Rings.com Research & Data

Rings.com Research  ·  Edition 3

The Mall Jewelry Chains Are Fading From AI Chat Recommendations

---

July 27, 2026  ·  OpenAI, Gemini, Perplexity, Claude + Google AI Overviews  ·  30 queries  ·  36 brands tracked  ·  2,400 responses + 600 AI Overviews

A third month of systematic testing means this study can finally do what it promised in June: separate real movement from measurement noise. This edition also upgrades how that line is drawn. Instead of the three-editions-in-a-row rule announced in June (a test that turns out to be too easy to pass by chance), every brand's change since May is now measured against its own run-to-run noise, computed from all 40 collection runs. Fifteen brands clear a strict version of that bar, and the clearest picture among them is a slide by the mall jewelry chains in AI chat answers. Google's own answer box, notably, disagrees.

Measured Against Its Own Noise: What Has Really Moved Since May

In June, this study committed to a rule: no brand's movement would be called a trend until three consecutive editions moved in the same direction. Edition 3 is the first edition where that test could run, and running it honestly means retiring it. With three editions there are only two month-to-month steps, so a brand moving at random has roughly a coin-flip chance of showing a consistent direction, and the 17 of 34 brands that did is almost exactly the coin-flip rate. A rule that noise passes half the time is not a bar. This edition replaces it with a stronger test: using all 40 collection runs (10 in May, 10 in June, 20 in July), we measured each brand's own run-to-run scatter, and movement is claimed only when a brand's change since May is larger than that scatter can plausibly produce. Fifteen of the 34 three-edition brands clear the strict version of that bar, set at 3.5 standard deviations, a level at which chance across 34 brands would be expected to produce essentially none (about 0.02 false positives). One thing this test does not claim: it establishes that a brand's visibility changed since May beyond measurement noise, not that the path there was a smooth month-by-month slope, and this report no longer asserts one.

The most coherent group among the decliners is the mall chains, and under the new test all four of them qualify. Helzberg has declined from 638 to 551 to 364 in comparable weighted score, a 42.9% drop from its May baseline and 6.6 standard deviations beyond its measured noise. Kay Jewelers has fallen from 857 to 846 to 617, down 28.0% (4.4 SD). Zales has fallen from 1,267 to 1,199 to 1,099, down 13.3% (3.6 SD). And Jared, which June's rule would have excluded because its slide paused in July, sits 31.1% below May (981 to 661 to 676), a 4.2 standard deviation decline the noise test correctly counts: the drop happened in June and has simply held.

The decline is a chat-model phenomenon, and Google's answer box is the counterexample in the same dataset. In Google AI Overviews this month, Zales ranks 5th (90 mentions) and Kay Jewelers ranks 7th (78 mentions), far above their chat-model ranks of 11th and 18th. Whatever is eroding the mall chains' position in ChatGPT, Gemini, Perplexity, and Claude answers has not reached the surface most shoppers meet first on Google.

What remains unresolved is why. Movement beyond measured noise establishes that the change is real; it does not establish the cause. The declines could reflect the models' source material turning over, retrieval systems reweighting which sites they read, or genuine shifts in how the web talks about these brands. This study measures the output, not the mechanism, and the source-mix data below shows the retrieval side did move this month, which is reason to hold the causal question open.

A note on reading this edition's numbers. Edition 3 doubled collection depth to 20 runs (2,400 core responses), versus 10 runs in May and June. Raw mention counts and weighted scores in this report are therefore roughly twice the size of prior editions by construction. Every cross-edition comparison and delta in this report uses the 10-run equivalent: the July weighted score divided by two. The scoring formula itself is unchanged. Changing collection depth in the same edition that introduces a movement test deserves a skeptical flag, so here it is, addressed: the doubled run count is what makes the noise measurement possible at all, it cuts July's sampling noise roughly in half, and the divide-by-two normalization was verified by exact replication of every published total. Details in the methodology.

The Overall Leaderboard

At the very top, the series has its first lead change, and it should be read cautiously. Blue Nile posts the edition's highest weighted score: 15,698 raw, or 7,849 in 10-run-equivalent terms, against Brilliant Earth's 15,552 raw and 7,776 equivalent. The 73-point equivalent gap is well inside measured noise: computed directly across July's 20 runs, one standard deviation on the gap between the two leaders is 58 points, which puts the 73-point lead at 1.3 standard deviations, statistically a tie. Brilliant Earth still leads on raw mentions, 1,851 to 1,744. Blue Nile leading on weighted score while trailing on mentions means the models name it earlier in their lists when it appears. The two leaders separate cleanly under the noise test, though: Blue Nile's rise since May clears the strict bar at 8.1 standard deviations, the largest supported move in the study, while Brilliant Earth's rise, though pointed the same way in all three editions, totals only 1.7 standard deviations and stays inside its own noise. This report claims the first and not the second.

