- 5/5 products readable in its own storefront catalog.
- 4/5 products matched to the official domain in Shopify’s Global Catalog, ranked first.
- Checked products rendered official ChatGPT product cards with correct product and price.
- One edge product (an add-on plan) missing from the global layer: the one item to fix.
What a Caeliai diagnosis looks like.
This is a real deliverable format filled with real, anonymized data from Caeliai’s July 2026 jewelry-category research. It is a demonstration, not a client case study: the brands below did not hire us, so their names are withheld.
Executive summary.
We checked how ChatGPT handles exact-product shopping queries for two jewelry brands of similar size, here called Brand A and Brand C. Brand A owns its buy path: its products sit in Shopify’s catalog layer under its official domain and render as official product cards. Brand C is being recommended and still losing the sale: it is missing from the global catalog layer, and every one of its exact-product queries surfaced a non-official seller first, including resale listings. Same platform, same category, opposite outcomes. The fix list for Brand C starts with its product data, not its marketing.
2
Anonymized jewelry brands, 5 products each: bestsellers, high-ticket, bundles, edge items
3
Layers checked per product: brand storefront catalog, Shopify Global Catalog, live ChatGPT
10
Exact-product prompts, one fresh ChatGPT conversation each
30
Category-level ChatGPT answers backing the category context (248 brand mentions)
Queries and platforms tested.
Exact-product prompts of the form “I’m shopping for the [brand] [product name]. Show me product cards or PDPs for this exact item” were run in fresh ChatGPT conversations, one product at a time. Category context comes from 30 ChatGPT shopping answers across 10 jewelry buying prompts, run 3 times each. A full paid diagnosis runs the same design on ChatGPT and Gemini; this public sample shows the ChatGPT side.
In a client deliverable, every observation below links to a dated screenshot and the recorded response. Evidence for this public sample is withheld because it would identify the anonymized brands.
What we observed, product by product.
- 0/5 products readable in its storefront catalog endpoint.
- 0/5 products matched to the official domain in the Global Catalog.
- 5/5 queries returned a non-official seller first: resellers and stale listings wearing the brand’s name.
- Still named in category answers, so shoppers meet the brand and get routed past its store.
Every product in a diagnosis lands in one of four classes: owns the path (official card, official store link), leaks the path (recommended, but a retailer, marketplace, or reseller gets the click), named only (mentioned with no usable buy path), or absent (never appears for prompts it should win). At the category level in this data, only 35% of brand mentions carried a brand-owned link (88 of 248 mentions across 30 answers, measured June 2026).
Recommended actions for Brand C.
01
Restore catalog readability.
Its products are invisible to the catalog layer AI systems read. First fix: get the storefront catalog endpoint serving its products, then verify the Global Catalog attributes them to the official domain.
02
Displace the wrong sellers.
Resale and stale listings currently win its exact-name queries. Clean product titles, canonical PDP URLs, and correct availability give the official listing something to win with.
03
Retest on a schedule.
Rerun the same product checks after the fixes, then again at 14 and 30 days. AI systems re-crawl on their own clock; the retest is what shows whether the fix landed.
Measurement limitations.
These are single-point-in-time observations of systems that vary run to run; a paid diagnosis repeats prompts and reports rates, per the methodology. Catalog presence is a diagnostic signal, not proof that it causes ChatGPT’s behavior. And live ChatGPT checks are rate-limited, so coverage per product is stated, never assumed.
What a paid engagement adds.
The five-day baseline runs this design at full width: 100 brand- and product-focused checks across ChatGPT and Gemini, your competitors mapped by name, your Shopify and GA4 numbers wired in, dated evidence for every claim, and an action plan ordered by expected impact. Implementation work then executes that plan in your store: product structured data, catalog and feed corrections, product-page data cleanup, and analytics wiring, retested at 14 and 30 days until the path points at your store.
Want this for your store?
Start with the free score: we run real shopping prompts for your category and show you where the recommendations land. One form, no meeting.