9 Claude prompts for checking what AI agents see on your store
Fill in every metafield and a product can still be invisible to an AI shopping agent.
Fill in every metafield on a product and it can still be invisible to an AI shopping agent.
Might not have reached the system doing the reading.
"AI visibility" gets treated like one system - when it's really three.
ChatGPT and similar shopping agents work from a payload built out of catalog data, queried live.
Google, Bing, and some of what Perplexity indexes read the JSON-LD schema embedded directly in the page instead.
Shopify's own agentic channels, Agentic Storefronts and the Shop App, decide whether a product is distributed to any of this at all, independent of how good the answer would be if it got there.
Sharing 9 prompts that check these three systems from Claude.
Run them all for the full picture - what an agent would see, what's causing the gaps, and whether the product is even live for agents.
1/ Reconstruct what the payload contains
Before checking whether a query would match, it helps to know what an agent is working with - the structured record behind it, built to a spec called Universal Commerce Protocol.
The prompt:
For each product in [COLLECTION], show me the attributes that would be included in a Universal Commerce Protocol payload sent to UCP-compatible agents. Use the product's current data: title, descriptionHtml, productType, vendor, tags, all visible metafields, image alt text, variant options.
What comes back:
→ Every UCP field present in the current data
→ Every field the payload would send empty, category and metafields most often
→ A rollup across the whole collection
Shopify doesn't expose the outbound payload through any public API, so this prompt reconstructs it rather than reading it live. What comes back is the shape of what an agent would work with, built from data already visible in the admin.
2/ Simulate which queries would surface your products
With this prompt, Claude predicts which natural language agent queries would match with your products and which it would lose to a better-structured competitor.
The prompt:
Given the product [PRODUCT], predict which natural-language shopper queries on ChatGPT would surface this product. Use the product's title, description, category, vendor, and metafields. Show me the queries it would match strongly, the queries where it would lose to a better-structured competitor, and the structural reasons for each.
What comes back:
→ Queries the product matches
→ Queries it loses, each tied to the specific empty field behind the loss
→ A short, ranked list of which metafields would recover the most queries if filled
Claude reasons from the catalog data the same way an agent would, since there's no "why did you rank this" endpoint to query directly.
To see live Shopify inferred attributes → run the free audit
3/ Run a shopper query against your catalog
The first two work from catalog data alone. This one is different - it runs a real query against Shopify's own search index, the one a shopping agent would draw from.
The prompt:
Treat this as an agent shopping query: [QUERY]. Search my catalog for the top 5 products that would match. Rank them by match strength.
What comes back:
→ The top 5 real matches from the catalog, ranked
→ A one-line reason for each match or near-miss
→ Whichever result should have won on paper but didn't
Run it for the 5 to 10 queries a product would rank on. If that product is not in the top 3 results in your own catalog, it will not be at the top of an agent search either.
4/ See the structural gap behind the loss
Queries lost in the last two checks trace back to the same root cause - empty metafields. This prompt is a check across a collection rather than just one product.
The prompt:
For each product in [COLLECTION], list which Shopify Standard Product Taxonomy metafields are empty. Cross-reference against the standard fields for the product's category: color, material, size, age group, target gender, country of origin, plus category-specific ones. Return a table per product: filled, missing, suggested.
What comes back:
→ A per-product table of filled versus missing fields
→ A summary across the collection - how many products are missing each field
→ A rough time estimate for fixing it manually
Sample Claude response
5/ Fix the single highest-impact field
Of the nine checks here, category mapping is the biggest lever. Wrong category and the product is effectively invisible to an agent no matter how good the rest of the data is.
The prompt:
Audit the Shopify Standard Product Taxonomy mapping on each product in [COLLECTION]. Show me: products with no category assigned, products with a category that looks too broad, products in a category that does not match the product type. Group by issue.
What comes back:
→ Every product with no category at all
→ Every product sitting one level too high in the tree instead of at the leaf
→ Any outright mismatch, wrong category for the product type
The Shopify Standard Product Taxonomy has thousands of categories. Most only need the leaf-level category. Claude can handle the assignment if you tell it the right node.
For bulk remapping, Atomz AI handles it programmatically.
6/ Product schema validation
The first five checks all read catalog data through the Shopify connector.
JSON-LD schema works completely differently - it's markup the theme renders directly into the page, and it's what Google's Shopping Graph, Bing, and some of what Perplexity indexes parse. The fix happens at the theme level, outside what’s covered so far.
The prompt:
Fetch the public product page for [PRODUCT_URL].
Parse the JSON-LD product schema embedded in the page.
Validate against schema.org Product spec.
Flag missing: offers, brand, aggregateRating, availability, image, sku.
Compare what is rendered vs what is in the underlying product data.
What comes back:
→ Which required schema.org fields are present on the live page
→ Which are missing, most often aggregateRating when no reviews app is installed
→ Anywhere the rendered schema disagrees with the product data, worse than a gap since it feeds an agent a confident wrong answer
Since the schema lives in the rendered page rather than the admin data the earlier prompts pulled from, this one needs web fetch access rather than the Shopify connector.
Run it on the highest-traffic products first.
7/ Agentic Storefronts publication status
None of the checks above matter if the product never reaches the AI channel.
This one is to confirm whether Agentic Storefronts is currently active as a sales channel and which products are published to it.
The prompt:
Tell me whether Agentic Storefronts is currently active as a sales channel on this store. For each product, show me whether it is currently included in the Agentic Storefronts publication or held back. Group: published, held back, draft (cannot be published).
What comes back:
→ Whether the channel itself is active
→ Every product published versus held back and why
→ The held-back reasons
Shopify auto-holds products that fail the data quality bar, usually the same issues an audit flags - empty description, no images or a stale draft collection.
8/ Distribution check on Shopify's other native surface
Check against a different channel - the Shop App feed and Shop Pay agent eligibility. Both running on Shopify's own infrastructure, which makes this the easiest of the nine to get a clean answer on. Worth a check across 15 to 20 products quarterly.
The prompt:
Check whether [PRODUCT] is currently included in the Shop App feed and visible to Shop Pay agents. If not, show me what is blocking inclusion. Common blockers: status DRAFT, no images, missing price, sales channel not selected, geographic restriction.
What comes back:
→ Current publication and eligibility status
→ A field-by-field readiness check against the known blockers
→ The specific blocker holding it back, if any
9/ Benchmark against your top competitors
Once the store's own picture is clear, the last useful check is relative - how the best product compares against whoever ranks above it for the same query.
The prompt:
Compare [PRODUCT] against [COMPETITOR_URL]. Show me the side-by-side payload difference: filled metafields, description length and specificity, image count, price positioning, structured attributes, reviews count.
What comes back:
→ A field-by-field comparison, one store against the other
→ Exactly which structured attributes the competitor has and the store doesn't
→ A gap list ordered by what to fix first
Pick one of your best sellers and the competitor currently outranking it on a query. The gap is small usually, and it comes down to metafields and reviews.
Three of the nine hit a wall Claude can't get past on its own: /ucp and /gpt-test above, plus part of /shop-feed's ranking check.
ChatGPT and Shop don't expose an endpoint for why they ranked something the way they did. The free audit checks it live instead, querying the same Shopify Catalog API that ChatGPT and Copilot actually shop from.
These 9 are a part of a library built for the Shopify and Klaviyo connectors in Claude. The others cover catalog hygiene, orders and customers, flows, and reporting.
- Ankit
If this was useful, the next one will be too.
Weekly. Free.
Unsubscribe with one click.