James Allen holds third, the position it has occupied in all three editions, with 1,367 mentions and an 11,425 weighted score. In equivalent terms it is essentially flat versus June (+14.5), which means June's sharp drop, the largest single-month decline of Edition 2, did not continue. The noise test reads the situation precisely: James Allen sits significantly below its May level (2.9 standard deviations), but the entire decline happened in June and has since held flat. A real drop, then a stop, not a continuing slide.

Brand	Mentions	Weighted	vs Jun*
Blue Nile	1,744	15,698	+499
Brilliant Earth	1,851	15,552	+164
James Allen	1,367	11,425	+14.5
Ritani	740	4,670	+413
VRAI	710	4,577	+82.5
Tiffany & Co.	558	3,977	+0.5
Whiteflash	549	3,812	-131
Rare Carat	427	2,933	-126.5
Mejuri	406	2,816	+269
Clean Origin	378	2,409	-77.5
Zales	365	2,198	-100
Cartier	286	2,189	-8.5
Grown Brilliance	261	1,644	+17
Bario Neal	192	1,607	+9.5
Harry Winston	199	1,437	+50.5
Jared	215	1,352	+15
With Clarity	238	1,315	+59.5
Kay Jewelers	217	1,234	-229
Aurate	154	932	-75
Shane Co.	143	767	-429.5
Helzberg	119	728	-187
Taylor & Hart	91	649	+94.5
David Yurman	84	570	+3
Costco	90	520	-18
De Beers	75	501	-32.5
Frank Darling	98	423	+33.5
Do Amore	57	389	-346.5
Brian Gavin Diamonds	57	336	-162
Pandora	59	308	+73
Angara	63	290	-132
Tacori	41	260	-26
Swarovski	27	117	-10.5
Keyzar	24	93	-81.5
Lightbox	7	43	-6.5
Leibish & Co.	7	38	-25
Michael Hill	1	7	new

Show all 36 brands

36 brands recorded at least one mention across 2,400 scored core responses (20 runs, July 2026). Weighted score: 1st mention = 10 pts, scaling to 1 pt for 10th or later. *"vs Jun" compares the July 10-run-equivalent weighted score (raw score divided by 2, since July collected twice the runs) against the Edition 2 (June 2026) score; fractional values are exact halves. Hearts on Fire, tracked since Edition 1, recorded zero mentions this edition. Michael Hill appears for the first time (see coverage note in the methodology).

Leaderboard Across Editions: May, June, July

This study runs every month. The table below tracks the weighted score of the top brands across every edition, so the change is visible rather than just the latest snapshot. A new column is added each edition; the July column is stated in 10-run-equivalent terms so all three columns are directly comparable. The final column reports the noise test: whether the brand's change since May exceeds its own measured run-to-run noise, at a strict bar of 3.5 standard deviations or a supporting bar of 2.

Brand	May 2026	Jun 2026	Jul 2026*	Change vs May
Blue Nile	7,333	7,350	7,849	Up (strict bar)
Brilliant Earth	7,581	7,612	7,776	Within noise
James Allen	6,154	5,698	5,712.5	Down (2 SD)†
Ritani	2,011	1,922	2,335	Up (strict bar)†
VRAI	2,017	2,206	2,288.5	Up (strict bar)
Tiffany & Co.	1,965	1,988	1,988.5	Within noise
Whiteflash	2,043	2,037	1,906	Down (2 SD)
Rare Carat	1,522	1,593	1,466.5	Within noise
Mejuri	1,107	1,139	1,408	Up (strict bar)
Clean Origin	1,385	1,282	1,204.5	Down (2 SD)
Zales	1,267	1,199	1,099	Down (strict bar)
Cartier	1,022	1,103	1,094.5	Within noise

