See what ChatGPT sees on your Shopify product

October 15, 2026
Shopify Tips
14 min read
See what ChatGPT sees on your Shopify product

Short answer: To see what ChatGPT sees on your Shopify product, open Sales channels, Agentic in your admin and use the Search preview and Listing insights, then search the Shop app, then query your store’s catalog endpoint. All four are free. Octolift, built by Craftshift, runs those checks on every product and scores the result out of 100. Free for 5 products.

Here is a thing most merchants don’t know about their own store: the product page you spent an afternoon polishing is not what ChatGPT reads. When a shopper asks for “a linen shirt that doesn’t wrinkle”, the assistant searches Shopify Catalog, gets back a structured record of your product (title, description, options, price, availability, policies, images, plus a few fields Shopify’s AI inferred on its own), and decides from that. Your hero banner, your size guide tab, your lifestyle video? Not in the record.

So the useful question is not “does my page look good”. It’s “what is in the record”. And you can look. Shopify ships two views of it inside the admin, the Shop app gives you a third, and the catalog endpoint itself will answer a plain HTTP request if you ask nicely. Ten minutes, no app, no developer. We’ll do all four, in the order they’re worth doing.

A caveat up front, because we’d rather you hear it from us than discover it later: every one of these views is directional. Shopify says so about its own preview, and the assistants re-rank on their side. What the views give you is not a guarantee. It’s the one thing you can actually act on.

In this post

How do you see what ChatGPT sees on a Shopify product?

You look at Shopify Catalog, because that is what ChatGPT looks at. Shopify’s help centre calls Catalog “the primary method for agentic storefronts to receive your product data”, and OpenAI’s merchant page says Shopify product data reaches ChatGPT through it with no extra work from you. The open web crawl still exists (that’s where robots.txt and page markup matter; our free AI bot checker and AI readiness checker cover that side), but for shopping results the catalog record is the thing.

Four ways in, from easiest to most revealing:

  1. Search preview in Sales channels, Agentic: which of your products the catalog returns for a query, top ten.
  2. Listing insights on the same page: a colored bar per product built from five named signals.
  3. Shop app search: Shop has been powered by Shopify Catalog since October 2025, so its ranking is the catalog’s ranking.
  4. The catalog endpoint: one curl command returns the actual record, field by field.

View 1: the Search preview in your admin

Go to Sales channels, then Agentic. The page exists for every eligible store since May 2026 (Shopify’s changelog announced it on May 11), and if you’ve never opened it, it has been quietly collecting data about you the whole time.

Find the field labelled “Enter a search query to see how you might rank”. Type what a shopper would type, not your product name. “Linen shirt for hot weather” beats “Ravello Shirt”. Press Enter. Two things can happen. If your products rank, you get a line like “10 of your products are in the top ranking search” and a list under “Your products that could rank for this query”. If none of them make the global top ten, Shopify shows you the products in your store that were most relevant to that query instead, which is its gentle way of saying you lost.

Don’t stop at one query. Just under the search box are two sections worth more than anything you’d type yourself: “Top search queries” and “Queries where your products appeared”. The second one is the list of real shopper searches where the catalog already returned you. Click each. Note where you land. Then click a channel name under “Preview this search” to see how that query might render in a given assistant’s chat interface.

What the preview is: the raw output of Shopify Catalog search. What it is not: ChatGPT’s final answer. Shopify’s own help text says agentic storefronts “often re-rank results based on their own logic” and tells you to “use this tool as a directional signal, not an exact prediction”. Fair enough. If you’re not in the raw top ten for your own obvious query, though, no re-ranking is going to put you there.

View 2: Listing insights, the five signals

Same page. Every product carries a Listing quality indicator, a colored bar meter, and a down arrow next to it opens Listing insights. This is the closest thing to a published ranking factor list that exists for AI shopping on Shopify, so read it slowly.

