stopped the check, like a robots.txt it couldn't read or a site that didn't answer, you'll see Couldn't check instead.

Can agents find you?

This looks at three things: whether your robots.txt lets the search crawlers of Google, OpenAI, Anthropic and Perplexity, and Claude-User, reach your product pages, whether those pages tell search tools not to index them, and whether a sitemap or links lead to your products.

• Yes: all five crawlers can reach your products, none of the pages say noindex, and your sitemap lists them.
• Partly: some crawlers are blocked, some pages say noindex, or the only route to your products is through links.
• Not yet: robots.txt blocks every search crawler, every page says noindex, or nothing leads to a single product page.

Claude-User counts too: it fetches pages when someone asks Claude, and it follows robots.txt, so blocking it keeps your products out of those answers. AI training crawlers like GPTBot and ClaudeBot, and on-request fetchers like ChatGPT-User that may ignore robots.txt, are listed in the report but don't affect the verdict. Blocking training is a fair choice.

Can they understand your products?

This reads the details an agent needs to describe what you sell and compare it with the alternatives.

• The basics: name, description, photo, price with a currency code, and stock.
• The extras: brand, a barcode or part number (GTIN or MPN; your own SKU on its own doesn't count), size and colour labels when a product comes in versions, shipping and returns.

• Yes: the basics and the extras on most products, meaning at least 8 in every 10 we read. A rating helps but isn't needed.
• Partly: what you sell comes through, but some details are missing on too many products.
• Not yet: your product pages publish no product data, the data on most of them is broken, or three or more basics are missing on at least half of them.

If you tell us you sell by quote or trade account, a missing public price isn't held against you.

Can they buy?

An agent can only buy for someone, or send them to pay, when it knows the price and whether the product is in stock. Some can go further and check out through a checkout profile your store publishes.

• Yes: clear prices and stock on most products, plus a valid checkout profile (UCP) at /.well-known/ucp that declares checkout. The profile is checked against the official UCP schema (version 2026-08-25). No AI agent can check out on a New Zealand store yet, so stores there top out at Partly.
• Partly: prices and stock are clear but there's no valid checkout profile, or there's a profile but prices or stock aren't clear.
• Not yet: most products don't show a clear price and stock.

If you sell by quote or trade account, Not yet is expected, and the report shows how to make the route to a quote or an account obvious. If you only want agents to find you, this verdict can wait.

What it can't see

• Anything that only appears after JavaScript runs.
• Prices behind a login, or prices that change by location. Each check runs once, from one place, and the report says where.
• Whether your stock levels are right, or whether checkout works end to end. It never places an order.
• Product feeds you send to Google Merchant Center or AI shopping programmes, and ChatGPT's own checkout programme. None of these shows from outside your store.
• How any particular AI ranks or recommends you.
• Pages your robots.txt or bot protection keeps it out of. These show up in the report as not checked.
• The rest of your catalogue beyond the ten pages it reads.

The fixes

Each verdict comes with the one fix that matters most, ranked by how many of the products we read it would improve. If your store runs on Shopify, WooCommerce or BigCommerce, the fix is written in that platform's terms. Otherwise it's written for whoever edits your site's templates.

What we keep

Reports are kept for seven days. A check of the same store from the last 24 hours may be reused, so asking again that day usually costs your site nothing.

Your answers to the questions under the report stay in your browser. They change how the report reads your situation, like not counting a hidden price against a quote-only store, but never what the scan saw. The help link fills in the contact form with your store, verdicts and answers, and nothing is sent until you press send.