Weighted score by edition for the current top 12 brands. *July shown as the 10-run equivalent (raw score divided by 2) because July collected 20 runs versus 10 in May and June. The strict bar is a change since May of at least 3.5 standard deviations of the brand's own measured run-to-run noise; at that level, chance would produce about 0.02 qualifying brands across 34 tested, effectively zero. The 2 SD bar is supporting evidence only, since chance would produce 1 or 2 there. Beyond the top 12, the strict-bar movers are: up, Taylor & Hart (5.9 SD), Pandora (4.2), With Clarity (4.1), and Aurate (3.7); down, Shane Co. (9.4), Helzberg (6.6), Brian Gavin Diamonds (5.0), Kay Jewelers (4.4), Jared (4.2), and Angara (3.6). At the 2 SD bar: De Beers and Keyzar, both down. Brands that moved the same direction in all three editions but whose total change stays inside noise, including Tiffany & Co. (+23.5 points, 0.3 SD), David Yurman, Grown Brilliance, and Swarovski, are listed as within noise: direction alone is not evidence, because a brand moving at random shows a consistent direction about half the time. † marks a path that was not one-directional: Ritani fell in June before surging in July, and James Allen's entire decline happened in June and has held flat since. Among the movers beyond the top 12, Aurate, Shane Co., Jared, Brian Gavin, and Angara also moved non-monotonically. The test measures where a brand stands versus May, not the shape of the path.

This Month's Largest Moves

The single-month movers are listed below in 10-run-equivalent terms. The largest single-month increase (Blue Nile, +499) and decrease (Shane Co., -429.5) tell you about July, while the since-May test above covers the full period. Both of this month's extremes also clear the strict bar on their net change since May, but Do Amore's reversal shows why even a statistically real single month needs another edition before it says anything about where a brand is heading.

Change in 10-run-equivalent weighted score versus the June 2026 edition.

Which Model You Ask Still Decides What You See

The four models still agree at the very top and diverge below it, but two of June's signature divergences moved this month. OpenAI remains the most James-Allen-friendly model (480 mentions, 24.0 per run versus June's 24.3). But Gemini cut its James Allen rate roughly in half (16.5 per run in June to 10.4 in July), while Perplexity, June's most James-Allen-skeptical model, doubled its rate (7.5 to 15.1 per run). Both are single-month moves. The Rare Carat split, meanwhile, persists: OpenAI mentioned it exactly once in 600 responses this edition, after exactly once in June and zero times in May, while Claude mentioned it 165 times this month alone.

Brand	Claude	Gemini	OpenAI	Perplexity
Brilliant Earth	501	454	499	397
Blue Nile	428	423	487	406
James Allen	377	208	480	302
Ritani	223	275	111	131
VRAI	181	272	105	152
Tiffany & Co.	158	140	188	72
Whiteflash	152	188	112	97
Rare Carat	165	160	1	101

Total mentions by model across twenty runs (600 responses per model). Gemini logs the most total brand mentions (3,705) and Perplexity the fewest (2,393); OpenAI remains the most James-Allen-friendly and the most Rare-Carat-averse.

Brand Category Concentration by Model

June's sharpest model divergence partially collapsed this month, and that itself is the finding. The heritage-luxury gap persists: heritage brands are 15.4% of OpenAI's total brand mentions versus 6.4% of Perplexity's, a 2.4x gap (down from 3.1x in June). But the mall-chain gap nearly closed: Perplexity's mall-chain share fell from 14.2% to 6.6%, Claude's rose from 3.4% to 6.9%, and the widest remaining gap is just 1.3x (Gemini's 8.5% versus Perplexity's 6.6%), versus 4.2x in June. A convergence this large in one month across multiple models is more consistent with retrieval-side changes than with four independent editorial shifts, but that is an inference; this study can only report that the divergence narrowed.

Model	Heritage luxury share	Mall-chain share
ChatGPT (OpenAI)	15.4%	8.3%
Claude	9.9%	6.9%
Gemini	6.6%	8.5%
Perplexity	6.4%	6.6%

Heritage luxury: Tiffany & Co., Cartier, Harry Winston, David Yurman. Mall-chain: Zales, Jared, Kay Jewelers, Helzberg. Each figure is that category's share of the model's total brand mentions this edition. Across all four models combined, the mall chains' share of total mentions fell from 9.8% (May) to 8.9% (June) to 7.7% (July), the aggregate view of the mall-chain declines above.

Google's Answer Box Is Still a Different Game

Google AI Overviews fired on 89.8% of the jewelry queries tested (539 of 600), down from June's 94.3%. A 4.5-point single-month drop in fire rate is worth watching but not yet a trend; it still means roughly nine in ten jewelry searches meet an AI answer before any traditional result.

