Why did ChatGPT return other store links for jewelry brands?
Being recommended and being linked are different observations. In our June field study, ChatGPT named jewelry brands 248 times across 30 shopping answers, and only 88 of those mentions, 35.5%, included a link to the brand’s own store. The remainder pointed to retailers, marketplaces, articles, or no useful link. Here is what the sample supports checking.
What the returned links show.
The decoded payload behind those 30 answers included reddit.com (38 appearances) and youtube.com (35), with Forbes roundups and competitor blogs ahead of almost every brand’s own site. This records which domains appeared alongside the answers; it does not establish what caused any recommendation.
In a separate July product-level check, an official-domain match in Shopify’s Global Catalog co-occurred with an official product card in 88% of 11 live checks. When the match was missing, the same query returned other sellers, including a “Katbird Bracelet” on a store that is not Catbird. This is a small, correlational observation; it does not prove that a catalog match caused the returned card.
The fix, free, in five moves.
Audit yourself in five minutes.
Open a fresh ChatGPT session. Ask the question your buyer asks (“best solid gold jewelry brands”), then ask for your top three products by exact name. Write down where every link lands: your product page, a retailer, an article, or nowhere.
Make every product page machine-readable.
Product structured data on every PDP: name, image, price, availability, and your brand as the seller, all agreeing with the visible page. We have observed results exposing two different prices for the same product; conflicting data reads as untrustworthy to the systems choosing links.
Fix your catalog layer.
On Shopify, confirm your products are present and readable in your storefront catalog and matched to your official domain in the Global Catalog. A missing product record can leave an unowned or other-seller result, so inspect the returned path before drawing conclusions.
Reduce stale links.
Keep exact product titles consistent everywhere you sell, redirect stale product pages instead of letting them 404, and make sure the intended PDP exists. Every dead or drifting PDP is a path to inspect, not evidence of a lost sale.
Test an open question.
Some buying intents have no settled answer. In jewelry, our study saw the birthstone and personalized intent name just 5 brands across three runs, while occasion gifts scattered across 19. Publish a useful, honest answer to one specific buying question, then measure returned links across repeated runs.
The numbers come from Caeliai Field Study 01: 30 ChatGPT Instant-mode shopping answers, 10 jewelry buying intents run 3 times each, captured June 29, 2026 as public share links. The catalog findings come from a separate July product-level check of 40 Shopify products. Directional evidence, dated on purpose: these systems change without notice.
Check your public store signals.
Run the five moves yourself, then use the public store checker to review the product, buying-path, and crawler signals available to AI shopping and search systems.