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

Rhode scores 62 out of 100, a mixed signal that undersells how good the foundation is and how strange the gaps are. The site does the hard part well: every AI crawler in the tracked set (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot, anthropic-ai, Applebot-Extended, Bytespider, Amazonbot) is allowed through robots.txt, an llms.txt file exists at the root, all nine sampled products carry Product plus Offer JSON-LD, and Shop Pay sits alongside Apple Pay and PayPal at the cart. The problem is that the checkout and cart layer, the exact surface an agent needs to actually transact, is hostile in ways that appear unintentional.

The most expensive finding is the checkout flow itself, which lost four points. The classifier could not render the checkout page at all, returning "The checkout page could not be accessed or rendered, indicating the checkout flow requires an initialized cart to proceed." Compounding this, the guest checkout check scored zero of four: the flow appears to gate behind account creation, with the classifier capturing "Login now to activate early access. Email Password sign in ... Don't have an account? Sign up here." A merchant on Shopify defaulting to a login wall on a public storefront is a choice, and for an agent it is a wall.

Next, CAPTCHA prevalence is a hard zero of four. Both reCAPTCHA and hCaptcha are present on the cart and homepage, and the check flags transaction_blocking: true. Stacking two CAPTCHA vendors on the cart page is unusual, and it is the single most agent-hostile signal on the site. Paired with the one 403 observed at https://rhodeskin.com/checkout, the transaction path reads as defended against exactly the kind of automated buyer that Shop Pay, sitting on the same page, is designed to accommodate.

Variant handling and inventory signals bleed points quietly. Every one of the nine sampled products has variants (Shopify product JSON exposes up to 52 variant IDs on the Pocket Blush Set), but zero are enumerated in a way the check recognizes, capping the score at one of one on a partial band. Inventory scored two of five because schemaHasShipping is false across every product and none carry structured ETAs, only prose like "free EU shipping on orders over €70." That same absence tanks delivery date transparency to zero of two: four products have unstructured speed hints, none have a structured shippingDetails node.

Smaller but telling: product attribute coverage sits at three of six because two of nine products (the original Peptide Lip Treatment and the Toast lip tint) return empty structuredHits arrays entirely, while their siblings expose five attributes each. This is inconsistency inside a single catalog, not a platform limit. Stated agent policy scored zero of two because neither terms nor privacy content was classifiable, and merchant verifiability lost two points for the absence of a physical address or registration ID beyond the name "rhode."

On the credit side, the returns policy is fully machine-readable with a 30-day window quoted verbatim, third-party trust anchors surface an aggregate rating with 17,794 reviews, and the cart API is directly addressable via /cart/add. The manipulation check caught only a mild "WHILE SUPPLIES LAST" banner, which is minor.

The biggest opportunity is unmistakable: remove the login gate and dual-CAPTCHA stack from the cart and checkout, because everything else Rhode has built for agents dies at that door.