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

Peloton scores a 55, which lands squarely in mixed-signal territory: the fundamentals an agent needs to find and parse the catalog are mostly in place, but the moment a machine tries to actually transact, evaluate trust, or estimate delivery, the site goes dark. This is a company that has done the easy structured-data work and then stopped, leaving meaningful points on the table in exactly the dimensions that determine whether an AI agent will recommend, let alone purchase, a Peloton product. The biggest single failure is the checkout stack, where the transactability dimension hemorrhages eight points across guest checkout, price transparency, and flow structure. Both the cart and checkout pages render only "You need to enable JavaScript to run this app" to a non-executing fetcher, which means agents cannot observe step count, cannot confirm guest availability, and cannot see shipping or tax preauth. The classifier flagged low confidence on all three checks for the same reason. Shop Pay, Apple Pay, and Google Pay are detected on product and homepage, and the Shopify cart endpoint at `/cart/add` is exposed, so the rails exist; the problem is that everything upstream of those rails is invisible until JavaScript executes against an initialized cart. For a brand selling a $1,500-plus hardware decision, opacity at the checkout step is the wrong place to be opaque. Trust signals are the second wound, with five of ten points lost. There are zero third-party review anchors detected across Trustpilot, BBB, Yotpo, Bazaarvoice, Google Shopping, and eight other platforms; the return policy at `onepeloton.com/returns` returned no parseable content to the classifier; and merchant verifiability captured the business name "Peloton" but no physical address or registration ID. Peloton is a public company with a real headquarters and a real returns policy. None of that is reaching the agent layer, which means an LLM weighing Peloton against a competitor with Yotpo reviews and a structured policy page has no signal to lean on. Sitemap quality is a clean zero out of four because `onepeloton.com/sitemap.xml` returns a 404. This is a fixable afternoon's work and there is no defensible reason for a catalog this large to ship without one. Compounding it, the agent-specific endpoints check finds no `llms.txt` and no `ai-plugin.json`, costing two more points in discoverability even though the robots.txt cheerfully welcomes GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and six other crawlers. Inviting the agents in and then giving them no map is a strange posture. The agent-native content gaps round out the story: delivery date transparency scores zero (no structured ship dates, only generic "Orders over $100 receive free standard shipping" text), price stability scores zero because not a single offer carries a `priceValidUntil`, and decision-support metadata loses two points despite strong size guides on apparel because the dumbbells and the Cody book pages strip out compatibility and use-case framing. The bones of good copy are there, particularly on the Bike PDP with its 4'11"-to-6'5" height range and 297 lb weight cap, but it is unevenly applied. The single highest-leverage fix is making the checkout flow legible to non-JavaScript fetchers, because that one change unlocks ten points across three checks and converts Peloton from a catalog an agent can read into a store an agent can use.