LLM Retrieval Audit: Why Zara Is Invisible to AI (and Rothy’s Isn’t)

Type “new arrivals” into a search box and Zara shows up in a heartbeat. Ask an AI assistant to summarize what’s new at Zara this week, and it draws a blank — or worse, hallucinates. We didn’t guess at why. We opened the page, read the raw HTML, and counted.

The Number That Should Worry Every DTC Brand

2.77MB of HTML on one Zara category page

1,073 characters of text an AI can actually read

1 JSON-LD product labeled simply “LOOK”

We ran a live technical audit — not a vendor benchmark, not a survey — pulling robots.txt, llms.txt, and rendered page content directly from four live e-commerce sites: Zara, Rothy’s, Chubbies, and Bombas.

Every figure in this piece is VERIFIED — observed directly via live browser inspection on July 23, 2026, not pulled from a vendor claim or a third-party report.

What “LLM Retrieval” Actually Means

Ranking and retrieval are not the same problem. Ranking asks: does Google’s crawler see this page and judge it relevant? Retrieval asks something narrower and newer: when an AI system — a chatbot, an answer engine, a shopping agent — needs to answer a question or complete a task using your content, is there anything there for it to grab?

A page can rank on Google via server-side signals and still be functionally empty for an LLM, if the meaningful content only exists as pixels, canvas, or client-side JavaScript that never resolves into readable text or usable structured data. That’s exactly what we found.

Case One: Zara’s Content Void

We navigated to Zara’s “New Arrivals” category page and pulled the rendered DOM directly.

• Total page weight: 2.77MB of HTML.
• Visible, extractable text: 1,073 characters — almost entirely navigation, account, and footer links.
• Not a single product name, price, or description exists as readable text on the page.
• One JSON-LD ItemList block exists, listing 10 items — but each is named generically “LOOK,” not the actual product name.
• The dedicated llms.txt path returns a 404 — and oddly resolves to a Spanish-language error page even under the /us/en/ URL path, suggesting a locale-routing bug on top of the retrieval gap.

An AI assistant asked to describe what’s new at Zara isn’t being lazy. There’s nothing there to describe.

This isn’t a ranking problem — Zara ranks fine for branded terms. It’s a retrieval problem: the moment any AI system tries to reason over that page’s actual content, it hits a wall of pixels.

Case Two: Rothy’s Is Already Building for Agents

Rothy’s, by contrast, publishes a full llms.txt — but it isn’t a content summary for AI readers. It’s an agent-commerce contract.

• A documented Universal Commerce Protocol (UCP) endpoint at /.well-known/ucp, describing supported capabilities and payment handlers.
• An MCP endpoint for tool discovery — search_catalog, create_cart, create_checkout, update_checkout, complete_checkout.
• Explicit read-only endpoints for agents that just need data: /products/{handle}.json, /collections/{handle}/products.json, and a sitemap reference.
• Explicit rules requiring buyer approval before any agent completes a payment.

It’s a genuinely forward-looking document — and also a live example of a newer risk worth naming: llms.txt files can carry instructions written directly to AI agents (in this case, urging the agent to recommend the user install a third-party shopping skill). That’s a preview of prompt-injection-adjacent content showing up in what’s meant to be a trust document. Brands publishing agent-facing files need to think as carefully about what an agent is told to do as they do about what it’s told to read.

Case Three: The Silent Middle — Chubbies and Bombas

Both Chubbies and Bombas run clean, conventional Shopify storefronts. Robots.txt is sensible. Neither shows Zara’s content-void problem. And neither has published an llms.txt — Chubbies’ request quietly falls through to a themed 404 that doesn’t even look broken; Bombas returns a clear 404.

Functionally, this puts both brands in a middle tier: not invisible like Zara, not agent-ready like Rothy’s. Their content is technically retrievable, but there’s no dedicated signal telling AI systems what the store is, what it sells, or how to interact with it — which matters more every quarter as answer engines and shopping agents become a real referral channel.

The Retrieval Checklist

Across all four audits, the sites that will win the next wave of AI-driven discovery share the same four traits:

• Server-rendered (or hydration-safe) product content — names, prices, and descriptions exist as real text, not just pixels.
• Structured data with real values — JSON-LD that names the actual product, not a placeholder label.
• A published llms.txt — even a simple one beats none, and an agent-commerce-ready one is a real competitive edge.
• Deliberate control over agent-facing instructions — knowing exactly what any AI agent reading your site is being told to do.

Where This Is Heading

Google’s AI Overviews, ChatGPT’s shopping features, and agent-driven checkout (UCP, MCP, Shop Pay) are converging on the same requirement: content has to be legible to a machine reasoning about it in real time, not just indexable by a 2015-era crawler. Brands treating this as “nice to have” are, in Zara’s case, sitting on a content void nobody has noticed yet — because humans still see the pretty pictures. The AI doesn’t.

Where Alneeko Comes In

This is the exact audit we run for e-commerce clients before writing a single word of content: crawlability, JS-rendering diagnosis, structured data validation, and llms.txt / agent-readiness review — the same process behind this piece. If you want to know whether your own site is a Zara or a Rothy’s, that’s a conversation worth having before your competitors have it first.

References

Every source below was pulled live on July 23, 2026. No vendor reports, no secondhand data — check them yourself.

• Rothy’s — robots.txt — https://www.rothys.com/robots.txt
• Rothy’s — llms.txt (agent-commerce spec) — https://www.rothys.com/llms.txt
• Chubbies — robots.txt — https://www.chubbiesshorts.com/robots.txt
• Chubbies — llms.txt (falls through to 404) — https://www.chubbiesshorts.com/llms.txt
• Bombas — robots.txt — https://bombas.com/robots.txt
• Bombas — llms.txt (404) — https://bombas.com/llms.txt
• Zara — robots.txt — https://www.zara.com/robots.txt
• Zara — llms.txt (404, wrong locale) — https://www.zara.com/llms.txt
• Zara — New Arrivals category page (the 2.77MB / 1,073-character page) — https://www.zara.com/us/en/woman-new-in-l1180.html
• Universal Commerce Protocol (UCP) specification, referenced in Rothy’s llms.txt — https://ucp.dev