How an agent searches a Shopify catalog
The agent calls search_catalog on your store's UCP endpoint with the shopper's words and gets a ranked list of products. It reads the same fields a shopper-facing summary would: title, description, options, price, tags and a Shopify product category. There are no separate fields for brand, material or barcode, so a fact that isn't in that text can't match a question.
How the list is ranked is Shopify's business and changes. In a large catalog with many near-identical products, such as one product per colour, several listings of the same model compete for the same question.
Why a product misses its own description
A miss on one question is not a verdict. It tells you which words the product doesn't carry.
- The title is a code or a name only. "Aria – Sand" says nothing about what it is. Put the product type in the title.
- The description is thin or all marketing. Material, fit, use and size belong in sentences, not only in images or variant names.
- The type or category is missing. Questions often start with the type, and agents can filter on category.
- Near-duplicates. The same model in ten colours can fill the top results, and the one you meant isn't first.
How we test it
The catalogaudit audit takes your homepage products and writes one natural question for each from its type, options and description, never using the brand or the product's name, for example "breathable ballet flats for women". It runs each through your catalog's agent search, takes the first ten results, and shows whether the product that inspired the question came back, at what position, and what came back instead. Without AI, the question is only the product type plus a gender word from the title.
It is a spot check of the catalog Shopify serves agents. It is not a ranking test of any assistant, and a miss isn't counted against your score.
What to change
Then run the audit again and compare the questions.
- Lead the title with what it is: type, then the feature people search for.
- Write two or three plain sentences: what it is made of, what it is for, what makes it different from the next one.
- Set the product type and the Shopify category.
- For colour variants listed as separate products, make the descriptions differ in more than the colour name.
The checks behind this guide
Agents don't find it by its own name
Searching the store's agent catalog for the product's exact name doesn't return it in the top five, so a shopper asking an agent for it by name may be shown something else.
How to fixMake the title specific (brand, product line, product type) and distinct from similar products. Don't leave the distinguishing details only in images or variant names.
No Shopify product category
Agents can narrow a catalog search by category ("running shoes under $150"). A product without one can't match those searches.
How to fixSet the category in Shopify admin (product → Category) from Shopify's Standard Product Taxonomy. You can bulk-edit it from the products list.
Very thin description (< 150 characters)
Short copy limits which searches the item can match in Shopping and gives agents nothing to quote.
How to fixExpand the description past 150 characters with material, dimensions, use and care — the facts agents quote and feeds match on.
Questions
Does a miss in the shopper questions lower my score?
No. Whether a product tops a free-text search depends on the size of the catalog and on Shopify's ranking, so the questions are shown in the report but not scored. Searching a product by its exact name is scored.
Are the questions the ones real shoppers ask?
They are plausible ones written from your own product data, not measured shopper queries. Treat them as a spot check, not as demand data.
Updated 2026-10-29. catalogaudit isn't affiliated with Shopify or Anthropic.