What AI Assistants Can (and Can't) Read on 9 Well-Known Shopify Stores

Published 2026-11-05 · ipok.fyi

More shoppers now start product research inside an AI assistant. They ask ChatGPT, Perplexity or Claude for "the best merino sneakers" or "a washable rug that survives a dog", and they get a short answer with a few brand names in it. If you run a Shopify store, you want to know whether those assistants can reach your pages, and whether they understand what's on them once they get there.

To get a rough answer, I ran the same technical check on ten well-known Shopify and DTC brands on November 3, 2026: Allbirds, Bombas, Brooklinen, Chubbies, ColourPop, Fashion Nova, Gymshark, Kylie Cosmetics, MVMT and Ruggable. For each store, the check loaded the homepage and one sampled product page. It looked at the things that decide whether a machine can fetch and interpret a store: robots.txt rules for AI crawlers, /llms.txt, the sitemap, site-level JSON-LD, Product schema, Open Graph tags and whether real content shows up in the raw HTML.

Before the findings, here are the limits. This is a small sample, and it's one product page per store, not a crawl. One store dropped out completely, and one product page result is probably misleading. I'll flag both where they come up. Read this as a snapshot of patterns, not a benchmark.

First, the two data caveats

Bombas returned HTTP 429 (Too Many Requests) to the checker, and to a plain curl request too. Its score of 25 reflects being rate-limited, not the quality of its site, so I left it out. That leaves 9 stores.

Brooklinen's sampled product URL was a content or landing page that lives under /products/ (/products/content-nov-cat-sale-bed), not a real product page. So its "no Product schema" result is probably a false positive. For product-page stats I used the 8 stores whose sampled page was a real product.

The Bombas result is a lesson in its own right. If an aggressive bot-protection rule blocks a checker, it may also block an AI assistant's live fetcher. Rate limiting protects you, but it's worth knowing who it turns away.

What these stores get right

Every store let AI crawlers in. On all 9 stores, robots.txt allowed GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended, Bingbot and Applebot-Extended. Shopify's default robots.txt doesn't block them.

However, none of the 9 named any of these bots explicitly. They were all allowed only through the general User-agent: * group. That works today. But access that depends on a default can change quietly when someone edits robots.txt.liquid, installs an app or switches themes, and nobody will notice.

Most already publish an llms.txt. 7 of 9 served a /llms.txt file (all between roughly 5.4 and 6.2 KB). Gymshark and Ruggable didn't. llms.txt is a proposed convention, not an official standard, and AI systems differ in whether they use it. Still, it was more common in this group than I expected.

The basics are in place. Every store returned HTTP 200 over HTTPS with a valid sitemap. Scores ranged from 76 (Brooklinen, pulled down partly by the false-positive product result) to 94 (Gymshark), with a mean of about 86 out of 100. Big brands get the fundamentals right. The gaps are in the details.

Gap 1: Site-level identity in structured data

Assistants need to work out who a brand is, not just what it sells. Organization and WebSite JSON-LD on the homepage is the most direct way to tell them: name, URL, logo and social profiles.

Takeaway: Add one Organization (or OnlineStore) block and one WebSite block to theme.liquid. Include name, url, logo and sameAs links to your official social profiles. This is a one-time edit.

Gap 2: Product identifiers (GTIN)

All 8 real product pages had Product schema with offers, price and availability. That's the part themes usually handle well. Identifiers were another story:

A GTIN lets any system, whether a search engine, a shopping feed or an AI assistant, match your product to the same item listed elsewhere: reviews, retailer listings, comparison articles. Without one, the system has to guess from the product name.

Takeaway: Fill in the barcode field on your variants in Shopify admin first. Then confirm your theme actually outputs it as gtin12, gtin13 or gtin14 in the Product JSON-LD. Many themes skip it even when the barcode field is filled. If you sell your own products without GTINs, use a stable mpn or sku together with an accurate brand. On that note, ColourPop's sampled product schema had a GTIN but no brand.

Gap 3: Review data that machines can see

Reviews are one of the strongest signals an assistant can pass on ("4.7 stars from 2,000 reviews"). But the data only counts if it's in the markup.

These are brands with plenty of customer reviews. Most likely the review widget loads in the browser and never writes structured data into the page's HTML.

Takeaway: Most review apps have a rich snippets or JSON-LD setting. Turn it on, then confirm the result by viewing the page source, not the rendered page. Search the source for aggregateRating. If it only appears after JavaScript runs, assume many crawlers won't see it.

Gap 4: FAQ content

0 of 9 stores had FAQ markup or detectable FAQ content on the pages checked. This carries the least weight of anything in the check. But buyers ask assistants the same questions they'd ask in a store: sizing, care instructions, shipping times, return windows. Clear, plainly written answers on the product page or a linked FAQ page give an assistant something to quote.

Takeaway: Put your 4–6 most common pre-purchase questions in plain HTML on your product template or a dedicated FAQ page. Treat FAQ schema as a bonus, not the goal.

Gap 5: Social preview tags

3 of 9 had incomplete or missing Open Graph tags. Ruggable had no OG tags detected on the homepage, and Chubbies and Fashion Nova had no og:image. That's mostly a link-preview problem, but it costs nothing to fix and it's another clean description of the page.

A note on raw HTML

All 9 stores passed the raw-HTML check: real text was present in the server response. But the amount varied enormously. Fashion Nova's homepage had about 336 words in raw HTML, against more than 57,000 for Gymshark (though raw HTML word counts can include navigation and script-adjacent text). Assistants that fetch a page live often don't run JavaScript. Whatever matters most, like your positioning, key categories and shipping or returns promises, should be in the HTML the server sends.

A 30-minute checklist for your own store

  1. Open /robots.txt. Make sure no rule blocks the AI user agents you care about. If you want access guaranteed, add explicit Allow groups for them in robots.txt.liquid.
  2. Check your firewall and bot protection. If you use rate limiting or bot challenges, make sure they don't return 403 or 429 to legitimate AI fetchers.
  3. View the source of your homepage. Look for "@type": "Organization" and "@type": "WebSite". Add whichever is missing.
  4. View the source of a real product page. Look for gtin, brand and aggregateRating inside the Product JSON-LD.
  5. Paste a product URL into Google's Rich Results Test to validate the markup.
  6. Add a short /llms.txt: who you are, your main collections, and links to your shipping and returns policies.
  7. Write down your top pre-purchase questions and answer them in plain HTML.
  8. Ask the assistants directly. Ask ChatGPT, Perplexity and Claude, with web search on, the questions your customers would ask. Note whether you're mentioned and which page gets cited.

The bigger picture

The good news from this small sample is that access isn't the problem. Big Shopify brands aren't blocking AI crawlers. The weak points are the details that help a machine understand a store once it gets in: who the brand is, which exact product this is, and what customers think of it. On Shopify, most of these are a settings change or a short theme edit. None of them needs a replatform.

Disclosure: ipok.fyi (unrelated to ipok.io) is my own product: I'm Minh, its solo founder, and it's the tool I used for these checks. Basic checks are free; the live AI test is 1 free per day, then credits. Everything in the checklist above can also be checked by hand.

Run the free GEO / AI visibility check on your store →