Ask Google for an independent fine jewelry designer and you get one shortlist. Ask ChatGPT, Gemini or Perplexity the same question and you get another. Across 20 shopper questions, only 3% to 8% of the designer recommendations the assistants made had the designer’s own website in Google’s matching top 10.
That is a narrower claim than it may sound. We counted a designer as present only when the designer’s own domain was among Google’s top 10 organic results for the same question, typed as a search. A designer described on a retailer page that ranks, or ranking for a different search, counts as absent. So the finding is about where Google sends a shopper for these questions, not whether Google knows the designer exists.
Key findings
- 92–97%
of designers recommended by AI in our tests had no own website in Google’s top 10 for the same search. It holds without ARTËMER, a brand Found & Form has researched (93.0%, 97.0%, 97.1%). Resampling the 20 questions puts the 95% range at 88–96% (Gemini), 92–99% (ChatGPT) and 94–99% (Perplexity). Counting any of the 20 Google searches lifts the “present” share only to 19.1%, 14.2% and 9.9%.
- 27%
of Google’s 179 results were designers’ own sites (about 2.5 per search, none for three searches). Retailers and galleries held 27%, editorial 23%, large or fashion brands 15%. Of the 49 designer results, 29 were collection or category pages and 11 homepages.
- 18
designers with a top-10 own site were never named by any platform. Overall, 33 of the 49 designer-site results were for designers Gemini did not name for that question (41 for ChatGPT, 44 for Perplexity).
- 82.6%
of Perplexity’s designer mentions cited another source; only 8.0% cited the designer’s own site. 48% of its cited domains were editorial, and 12.8% also ranked in Google’s top 10 for the matching query.
- 15–42%
of designer recommendations for a given search appeared in all three identical runs (Gemini 14.7%, ChatGPT 17.6%, Perplexity 41.5%). Platforms agreed with each other on 13–20% of designers.
- 0%
overlap for sculptural and artisan jewelry on all three platforms; alternative engagement and wedding rings overlapped most (Gemini 17.4%, ChatGPT 11.8%, Perplexity 5.6%).
How we ran the research
We wrote 20 non-branded questions a shopper might ask, four in each of five categories: alternative engagement and wedding rings, everyday solid gold, colored gemstone jewelry, sculptural and artisan jewelry, and ethical and meaningful jewelry. Each has a keyword-style Google equivalent and none names a brand. They were written before any data was collected and are not drawn from observed shopper behavior.
| Source | What we captured | When |
|---|---|---|
Organic results (7–10 per query, 179 in total) for 20 queries on google.com with gl=us, hl=en, pws=0. Ads excluded. One capture per query. | Oct 23, 2026 | |
| Gemini | 60 runs: 20 questions × 3, each in a fresh Temporary Chat (Gemini Auto) | Oct 23, 2026 |
| ChatGPT | 60 runs: 20 questions × 3 in Temporary Chat (Instant, Plus). 12 runs before its memory setting was switched off, 48 after. | Oct 23–24, 2026 |
| Perplexity | 60 runs: 20 questions × 3 in default Search mode (Free plan) | Oct 24, 2026 |
We read each answer from the page, recorded every brand named in order, and coded each brand and Google result by type. Three counting units appear and are not interchangeable:
- Responses (runs): 180 AI answers, 60 per platform.
- Mentions: every time a brand is named. Independent designers were named 302 times by Gemini, 337 by ChatGPT and 362 by Perplexity, among 117, 116 and 134 unique designers.
- Designer–question pairs: a designer counted once per question and platform, however many of the three runs named them (204, 204 and 171). All overlap percentages use pairs.
For “Where can I find unique alternative engagement rings from independent designers?”, Gemini named Sofia Kaman, whose site ranked third in Google’s top 10 for the matching search: that pair is present. Gemini also named Jennie Kwon Designs in all three runs, and her site was not among Google’s results for that search: that pair is absent, even though she may rank for other searches or appear on retailer pages that do. The 92–97% is the share of pairs like the second.