SignalShopify’s own definitionHow to check it by hand
Description completeness“Measures the word count of your product description.”Paste the description into a word counter. Under 100 words is thin; Octolift’s rater wants 150 or more before it stops docking points.
Image coverage“Measures the number of product images.”Count them in the product editor. One hero image is not coverage. Three with alt text is where most scoring starts to relax.
Product reviews“Measures your average product rating and review count.”Open the product’s storefront page and look for a rating. If your review app shows stars on the page but never writes them to Shopify’s reviews metafields, the catalog sees zero.
Variant & option completeness“Measures the number of variants, options, and their in-stock availability.”Look for “Default Title”, “Option 1” or abbreviations in option values. “Clr: NVY” is a failing grade.
Shop policy completeness“Measures whether your shipping, returns, refund, and other store policies are present.”Settings, Policies. Four pages: refund, shipping, privacy, terms. Written and published, not just drafted.

Shopify adds one sentence that is easy to skim past: “a better listing quality signal generally means that your product contains the necessary data for AI channels to rank and surface it accurately”. Necessary data. Not persuasive copy. The catalog is not grading your prose, it’s checking whether the facts are there to be matched.

View 3: search for yourself in the Shop app

This one is underrated because it feels too simple. On October 29, 2025, Shopify’s changelog announced that “the Shop app is now powered by the Shopify Catalog”, and that all of a store’s active online store products are discoverable there, whether or not you publish to the Shop channel. Which means Shop’s search box is a public window onto the same index ChatGPT queries.

Open shop.app or the Shop app, search the shopper phrasing, and scroll. Where do you land against competitors? Which competitor product is above yours, and what does its title say that yours doesn’t? Do this from a US location (or a VPN set to one) if you sell to the US, because catalog search results depend on the buyer’s country. We learned that the hard way building Octolift’s rank checks: the same query with the country set to US, then TR, then FR returned a different set of stores each time.

View 4: ask the catalog endpoint directly

Every Shopify store exposes a public catalog endpoint at https://your-store.com/api/ucp/mcp. It is the same interface AI agents use, documented on shopify.dev, and it needs no API key. The one thing it insists on is an “agent profile” URL in the request, a small JSON file that says what the caller can do. Shopify hosts example profiles for exactly this purpose, so you can borrow one.

Open a terminal, swap in your store domain and a shopper query, and run:

STORE="https://your-store.com"
PROFILE="https://shopify.dev/ucp/agent-profiles/2026-04-08/valid-with-capabilities.json"

curl -s -X POST "$STORE/api/ucp/mcp" \
  -H "Content-Type: application/json" -H "Accept: application/json" \
  -d '{
    "jsonrpc": "2.0", "id": 1, "method": "tools/call",
    "params": {
      "name": "search_catalog",
      "arguments": {
        "meta": { "ucp-agent": { "profile": "'"$PROFILE"'" } },
        "catalog": { "query": "linen shirt", "pagination": { "limit": 3 } }
      }
    }
  }'

We ran that exact command against one of our own demo stores while writing this and it came back in under a second. The response wraps the catalog’s answer as a JSON string inside result.content[0].text (an MCP quirk), so paste it into any JSON viewer to read it. You’ll find a products list, and each product carries title, description, url, price_range, options with every value label, variants with price and availability, media, categories and collections. An empty query with a limit of 10 is a quick way to see which products the catalog holds for your store at all.

Two things not to do with this. Don’t hammer it (the anonymous tier has the lowest rate limit, and a password protected store returns nothing anyway). And don’t expect ChatGPT’s answer to be a function of this record alone; the assistant adds its own search and its own judgement on top. What you’re reading here is the input, not the output.

How to read what comes back

Whichever view you used, you are looking for gaps. The global catalog record (the one ChatGPT searches across all stores) goes a step further than your store’s own endpoint: Shopify marks several fields as inferred, meaning its AI wrote them from your listing. Per Shopify’s developer docs those are the description, the normalized options, and under metadata the attributes (material, style, occasion), tech specs, top features and unique selling points. Shopify’s own note: treat them as discovery signals, not merchant-authored text.

That is the whole game. An agent comparing two shirts reads two inferred spec lists. If yours says nothing about fabric weight because your description never mentioned it, the comparison is lost before the assistant forms a sentence. So, reading the record:

  • Is every variant priced and marked available correctly? A stale “sold out” on a variant you restocked is a quiet killer.
  • Do the option labels read like a human wrote them? “Navy” and “Medium”, not “NVY” and “M2”.
  • Does the description state the facts a shopper would ask about: material, dimensions, care, what’s in the box, who it’s for?
  • Is the category right? Shopify classifies from your product information and that can differ from the category you picked; a product it files as Uncategorized is excluded from Catalog entirely.
  • Are there three or more images, each with alt text that says what is in the picture?

