Rings.com Research

Which AI You Ask Changes What Jewelry Brand You Get Recommended

May 24, 2026  ·  GPT-4o, Gemini 2.5, Perplexity, Claude  ·  30 queries  ·  40 brands tracked  ·  1,192 scored responses

A study of 1,192 scored responses, generated from 30 shopping queries across ten independent runs and four AI models, reveals that major AI systems produce materially different jewelry recommendation ecosystems (different sets and rankings of brands) for the same consumer intent.

Key Findings: May 24, 2026

  • Three brands (Brilliant Earth, Blue Nile, and James Allen) account for roughly 40% of all AI jewelry recommendations across every model and query type tested. The gap between them and the fourth-ranked brand is substantial. The top three are not a toss-up: they are a structural feature of how AI recommends jewelry.
  • The AI model a consumer uses shaped their consideration set before they looked at a single product. ChatGPT skewed toward luxury brands. Gemini returned the broadest set. Perplexity surfaced mall chains; Claude avoided them.
  • Four Canadian brands were tracked across all queries and all ten runs: Birks, Michael Hill, Spence Diamonds, and People's Jewellers. None appeared in a single response from any model.
Share:

This page is the full research report. The consumer-friendly summary with shopping takeaways is in the consumer guide.

Consensus at the Top, Divergence Below

Three brands dominate AI recommendations regardless of which model is queried: Brilliant Earth, Blue Nile, and James Allen. Together they account for roughly 40% of all brand appearances across all models and query types. The gap between them and the fourth-ranked brand is substantial.

Brilliant Earth leads Blue Nile on both weighted score and total mentions, but the gap tells only part of the story. Blue Nile leads the general purchase-intent cluster and performs uniformly across all four models in that cluster. Brilliant Earth earns its overall lead through the ethics and sustainability cluster, where it received nearly 200 total mentions across ten runs, roughly 49 per model: a level of cross-model consensus not seen for any other brand on any other cluster. James Allen is consistent across clusters without leading any of them.

James Allen has merged into Blue Nile. Its parent company merged the brand into Blue Nile; jamesallen.com now redirects to bluenile.com. James Allen ranked third in this study, which was conducted in May 2026 before the migration completed. The Big Three became the Big Two, and the AI recommendation share James Allen held shifted without a formal successor. Blue Nile, which already led on the general purchase cluster, is the most direct beneficiary.

What a consumer finds at the top of their AI results depends on how they phrase the question.

Brand Mentions Weighted Score
Brilliant Earth9087,581
Blue Nile8407,333
James Allen7226,154
Whiteflash2972,043
VRAI3102,017
Ritani3142,011
Tiffany & Co.2591,965
Rare Carat2171,522
Clean Origin2221,385
Zales2011,267

35 of 40 tracked brands appeared in at least one response. Weighted score: 1st mention = 10 pts, scaling to 1 pt for 10th or later. Ten runs, May 2026. 1,192 scored responses after exclusion of 8 methodologically flawed rows.

The Models Diverge Below the Surface

The more significant finding is how sharply the four models diverge on everything below the top three.

ChatGPT (OpenAI)

Most concentrated; strongest skew toward heritage luxury brands.

ChatGPT returned the most concentrated recommendation patterns. Tiffany & Co., Cartier, Harry Winston, and David Yurman accounted for 15.7% of all ChatGPT brand mentions in this study, more than double Gemini's rate (6.9%) and more than four times Perplexity's (3.4%).

A consumer using ChatGPT was more likely to encounter a highly aspirational, premium consideration set.

Google Gemini

Broadest lists; highest brand diversity of any model tested.

Gemini produced the longest and most varied recommendation sets, surfacing 34 distinct brands across the dataset compared to fewer than 30 for each of the other three models. It led on mentions of VRAI, Ritani, Mejuri, and Grown Brilliance: brands that are lab-grown specialists or sustainability-oriented.

A consumer using Gemini got a meaningfully broader set of options than with any other model tested.

Perplexity

Strongest affinity for traditional retail chains.

Zales, Jared, Kay Jewelers, and Helzberg accounted for 17.1% of all Perplexity brand mentions, nearly double the rate of Gemini (9.7%) and ChatGPT (8.6%), and nearly four times Claude's (4.4%), potentially reflecting differences in how Perplexity retrieves and synthesizes web sources.

A consumer using Perplexity was more likely to be directed toward mall-based jewelry chains than with any other model tested.

Claude (Anthropic)

Most concentrated toward premium online retailers of any model tested.

Claude had the second-highest heritage luxury share (9.4%), behind only ChatGPT, and the lowest mall-chain share of any model (4.4%). No other model combined both: ChatGPT had a similar luxury lean but nearly double the mall-chain rate (8.6%).

A consumer using Claude was most likely to see an established premium online retailer and least likely to encounter a mall-based chain.

Brand Category Concentration by Model

Heritage Luxury Share

ChatGPT 15.7%
Claude 9.4%
Gemini 6.9%
Perplexity 3.4%

Mall-Chain Brand Share

Perplexity 17.1%
Gemini 9.7%
ChatGPT 8.6%
Claude 4.4%

Heritage Luxury: Tiffany & Co., Cartier, Harry Winston, David Yurman. Mall-Chain: Zales, Jared, Kay Jewelers, Helzberg. Both charts use the same 20% scale. Percentages reflect each category's share of that model's total brand mentions across 1,192 scored responses.

