ShelfProbe
Product data QA for Shopify Agentic Storefronts

See what shopping agents
find, miss, and misunderstand.

ShelfProbe queries your Shopify Catalog as a real UCP agent, compares every answer with your Admin product data, and backs each finding with the query, the response and the rule that flagged it.
Open consoleSee how it works
HighPIPELINE_PRICE_MISMATCHSample data

Price differs between your Admin data and the Catalog response

Affected query

“merino wool hoodie under $100”

Admin value

price
89.00 USD
availableForSale
true

Catalog response

price
109.00 USD
availableForSale
true

Rule

The Catalog price must match the lowest sellable variant price in Admin.

Suggested fix

Check the price list and currency settings for the target market, then re-test.

Why it matters

Agents only sell what they can read

Shopping agents in ChatGPT, Google AI Mode, Gemini, Copilot and Shop discover products through Shopify Catalog. Gaps between your Admin data and what the Catalog returns cost you sales you never see.
  • Products agents cannot find
    Real purchase-intent queries that should match a product return nothing, or return something else.
  • Attributes agents misread
    Titles, categories, prices, stock and variants that differ between your Admin data and the Catalog response.
  • Products agents cannot buy
    Published products without a sellable variant or a valid purchase link in the Catalog.
How it works

From probe to proven fix

Every scan follows the same loop, so each change you make can be measured against the last one.
  • Read your source data
    Products, variants, prices, stock, categories and alt text through the Admin API, read-only.
  • Probe the Catalog as an agent
    Exact-name, category, use-case, constraint and variant queries sent to your Storefront Catalog with a public UCP agent profile.
  • Compare and explain
    Deterministic rules compare the source values with the Catalog response and attach the evidence to every finding.
  • Fix and re-test
    Product updates trigger a targeted re-test, so the report shows what was fixed, what is new and what persists.

Scan #14 vs #13

Sample data
New
2
Fixed
5
Persisting
3

  • Trail running shoe

    CATALOG_VARIANT_UNRESOLVED

    Fixed

  • Down vest

    SOURCE_CATEGORY_MISSING

    New

  • Canvas tote bag

    SOURCE_IMAGE_ALT_MISSING

    Persisting
Principles

Evidence first, promises never

ShelfProbe measures data quality, query hits and purchasability. It does not promise rankings on Shopify or any AI channel.
Real requests, not guesses
Findings come from actual UCP Catalog responses, not from a model predicting what an agent might see.
Evidence on every finding
Each finding shows the rule, the Admin value, the Catalog value, the affected queries and a suggested fix.
Deterministic rules
Completeness, consistency and purchasability are judged by rules you can inspect, not by a score you have to trust.
Read-only by default
ShelfProbe asks only for product read access. Any write-back needs your explicit approval first.
Continuous regression
Scheduled scans and change-triggered re-tests keep the same queries running, so regressions surface early.
Storefront and Global Catalog
Your own Storefront Catalog is the baseline; Global Catalog discovery is reported separately, never blended into one score.

Start with the products that matter most

Install ShelfProbe from your Shopify admin, open the console and run your first scan.
Open console