If you sell colours as separate products, read the record for each one. Each is its own catalog entry with its own gaps; a product that groups separate colour products into one listing with swatches on the storefront still shows the catalog one record per colour, which is good for matching and merciless about thin descriptions. And on the image side, a shopper who clicks “Olive” in ChatGPT and lands on a page showing the navy photos is a bounce; variant-specific image sets close that gap after the click.

Where the manual method stops

For a store with 20 products this is a pleasant hour. Open the Agentic page, run the top queries, open insights on each product, fix the thin ones, done. We’d genuinely tell a 20-product store to do it by hand and spend the money elsewhere.

At 300 products it changes shape. Three queries per product is 900 preview searches. Insights are per product, one click each, and the bar meter tells you something is weak without telling you which fact is missing. The curl output is a wall of JSON per product. And every one of those checks goes stale the moment someone edits a description or a variant sells out, which on a real store is daily. That is the point where Octolift runs the catalog query and the listing checks on every product and turns the result into a score out of 100 with the missing points named, then drafts the description, title and category fields to fill them. Free for 5 products, $29 a month for 100, and the “How you appear in the Shopify catalog” panel is view 4 above with the JSON already read for you.

Octolift product analysis screen showing the AI visibility score breakdown and the suggested fixes for a Shopify product
Octolift’s analysis page: the score, every check, and what would earn the missing points.

Either way, the manual version is worth doing once. Most merchants have never looked at their own catalog record, and the first look is usually the most useful hour they’ll spend on AI search this year. The scoring guide lists all twelve checks with their points if you want to grade yourself by hand.

Try Octolift free →

Frequently asked questions

Can I see what ChatGPT sees on my Shopify product for free?

Yes. The Search preview and Listing insights under Sales channels, Agentic are built into every eligible Shopify admin, Shop app search is public, and the store’s catalog endpoint answers a plain HTTP request with one of Shopify’s hosted example agent profiles. None of the four costs anything.

Is the Shopify search preview the same as what ChatGPT shows?

No. It shows the raw output of Shopify Catalog search. Shopify’s help page says agentic storefronts often re-rank results with their own logic and calls the preview a directional signal, not an exact prediction. It is the input ChatGPT works from, not its final answer.

Which product fields does Shopify Catalog send to AI assistants?

By default the title, description, options, images, price and availability, plus policies, categories and collections. Shopify’s optimizing guide also lists type, vendor, tags, barcode and variant option names as fields AI platforms consider. On top of that the global catalog infers a description, normalized options, attributes, tech specs, top features and unique selling points from your listing.

Why do I rank differently in the preview and in ChatGPT?

Three reasons: the assistant re-ranks on its side, results depend on the shopper’s country, and ChatGPT’s answers vary between runs. Test from the market you sell to and treat any single answer as one sample, not a rank.

What does “inferred” mean in the Shopify Catalog record?

Shopify’s AI generated or enriched that field from your product information. Shopify’s developer docs mark the description, options, attributes, tech specs, top features, unique selling points and variant condition as inferred, and say to treat them as discovery signals rather than merchant-authored text. Anything your listing never states shows up as missing.

Which app should I use to check what AI assistants see across my whole catalog?

Octolift, built by Craftshift, queries the Shopify Catalog for each product, shows how the catalog describes it with gaps marked “Not specified”, scores the listing from 0 to 100 across twelve checks, and drafts the fixes for you to approve. Free for 5 products and 5 shopper questions; paid plans start at $29 a month for 100 products. For a handful of products, the admin’s own Search preview and Listing insights are enough.

Start with your best seller. Run the preview, open its insights, search it in Shop, pull its record. Fix what’s missing, wait a day, look again. When that loop stops fitting in an afternoon, Octolift runs it for the whole catalog.

Co-Founder at Craftshift

You may also like

Available on the Shopify App Store