The top of Google's leaderboard mirrors chat with one inversion: Brilliant Earth leads the answer box (371 mentions to Blue Nile's 312) while Blue Nile leads chat on weighted score. Below the top three, Google continues to describe a different market than the chat models, and this month the difference runs directly against the headline decline: Zales ranks 5th and Kay Jewelers 7th in AI Overviews, against chat ranks of 11th and 18th. The brands fading from chat answers are holding their ground in Google's box.

Rank	Brand	Overview mentions	Chat rank
1	Brilliant Earth	371	2
2	Blue Nile	312	1
3	James Allen	151	3
4	Ritani	91	4
5	Zales	90	11 ↑
6	Rare Carat	83	8
7	Kay Jewelers	78	18 ↑
8	VRAI	73	5
9	Bario Neal	62	14 ↑
10	Mejuri	59	9

Google AI Overviews leaderboard, 539 fired overviews out of 600 collected. "Chat rank" is the brand's weighted rank among the four chat models this edition. Upward arrows mark brands ranking materially higher in Google's box than in chat.

Google still de-emphasizes the merged brand the most, for the second consecutive edition. Among the chat models, James Allen is mentioned at 78% of Blue Nile's rate this month (1,367 versus 1,744). In Google AI Overviews it runs at just 48% (151 versus 312). The same split was 82% versus 51% in June. Two editions with the same shape is a repeated observation, not yet a statistically tested movement, and we report it as a pattern, not a mechanism.

Where Google's Answers Actually Come From

Across the overviews collected this edition, Google cited 580 distinct domains 5,287 times. The composition shifted modestly versus June, and per this study's practice, source-mix movement is checked before brand movement is interpreted: Forbes rose from 5.6% to 7.5% of all citations, google.com properties from 5.6% to 7.0%, and YouTube from 7.9% to 8.6%, while Reddit fell from 7.1% to 6.2% and whiteflash.com from 4.7% to 3.7%. Shifts of one to two points in the source mix are exactly the kind of retrieval-side movement that can move brand counts for reasons that have nothing to do with the brands.

New this edition: the chat models' own citations. Starting in July, the study also captures which web domains the four chat models cite in their answers, a baseline that could not be collected retroactively for May or June. Across 2,400 chat responses, the models cited 887 distinct domains 21,361 times. The top sources: forbes.com (2,287), brilliantearth.com (1,114), reddit.com (820), youtube.com (513), pricescope.com (506), and whiteflash.com (477). One capture caveat: OpenAI's API exposes citation annotations far more sparsely than the other three (231 captured citations versus 8,086 for Claude, 6,583 for Gemini, and 6,461 for Perplexity), so cross-model totals should not be read as cross-model comparisons. From Edition 4 onward, these citation counts become comparable month to month.

The route into an AI answer continues to look the same on both surfaces: presence in community discussion, placement in a small set of high-authority publishers, and strong owned content the systems can cite directly. Forbes alone now accounts for the single largest share of chat-model citations captured, and its share of Google AI Overview citations grew this month.

AI Is Still Working From a Stale Map

Month three of tracking the market-structure gap, and the needle has not moved. James Allen has redirected to Blue Nile under Signet since the spring, yet 1,367 of 2,400 responses (57.0%) described it this month as an independent, standalone brand, essentially unchanged from June's rate of 57.1% (685 of 1,200). Descriptions of Blue Nile as lab-grown focused (its post-acquisition direction is a natural-diamond emphasis) ran at 33.8% of responses, up slightly from June's 31.8%. These are methodology flags, not scoring changes: scores are preserved exactly as the models cited them, and the discrepancy is documented alongside.

The persistence is the story. In June it was possible to argue the models simply had not caught up yet. A month later, with the consolidation reported in the trade press since March, the standalone-James-Allen rate is unchanged to within a tenth of a percentage point. Whatever update cycle would correct this has not run, on any of the four models.

Methodology

Rings.com tracked 36 jewelry brands across 30 queries in five intent clusters (general purchase, lab-grown diamonds, ethical sourcing, custom and gifting, and price-value) using four AI models with live web search enabled: OpenAI (gpt-4o), Google Gemini (gemini-2.5-flash), Perplexity (sonar), and Anthropic's Claude (claude-sonnet-4-6). Twenty independent collection runs were conducted on July 23 and 24, 2026, twice the run count of the May and June editions. A supplementary source, Google AI Overviews, was collected across the same 30 queries and 20 runs, firing on 539 of 600 (89.8%).