Checking the labels. “Independent” means a designer-led, founder-led or family-run label that sells through its own site, as we coded it. We then checked public sources for 52 designers: everyone behind a Google own-site result, every matched pair, and a further sample of frequently named designers. Seven Google results were reclassified (three scaled or non-fine brands, four retailers or galleries), and the 27% figure reflects those corrections. Eight more brands were borderline or unverified; we kept them and tested what happens if they are removed. About 220 other AI-named designers were not individually verified.
| Scenario | Designers’ own sites in Google’s 179 results | Own site not in matching top 10 (Gemini / ChatGPT / Perplexity) |
|---|---|---|
| As published (7 corrections applied) | 49 (27%) | 92.2% / 96.1% / 97.1% |
| Also remove the 8 borderline or unverified brands | 40 (22%) | 92.6% / 96.0% / 97.1% |
| Also remove 9 well-known scaled or multi-store names on the AI-named list, and Jean Dousset (analyst screen, not verified) | n/a | 92.8% / 96.5% / 97.5% |
Two shortlists, little overlap
Chart 1 is the central result. Gemini named 117 independent designers, and 16 of its 204 designer recommendations across the tested searches had the designer’s own site in Google’s top 10 for that same search. ChatGPT matched 8 of 204 and Perplexity 5 of 171, the same five in all three runs. The gap is not an artifact of the matching rule: letting a designer’s site count if it appeared for any of the 20 searches raises the share only to 10–19%. The assistants also named many who were not independent designers: large or scaled brands made up 10–13% of mentions and retailers 20% on Gemini and 16% on Perplexity.
AI-recommended independent designers whose own website is not in Google's top 10
Independent designers each platform named (counted once per question), by whether the designer's own website appeared in the top 10 organic results for the matching Google query.
92.2% absent (188 of 204). In top 10: 16 (7.8%). Excluding ARTËMER: 93.0% absent.
96.1% absent (196 of 204). In top 10: 8 (3.9%). Excluding ARTËMER: 97.0% absent.
97.1% absent (166 of 171). In top 10: 5 (2.9%). Excluding ARTËMER: 97.1% absent.
- Designer's own website in the matching Google top 10
- Not in the matching top 10
What Google actually shows
Designers’ own sites held about 2.5 slots per search, and none for “best independent brands everyday solid gold jewelry”, “emerging fine jewelry designers to know” or “independent fine jewelry anniversary gift”. The mix shifts by category: designer sites were 18 of 37 results for alternative engagement and wedding rings, but 7 of 34 for everyday solid gold and 6 of 37 for sculptural and artisan jewelry, where editorial pages took 15.
What Google ranks: organic results by source category
Share of all 179 organic results across the 20 matched queries (ads excluded). Bars are drawn on a 0–35% scale.
What the assistants do
Perplexity leans on other people’s pages
Perplexity shows a source chip beside each statement. Of 362 independent-designer mentions, 299 (82.6%) cited another source (an editorial piece, retailer or gallery), 29 (8.0%) the designer’s own site, and 34 (9.4%) showed no chip. Its cited domains were mostly editorial (71 of 148 domain–question pairs) or retailers and galleries (45); 19, including National Jeweler, The Wed and Vogue, also ranked in Google’s top 10, though mostly on different pages. We did not open the full source panels or measure whether Google’s third-party results mention the designers Perplexity named. ChatGPT linked 61.5% of the brands it named (usually a shop button); Gemini showed no links.
Where Perplexity gets its independent-designer recommendations
Which kind of source Perplexity cited beside each independent designer it named, and what kinds of domains it cited overall. Every mention is accounted for.
- Another source cited (editorial, retailer, gallery, marketplace)299 · 82.6%
- Designer's own site cited29 · 8.0%
- No citation chip shown34 · 9.4%
19 of 148 cited domains (12.8%) also ranked in Google's top 10 for the matching query.
