Mixed signal. Agents can reach the page, but guess from there.

Graza scores 53 out of 100, a mixed signal that reflects a brand built for humans browsing a beautiful site rather than agents transacting on behalf of them. The discoverability layer is excellent, the catalog markup is clean, and the payment rails are agent friendly. But the moment an agent tries to confirm what a delivery will cost, when it will arrive, or whether the merchant is who it claims to be, the page goes quiet. This is the profile of a Shopify store that did the easy structured data work and stopped. The largest single hole is product attributes. Every one of the nine sampled product pages returned zero structured attributes and zero labeled attributes, a flat fail on check 2.2 worth six points. Sizzle and Drizzle have rich human readable copy about Picual olives, harvest months, and 13lbs to 1 liter yield ratios, but none of it is exposed as machine readable specs. An agent comparing finishing oils across merchants cannot extract volume, varietal, or origin without scraping prose. That is six points left on the table for what is essentially a templating decision. Checkout transparency is the second wound. Check 3.3 fails outright: shipping and tax are invisible pre auth, with the cart rendering a literal "${{amount}}" template placeholder where totals should appear. Check 3.1 only scored two of six because the checkout route at https://graza.co/checkout returned 403 without an initialized cart, so the flow could not even be evaluated end to end. Combined with the CAPTCHA finding on 4.2, where reCAPTCHA and hCaptcha were detected on both the cart and homepage, an agent attempting a purchase faces a black box guarded by challenges. Machine native rails like Shop Pay are present, which is exactly why the friction elsewhere is so frustrating: the infrastructure is there, the surface is hostile. Trust signals underperform for a brand of Graza's stature. Check 5.4 returns zero across eleven checked review platforms including Trustpilot, Yotpo, Judge.me, and Okendo, despite the product pages themselves citing 4,436 reviews at 4.9 stars. Those reviews exist somewhere on the page but are not surfaced through any aggregator schema or third party anchor an agent would recognize. Check 5.2 confirms the business name as Graza but finds no physical address and no registration ID. For a perishable food product where check 5.1 already confirms returns are not accepted, verifiability matters more, not less. Smaller but telling: check 6.4 fails completely on delivery dates, with zero of nine pages exposing structured or unstructured ETA signals. Check 1.4 leaves two points on the floor for the absence of llms.txt or an ai-plugin.json manifest. Check 6.6 flags Subscribe and Save as the default pricing display next to One Time Purchase, a mild dark pattern that an agent acting for a user will either misread or have to explicitly override. Check 1.3 scores two of four because the sitemap at https://graza.co/sitemap.xml lists only four URLs, has no product URLs, and carries no lastmod timestamps. The biggest opportunity is fixing checkout transparency and product attribute markup, which together represent ten lost points and convert Graza from a site agents can read into a site agents can actually buy from.