Invisible. Agents can't meaningfully engage with this storefront.

Nike scored 32 out of 100, which lands in the Invisible band, and the diagnosis is almost darkly comic: a company that spends billions on brand visibility to humans is functionally opaque to the agents increasingly mediating commerce discovery. The crawler doors are wide open, the bot posture is welcoming, and yet an AI agent arriving at nike.com cannot reliably see a product, a price, or a return policy. Nike has invited the guests in and forgotten to set the table.

The single most damaging finding is that the scanner could not extract a usable sample of product URLs. Nike declares seven sitemaps in robots.txt, including sitemap-v2-pdp-index.xml, which should be the product detail page index, but the sitemap that actually resolved (sitemap-us-help.xml) contains 113 help center locations with a most recent lastmod of February 6, 2019, nearly seven years stale. That cascading failure zeroed out every product-dependent check: structured product data (0/6), price in raw HTML (0/8), product attributes (0/6), inventory signals (0/5), variant handling (0/1), specification completeness (0/4), conversational descriptions (0/2), decision-support metadata (0/3), delivery dates (0/2), and price stability (0/2). That is roughly 41 points evaporated because the crawlable sitemap surface points at a help archive frozen in the first Trump administration rather than the current PDP catalog. Whether the PDP sitemaps are gated, misnamed, or served conditionally, the observable reality is that agents get help articles instead of shoes.

The transactability layer compounds the problem. The checkout page returns only header and footer chrome unless a cart is already initialized client-side, which the classifier flagged with a medium-confidence note that no form fields, step indicators, or payment inputs render in the HTML. This produced a 2/6 on checkout structure and a 0/4 on guest checkout, because there is simply nothing there for an agent to reason about. Price transparency likewise scored 0/4: shipping and tax are invisible pre-authentication. The one bright spot is a detected Shopify-style /cart endpoint (3/3 on cart API accessibility), but a cart API without a machine-readable checkout is a door that opens onto a wall.

Trust signals are the third quiet disaster. The return policy at nike.com/returns rendered as navigation chrome with no policy body, scoring 0/3. Merchant verifiability caught only the "Nike, Inc." copyright string with no physical address or registration ID (1/3). Third-party trust anchors scored 0/2 across eleven checked platforms including Trustpilot and Google Shopping, and there is no stated agent policy in the privacy documentation (0/2). Nike's brand equity carries this weight with humans; agents evaluate what they can parse, and they can parse almost none of it.

The genuinely strong results (6/6 on crawler posture with GPTBot, ClaudeBot, PerplexityBot, and six others explicitly allowed, 6/6 on bot detection with a 0% block rate across 13 fetches, 4/4 on CAPTCHA, 3/3 on rate limiting, 2/2 on absence of manipulation tactics) prove the infrastructure welcomes agents. The content layer just does not speak to them.

The biggest opportunity is trivial to state and non-trivial to execute: publish a working, fresh PDP sitemap and render product pricing, attributes, and inventory in server-side HTML, because every other fix compounds on top of agents actually being able to see a product.