The platforms disagree with each other, and with themselves
For the same question, Gemini and ChatGPT shared 20.4% of the designers either named, ChatGPT and Perplexity 16.1%, Gemini and Perplexity 12.7%. Run to run, Perplexity was the steadiest (41.5% of pairs in all three runs, against 17.6% for ChatGPT and 14.7% for Gemini). Gemini’s details also shifted: one brand was given different founder names. We measured who is named, not whether the descriptions are accurate.
How much the AI platforms' shortlists overlap
Overlap (Jaccard index) of the independent designers two platforms named for the same question, computed per question (union of 3 runs) and averaged over the 20 questions. 100% would mean identical lists.
| Gemini | ChatGPT | Perplexity | |
|---|---|---|---|
| Gemini | — | 20.4%avg. of 20 questions | 12.7%avg. of 20 questions |
| ChatGPT | 20.4%avg. of 20 questions | — | 16.1%avg. of 20 questions |
| Perplexity | 12.7%avg. of 20 questions | 16.1%avg. of 20 questions | — |
How consistently each AI platform repeats the same independent designer
Each independent designer named for a question, by how many of that platform's 3 identical runs (fresh chats) included them.
All 3 runs: 30 · 2 of 3: 38 · only 1 of 3: 136
All 3 runs: 36 · 2 of 3: 60 · only 1 of 3: 108
All 3 runs: 71 · 2 of 3: 49 · only 1 of 3: 51
- All 3 runs
- 2 of 3 runs
- 1 of 3 runs
Category changes the picture
Alternative engagement and wedding rings is where AI and Google agreed most, and where designers’ own sites held the most Google slots. Sculptural and artisan jewelry never overlapped: Google’s results there were mostly editorial roundups, while the assistants named designers and galleries Google did not rank. Each bar rests on 25–51 pairs, so read the pattern, not the exact percentages. ChatGPT’s engagement-ring figure comes from the 12 runs collected before its memory setting was switched off.
Where AI and Google overlap most: category differences
Share of independent designers each platform named whose own website was in the matching Google top 10, by category. Bars are drawn on a 0–20% scale.
One real overlap: Jean Dousset
For “lab grown diamond fine jewelry designer brand”, Jean Dousset’s lab-grown diamond collection ranked first on Google, and the matching question produced the brand in all three Gemini runs, all three Perplexity runs and one of three ChatGPT runs. It is a founder-led label, though Renaissance Global Limited, a Mumbai-based jewelry manufacturer, bought a 38.75% stake in January 2025. Jessica Biales, on a signet-ring search, also ranked first and was named repeatedly. Two designers in a 20-question test, one Google capture each: an illustration of the pattern, not a rule.
ARTËMER: a research-context check, and what changes without it
Found & Form has independently researched ARTËMER and has been in direct contact with the brand. ARTËMER is not a Found & Form client or advisory partner. It is one of the designers in this data. We coded it under the same rules as every other designer and did not treat it differently in any figure. But the browser and accounts used for collection may be influenced by prior research or browsing activity, so its appearances could be inflated, and we tested for that. Its own site ranked first or second on Google for two engagement-ring searches, and it was named in 7 of 60 Gemini runs, 10 of 60 ChatGPT runs (including huggie-earring and sapphire-ring answers after ChatGPT’s memory setting was off) and 2 of 60 Perplexity runs.
| Platform | All designers: pairs / absent from top 10 | Without ARTËMER: pairs / absent |
|---|---|---|
| Gemini | 204 / 92.2% | 201 / 93.0% |
| ChatGPT | 204 / 96.1% | 199 / 97.0% |
| Perplexity | 171 / 97.1% | 170 / 97.1% |
The headline does not depend on one brand. The category view does move: in alternative engagement and wedding rings, the match share falls from 17.4% to 14.0% on Gemini and from 11.8% to 8.3% on ChatGPT without ARTËMER (Perplexity is unchanged at 5.6%), and it remains the highest-overlap category either way. Other brands in the data, such as WWAKE, have been the subject of earlier Found & Form Discovery Studies; they were coded and counted like all the others. We also coded designers as established or lesser-known by judgment, with no external popularity data behind the labels, so we draw no conclusion about whether being well known helps in either channel.
