We Asked ChatGPT to Recommend Independent Fine Jewelry. Here’s What Happened.

What happens when a shopper asks ChatGPT to discover independent fine jewelry? In five untouched sessions, the answer changed dramatically with the shopper’s intent.

For independent jewelry brands, the rise of AI shopping creates an obvious question: when someone asks ChatGPT what jewelry to buy, which brands actually get recommended?

Found & Form ran a small exploratory test. We opened five fresh, logged-out ChatGPT conversations in an incognito browser and asked five different versions of a jewelry discovery question. Every prompt explicitly described the shopper as being in the United States.

We did not regenerate responses. We did not correct ChatGPT. We did not follow up. We recorded the first answer exactly as it appeared.

The results were messier, and more interesting, than a simple list of winners.

The experiment

The prompts represented different ways a real customer might shop: broad discovery, under-the-radar discovery, a dark or slightly Gothic aesthetic, a specific material request for solid gold with black diamonds, and a purchase-intent query with a $500–$2,500 budget.

The first two prompts returned 10 brands each. The Gothic prompt returned only six named designers before shifting into broader shopping advice. The black-diamond query produced four clear solid-gold designer recommendations plus one caveated suggestion that did not meet the brief. The budget query returned only five brands.

AI discovery is not behaving like a conventional search-results page with a predictable number of blue links. The answer itself is being composed.

Some brands surfaced more than once

Most of the 32 brands appeared only once. Four repeated across different shopping intents: WWAKE, Sofia Kaman, Bario Neal and Automic Gold.

That repetition matters. A brand does not necessarily need to be the most famous name in the category to become useful to an AI system. It may instead be particularly legible for multiple shopping intents.

Specificity changed the brands dramatically

The broad query produced names including WWAKE, Brent Neale, Foundrae, Catbird, Valerie Madison, Cece Jewelry and Kataoka. Once the prompt became more specific, the list changed.

When we asked for unusual fine jewelry with a slightly Gothic or dark aesthetic, ARTËMER appeared first. Then we asked for independent designers making interesting solid-gold jewelry with black diamonds. Maya Geller appeared first.

That answer did not just mention the Tel Aviv designer. ChatGPT described several specific black-diamond designs and displayed four Maya Geller product cards. There is a difference between a brand being mentioned by an AI assistant and its products becoming part of the shopping interface. For independent brands, those may become two different forms of discoverability.

The location experiment did not behave cleanly

Every prompt explicitly identified the shopper as being in the United States. That influenced the recommendation pool, but it did not erase other location signals.

ARTËMER, based in Tel Aviv, became the first recommendation for the dark/Gothic query. Maya Geller, also based in Tel Aviv, became the first recommendation for black-diamond jewelry. More strikingly, product prices displayed during the test appeared in Israeli shekels.

The explicit U.S. shopping context therefore did not completely override the environment’s apparent geographic context. The same question may not produce the same discovery experience for every shopper.

Visibility is not the same as relevance

The black-diamond prompt specifically requested solid-gold jewelry. ChatGPT nevertheless included Remeinz as an additional suggestion while acknowledging that the brand’s standard metal was sterling silver and suggesting that the shopper contact the studio about a gold version. After publication, founder Maddi Kligerman confirmed that Remeinz currently offers sterling silver and black diamonds, not gold; gold is planned for the future. She also described Dark / Gothic as an appropriate category for the brand.

That is not a clean match. It is also a useful example of what gets lost when AI visibility is measured only by whether a brand appeared.

AI discoverability is not a leaderboard

Perhaps the biggest lesson from these five prompts is that asking “Which jewelry brands does ChatGPT recommend?” is the wrong question.

There was no single permanent list. The recommendation set changed substantially when we changed the customer’s language, aesthetic, material requirement and budget.

The more useful question may be: What does an AI system understand your brand to be relevant for?

A brand could be nearly invisible for “independent fine jewelry” and highly visible for “solid-gold black-diamond jewelry.” Another might dominate an ethical-jewelry query but disappear from a design-led one. In traditional search, brands spent years thinking about keywords. AI shopping may force them to think more deeply about context.

What this test does not prove

This was an exploratory experiment, not a controlled ranking study. Five prompts are not enough to determine why a particular brand surfaced. ChatGPT responses can vary, and geographic context, product data, public web content, third-party sources and other signals may all influence an answer.

We also cannot conclude from an appearance that a brand deliberately optimized for AI. In fact, that question is what interests us next.

Found & Form is investigating a larger pre-existing research universe of independent jewelry companies to understand what the brands that surface in AI discovery have in common with those that do not. We are particularly interested in a phenomenon we are calling accidental AI discoverability: brands becoming understandable and recommendable to AI systems without a reported deliberate AI-search strategy.

If that pattern holds, the implications could be significant. The next discovery advantage may not begin with learning how to “game” an AI system. It may begin with making the brand exceptionally easy to understand.