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

Warby Parker scores 26/97 and lands in the Invisible band because DataDome treats agent traffic as hostile: 63% of scanner requests were blocked or challenged, including every product page and the sitemap itself. The site has the right paperwork on the front door (open robots.txt, llms.txt, ai-plugin.json) but the actual store is unreachable, so none of the downstream signals (schema, price, inventory, returns) can be evaluated by an agent. This is the classic pattern of a brand that published AI-friendly manifests while its edge security silently negates them.

Where the points went

The evidence

• Bot detection, 0/6. DataDome challenge pages served for warbyparker.com/, /cart, /checkout, /terms, /privacy, /return-policy, and /.well-known/ai-plugin.json. All 9 sampled PDPs returned 403, and even the sitemap at warbyparker.com/sitemap.xml was blocked.
• Structured data, 0/6 and 0/8 on price. Zero of 9 products expose Product or Offer schema; price only surfaces in the rendered DOM as data-event-copy="Premium eyewear, starting at $95", which is marketing chrome, not a per-SKU offer.
• Return policy, 0/3. The classifier hit a 404 error page at /return-policy; there is no machine-readable window, conditions, or process for an agent to cite.
• Checkout, 2/6 and guest 0/4. Cart shows "Your cart is empty" and checkout renders only header nav until a cart is initialized; no "continue as guest" affordance was detected.
• Specs and descriptions, 0/4 and 0/2. Only 3 of 9 PDPs surfaced high-signal specs (Caldwell, Anselm, Dupree); the rest returned empty or "Page is down for scheduled maintenance."

What already works

• Crawler posture, 6/6. robots.txt explicitly allows ChatGPT-User, OAI-SearchBot, Claude-User, Claude-SearchBot, PerplexityBot, Amazonbot, and Meta-ExternalFetcher.
• Agent endpoints, 4/4. Both llms.txt and .well-known/ai-plugin.json are published.
• Payment and cart rails, 6/6. Apple Pay present on cart, checkout, and homepage; Shopify /cart endpoint is discoverable.

Fix first

1. Allowlist the declared AI user-agents in DataDome so the 63% block rate collapses (+10 and unlocks ~25 dependent points).
2. Emit Product + Offer JSON-LD with price, availability, and SKU in server-rendered HTML (+14).
3. Restore /return-policy and mark it up with window, conditions, and process (+3).

The single biggest lever is the DataDome allowlist: until agents can actually load a PDP, every other investment in schema and content is invisible.