Barely understood 0–49
soup-stock-tokyo.com is barely understood by AI shopping assistants.
AI assistants can reach soup-stock-tokyo.com and read the page perfectly well — the copy, the images, the prices on screen. What is missing is the layer underneath: almost nothing states, in a form an assistant can parse, what each product actually is, what it is made of, who it suits, or why a shopper should pick it over the next one. So an assistant can describe the shop in general terms, but has no verified detail to recommend a specific product on — and it recommends the store that gave it that detail instead. Soup Stock Tokyo is an online store that sells a variety of soups, with main product categories including art-inspired soups (such as Vermeer's and Gogh's onion soup), baby food and kids' sets, and seasonal gifts. It is designed for individual consumers looking for quality meals, families seeking child-friendly food options, and shoppers looking for gifts.
Score by category
Six weighted categories, one deterministic rubric — grouped by what each one actually changes. Open any row for what it checks and the exact fixes.
| Category | Points | Earned | Status | Lost | Details |
|---|---|---|---|---|---|
| Can AI reach you 20 / 20 The gate. Nothing below counts until a crawler can fetch the page. | |||||
| Crawler accessrobots.txt, llms.txt, and five declared AI crawler user-agents | 20 / 20 | Good | 0 | ||
What this checks
Before anything else can be read, an AI crawler has to be allowed in. We check robots.txt and llms.txt for the five declared AI user-agents — GPTBot, ClaudeBot, PerplexityBot, Google-Extended and Bingbot — and confirm each one can actually fetch a page rather than just being permitted on paper. It is worth 20 points, and every other category depends on it. Fixes for this category
Nothing to fix here — this category is at full marks. | |||||
| Can AI understand what you sell 19 / 60 The richness an assistant reads to name a product, its price and what it is for — the part that decides whether you get recommended. | |||||
| Structured dataProduct, Offer, and Breadcrumb JSON-LD across product and collection pages | 0 / 30 | Missing | −30 | ||
What this checks
Structured data is the machine-readable copy of a product page — name, price, availability, brand — in a format assistants parse directly rather than guessing at. Without it an assistant has to infer your catalog from prose, and usually declines to recommend anything it cannot verify. This is the heaviest category in the rubric at 30 points. Fixes for this category
+30Add missing product structured data5 of 5 pages we checked have no product structured data at all — without it, AI shopping agents can't tell what you're selling.
| |||||
| Titles & metaA unique title and meta description per page, at usable length and shape | 11 / 20 | Missing | −9 | ||
What this checks
The title and meta description are what an assistant quotes back to a shopper, so they have to be unique per page and land in a usable length — long enough to say what the product is, short enough not to be truncated. Worth 20 points, the joint-second heaviest category. Fixes for this category
+9Tighten your titles and meta descriptions5 of 5 pages have a title or meta description outside the ideal length — that's the summary AI assistants pull into search results and chat answers.
| |||||
| Content depthDescription length, attribute coverage, and specificity per product | 8 / 10 | Missing | −2 | ||
What this checks
A one-line product description gives an assistant nothing to recommend on. We look at description length, how many real attributes are covered, and how specific the language is — the difference between a product that can be matched to a shopper's question and one that cannot. Worth 10 points. Fixes for this category
Nothing to fix here — this category is at full marks. | |||||
| Search & sharing 7 / 20 How the page behaves in search results and when a link is shared. Real value, but it does not change what an assistant can say about a product. | |||||
| Social & canonicalCanonical URL, Open Graph, and Twitter card tags | 6 / 10 | Missing | −4 | ||
What this checks
Canonical URLs tell an assistant which version of a page is the real one when several resolve to the same product, and Open Graph and Twitter card tags decide how a link renders when someone shares it. Together they are worth 10 points. Fixes for this category
Nothing to fix here — this category is at full marks. | |||||
| Image alt textDescriptive alt attributes on product imagery | 1 / 10 | Missing | −9 | ||
What this checks
An assistant that cannot see a photograph still needs to know what is in it. Descriptive alt text on product imagery is what makes an image legible to a text model, and it is the same attribute a screen reader announces. Worth 10 points. Fixes for this category
+9Add alt text to your product imagesOnly 6% of your product images have alt text — descriptive alt text helps AI assistants understand what's in each photo.
| |||||
| Total | 46 / 100 | Barely understood | −54 | ||
Change since the last scan
Can AI crawlers reach the store?
5 allowedWeakest pages
Every missing point above has a fix inside Shopify.
CortexGuard writes the structured data, the titles, the meta, and the alt text from your existing catalog — then queues all of it for your review. You approve every change before it publishes.
See CortexGuard →A snapshot, not a promise
We fetch only what is publicly there — the same pages, the same declared bot user-agents, an AI crawler would see. No login, no admin access, nothing private.
Every score comes from the same deterministic 0–100 rubric across six categories. Same inputs, same score — nothing is randomized and nothing is weighted by who is paying.
A score reflects what we found the moment we crawled. Stores change constantly, so re-run a scan whenever you want a current read.