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

Allbirds scores 59 out of 100, a mixed signal that captures a site doing the easy things well and the hard things inconsistently. The crawler posture is wide open, the payment rails are agent-grade, and Shopify's plumbing means cart APIs and variant data are largely there for the taking. But three of the nine sampled product pages return no structured data at all, the cart page sits behind reCAPTCHA and hCaptcha, and the legal surface that an agent would consult before transacting is effectively missing. The result is a site that an AI shopper can find and parse, then stumbles when it tries to commit. The most expensive failure is CAPTCHA prevalence, a zero out of four. Both reCAPTCHA and hCaptcha are present on the homepage and on the cart page, and the check flags this as transaction-blocking. An agent that successfully parses a Wool Runner product page, picks a size, and proceeds to /cart now hits a human-verification wall on the exact surface where commitment happens. Pair this with the 403 on https://allbirds.com/checkout, which dropped the bot detection score to three out of six, and the picture is of a transactional funnel that punishes non-human traffic at precisely the wrong moment. Structured product data coverage is the second wound, three out of six. Six of the nine sampled URLs carry full Product plus Offer JSON-LD, but three (womens-tree-breezer-dark-navy-knit, mens-wool-runner-mizzles-hazy-indigo, mens-superlight-tree-runners) return nothing: no schema, no labeled attributes, no enumerated variants, only swatch pickers an agent cannot resolve. These same three URLs also fail the specification, conversational description, and decision-support metadata checks. This is not a platform problem; it is a template inconsistency. Some PDPs render the full Shopify schema payload, others render a degraded variant. An agent sampling the catalog will conclude Allbirds is unreliable. Checkout flow structure scored two out of six because the checkout page was unavailable to the classifier without an initialized cart, and guest checkout availability scored two out of four for the same reason. The classifier could not confirm a guest path exists, only that the platform hint suggests one by default. Combined with the CAPTCHA finding, an agent has no way to verify, pre-commitment, that it can complete a purchase without an account. Trust signals leak points across the board. Stated agent policy is a zero because no terms or privacy content was classifiable. Third-party trust anchors is a zero: no Trustpilot, no Yotpo, no Okendo, no aggregateRating markup anywhere in the sample. For a public company with a recognizable brand, the absence of any machine-readable review signal is a self-inflicted wound. Merchant verifiability returns the business name (Allbirds, Inc.) but no physical address and no registration ID, so an agent doing due diligence on the counterparty has only a copyright footer to work with. Sitemap quality (two of four) compounds discoverability: the sitemap at https://allbirds.com/sitemap.xml lists six locs, contains no product URLs, and has no lastmod values. The llms.txt file is a nice touch, but it does not substitute for a product index. The biggest opportunity is removing reCAPTCHA and hCaptcha from the cart surface and exposing a verifiable guest checkout path, because agent traffic that can browse but cannot buy is worth nothing.