An automated buyer reads markup. It finds a product page, or it does not. It reads a price, a currency and whether the item is in stock, or it reads nothing. This tells you which — from your own storefront, page by page, with the evidence beside every answer.
Two audits a month on the free plan — two, because one number is a snapshot and this product is about what changed.
A shop that turned our reader away at every page told us nothing about itself. A shop we read fully, that carries no product markup, told us a great deal. Only the second is a finding. The first comes back as "could not check" and scores nothing at all — publishing our own refusal as your failing would be the easiest lie this product could tell.
Agent-commerce standards are being written right now. A tool that grades you against them today is grading you against its own guess. So that category is not scored here — it is listed as not assessed, with the reason, and it will stay that way until an adapter arrives carrying a real specification and version.
The journey opens a product, adds to the cart, opens the cart and confirms the checkout entry exists. It never submits an order, never pays, never asks for an SMS code and never creates an account — and a selector that looks like it would is refused, including one you type into settings yourself.
Discovery starts with robots.txt, follows it to your sitemaps, and samples up to a dozen pages — product pages first. This is a sample, and it is described as one: a catalogue of fifty thousand items has the same markup problem on page one as on page fifty thousand, and crawling all of it would cost you money to learn the same thing.
Name, price, currency, availability — without these an agent cannot proceed at all. SKU, brand, image, description — without these it cannot tell your product from a similar one. GTIN, condition, price validity, seller, shipping — helpful, and reported at the bottom of the scale. Twenty products missing the same field is one finding carrying the count, not twenty findings burying the rest.
If a product page carries no product in the markup it served, we fetch it again through a real browser and look a second time. If the product appears then, the answer is that your data is client-rendered — reported plainly, at the middle of the scale, because plenty of excellent stores render client-side and simply cost an agent a browser.
When a page states a price twice, in two vocabularies, an agent may read the one your shoppers never see. That comparison is made only where the pairing is unambiguous — one product and one offer on the page — because lining up two lists by position would invent a disagreement between prices belonging to different items.