Canadian Consumers Get American Answers

Five of the ten runs were conducted from Vancouver, British Columbia, with no VPN. Five additional runs used VPN connections from five U.S. cities: San Francisco, New York, Chicago, Dallas, and Miami. Four Canadian jewelry brands were tracked across all runs: Birks, Michael Hill, Spence Diamonds, and People's Jewellers.

None appeared in a single response across any model, any query, or any run.

A Canadian consumer asking an AI where to buy an engagement ring receives the same U.S.-brand answer set as a consumer in San Francisco, New York, or Chicago. In this dataset, geographic location did not meaningfully alter AI jewelry recommendations. For Canadian retailers (and likely for non-U.S. retailers broadly), AI recommendation visibility is effectively absent.

Methodology

The study ran 30 queries across five intent clusters on four AI models with live web search enabled (ChatGPT, Google Gemini, Perplexity, and Claude). The clusters covered general purchase decisions, lab-grown diamonds, ethical sourcing, custom and gifting occasions, and price-value searches. Forty brands were tracked across ten independent collection runs in May 2026. Five runs were conducted from Vancouver, British Columbia with no VPN. Five additional runs used VPN connections from five U.S. cities: San Francisco, New York, Chicago, Dallas, and Miami. In this dataset, geographic location produced no detectable difference in results.

Brand appearances were scored by position. A first mention in a response earned 10 points, scaling down to 1 point for a tenth mention or later. The final dataset covers 1,192 scored responses, after removing 8 responses that contained a query later identified as methodologically flawed.

Query Design Disclosure

During data collection, an error in query design revealed something relevant to anyone conducting AI recommendation research. One of the original 30 queries asked models to compare two specific brands by name. Every model named both brands in response, not because it favored them, but because the question required it. That single query, representing 1.3% of the final dataset, was sufficient to flip the top two ranked brands.

The query was replaced before the final collection runs, and all prior results containing it were excluded from scoring. AI brand visibility studies are sensitive to query construction in ways that are not immediately obvious. A study that embeds brand names in questions will inflate those brands' scores.

All 30 Queries

Queries were grouped into five intent clusters.

Show all 30 queries ↓

General Purchase Decisions

  • "best place to buy an engagement ring online"
  • "where should I buy a diamond ring online"
  • "best online jewelry stores for rings"
  • "where to buy a wedding band online"
  • "best websites to buy fine jewelry"
  • "which online diamond retailer has the best reputation"
  • "best alternative to Tiffany for engagement rings"
  • "which online jeweler has the best diamond quality"
  • "most reputable online diamond retailers"
  • "best jewelry brands for engagement rings"

Lab-Grown Diamonds

  • "where should I buy a lab-grown diamond"
  • "best lab-grown diamond retailers"
  • "is it worth buying a lab-grown diamond engagement ring"
  • "best place to buy lab-grown diamond rings online"
  • "lab-grown diamond vs natural diamond where to buy"

Ethical Sourcing

  • "most ethical jewelry brands"
  • "best conflict-free diamond retailers"
  • "sustainable jewelry brands for engagement rings"
  • "ethical engagement ring brands"
  • "where to buy ethically sourced diamonds"

Custom and Gifting

  • "best place to buy an engagement ring"
  • "where to buy a wedding ring set"
  • "best jewelry brands for anniversary gifts"
  • "where to buy a custom engagement ring"
  • "best place to customize an engagement ring"

Price and Value

  • "best engagement rings under $3,000"
  • "affordable diamond engagement rings online"
  • "where to get the best value diamond ring"
  • "cheapest place to buy a real diamond ring"
  • "best engagement ring for the money"

What This Means for Consumers

AI systems are increasingly the first stop for purchase research. For jewelry (a category where a consumer may spend thousands of dollars and where brand trust matters), the model choice is shaping consideration sets in ways consumers are not aware of and have no tools to evaluate.

Asking ChatGPT where to buy an engagement ring produces a different brand list than asking Gemini. Asking about lab-grown diamonds produces a different list than asking about the best place to buy a ring generally. The same consumer, with the same intent, gets different recommendations depending on which app icon they tap first.

The brands at the top of AI recommendations did not earn those positions through a transparent ranking process. There is no disclosed methodology, no audit trail, and no mechanism for consumers to understand why one brand appears and another does not. This study documents how those recommendation patterns differ across major AI systems.

As AI systems increasingly replace traditional search behavior for purchase decisions, recommendation visibility inside those systems may become as commercially consequential as search ranking once was. That shift is already underway. This dataset documents where it stands as of May 2026.

Get future Rings.com Research in your inbox

New studies and data delivered when they publish. No other emails.

Published May 24, 2026. Data collected May 13 to May 22, 2026.

Rings.com receives no compensation from any brand included in this study for inclusion or ranking. No brand had editorial input into the methodology or results. Results may shift over time as model behavior, search integrations, and ranking systems evolve.

Full dataset available on request. Questions: rodney@rings.com