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. This formula is unchanged from Editions 1 and 2. Because July collected 20 runs against 10 in prior editions, all cross-edition comparisons in this report use the 10-run equivalent: the raw July score divided by two. Raw July totals appear in the main leaderboard. New this edition, the report also carries a measured noise model. For every brand, we computed the standard deviation of its per-run weighted score within each month (10 runs May 13 to 22, 10 runs June 16 to 29, 20 runs July 23 to 24). The noise on a monthly total is that per-run standard deviation times the square root of the number of runs, and the noise on a change between two months combines both endpoints as the square root of the sum of their variances. A movement since May is claimed at 3.5 standard deviations of that combined noise (the strict bar; expected false positives across 34 brands, about 0.02) and noted as supporting evidence at 2 standard deviations (where chance would produce 1 or 2). Those chance rates treat each brand's monthly totals as independent and approximately normal; in practice brands are not fully independent, because a single response can mention several brands at once. We accept that simplification and state it here rather than leave it for a reader to find. Two limits are worth stating. First, May's and June's runs were spread across 10 and 14 calendar days versus July's 2, so their scatter absorbs some real within-month movement; that inflates the noise estimate and makes the test conservative. Second, the test measures where a brand stands versus May, not whether its path was smooth month to month. The final dataset comprises 2,400 scored core responses and 600 Google AI Overview collections, with zero gaps, duplicates, or empty responses. Starting this edition, the collector also records the web domains each chat model cites (a July baseline; not available for May or June) and the resolved model version of every response.

Representativeness note: the OpenAI surface is measured through the gpt-4o API with web search, which keeps the longitudinal series internally consistent. Consumer ChatGPT may use newer models with different retrieval behavior, so this study's OpenAI figures describe the API surface, not necessarily the consumer app.

Coverage note (Canadian retailers): Four Canadian brands (Birks, Michael Hill, Spence Diamonds, and People's Jewellers) are tracked in every edition regardless of whether they appear. This edition produced the first-ever appearance: Michael Hill was mentioned once, by Gemini, in a "best conflict-free diamond retailers" response on July 24. That is 1 mention in 2,400 responses, a noise-level result, not a breakthrough. Birks, Spence Diamonds, and People's Jewellers appeared zero times across the 2,400 core responses and 600 AI Overview collections. These are brand-tracking counts, not a geographic-location test.

All 30 Queries

Queries were grouped into five intent clusters. The query set is unchanged from Editions 1 and 2.

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

Three editions in, the series can finally separate signal from noise, and the first signal to clear the noise bar is structural: the brands built on physical mall retail are losing ground in AI chat answers, while the brands gaining are the ones the models' favored sources write about. At the same time, the surfaces are diverging. A shopper who asks a chat model and a shopper who reads Google's answer box are now seeing measurably different markets, and neither is seeing the market as it actually exists, because the models' underlying picture of who owns whom has not updated in the three months this study has been measuring it.

For brands, the lesson of the first statistically supported movements is that AI visibility genuinely moves, measurably and sometimes quickly, which means it can be watched, and it rewards a specific content profile rather than legacy fame. For consumers, the practical advice is unchanged from June but now rests on firmer data: which app you ask determines what you see, and what you see may describe a market that no longer exists. This dataset documents where that stands as of July 2026; the August edition will show whether this month's single-month swings were the start of anything.

Published July 27, 2026. Data collected July 23 to 24, 2026. Edition 3 of an ongoing monthly series; Edition 1 published May 24, 2026, Edition 2 published June 30, 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. Rings.com also has an interest in how the broader story of AI visibility lands, a conflict no disclosure sentence removes; the response is that every claim here is published with the data and thresholds needed to check it independently. Rings.com is itself a participant in the market this study measures: it operates Ring Advisor, a free ring-shopping guidance tool that draws on Rings.com research, including this study. For that reason Rings.com is tracked in this study under the same methodology as every other brand. Through all three editions it has recorded zero mentions in the scored chat responses, zero appearances in Google AI Overviews, and zero of the 21,361 model-cited source domains logged in July. If that changes in a future edition, it will be reported in that edition. Results may shift over time as model behavior, search integrations, and ranking systems evolve.

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