What independent brands can do differently for Google and for AI
Each suggestion follows from something we measured. They are hypotheses to test on your own brand, not proven ways to rank or be recommended, and nothing here shows that either channel produces sales.
| What we saw | What to check or try |
|---|---|
| Google: build a page for the question, not just the brand. 29 of the 49 designer-site results Google ranked were collection or category pages (for example a lab-grown diamond collection or a non-traditional engagement rings collection); 11 were homepages. | For two or three questions your customers actually ask, check whether you have a collection page named in their words, and which pages Search Console shows ranking for them. We saw the pattern in ranking sites; we did not test that changing a page moves its rank. |
| Google: know who holds the slots in your category. Designers’ own sites were 27% of results overall (18 of 37 for engagement rings, 7 of 34 for everyday solid gold, none for three searches). Retailers, galleries and editorial pages took 50%. | Search your five most important questions and tally who holds each of the top 10. Where designers hold few slots, the realistic route may be the retailer or editorial pages that do. We did not test whether being listed on them helps your own rank. |
| AI: look at how other pages describe you. Perplexity cited a source other than the designer’s own site for 82.6% of mentions; 48% of the domains it cited were editorial publishers and 30% were retailers or galleries. | List the retailer, gallery and editorial pages that mention you. Check that they state what you make, your materials and your location accurately, and ask for corrections where they do not. We saw Perplexity lean on these pages; we did not test whether better coverage earns a recommendation. |
| AI: do not treat a Google rank as an AI result. Only 3–8% of AI-recommended designers had their own websites in Google’s top 10 for the same search, and 18 designers with a top-10 Google site were never named by any platform. | Track Google positions and AI mentions in separate columns of the same sheet, and do not report one as evidence of the other. |
| AI: check three times and record how you are described. Two-thirds of Gemini’s and about half of ChatGPT’s pairs appeared in only one run, and Gemini’s founder and sourcing details changed between runs. | Ask each question three times in fresh chats, monthly. Log whether you appear, where, and the wording used about you. One answer proves little, and an error in how you are described is worth correcting at its source. |
- Write three questions a customer might ask, without your brand name.
- Search each on Google and note who holds the top 10: your site, retailers, editorial, others.
- Ask ChatGPT, Gemini and Perplexity each question in fresh chats, three times each; log whether you appear and how you are described.
- List the third-party pages that name you; fix inaccuracies at the source.
Keep the two channels apart in your reporting. Search Console shows how Google treats your pages. AI visibility needs its own log. Neither number tells you about sales.
Limitations and what we did not measure
This is an exploratory study of 20 questions. It is not a representative study of US consumer behavior, and it does not show that AI recommendations or Google rankings lead to sales.
- Location and accounts. Testing used English-language prompts in signed-in accounts from a location outside the United States. Google was requested with US results (
gl=us), but geographic and account-level personalization cannot be ruled out. ChatGPT returned some location-specific recommendations even after its memory setting was switched off. Perplexity was also tested in a signed-in Free account, so its results may likewise reflect personalization. Results for a shopper in the US or another market may differ, and a US-located, logged-out replication is the main follow-up. - ChatGPT ran in two conditions. The 12 engagement-ring runs (A1–A4) were collected before its memory setting was switched off; the other 48 after. One extra answer collected in between, with memory on, drew on personal context and was excluded, and a logged-out attempt hit a sign-in wall and was not used.
- One Google capture per query. Google’s results varied between loads in a test, so individual ranks are tentative.
- Prior research and contact. Found & Form has researched ARTËMER and communicated with the brand; it is not a client or advisory partner. ARTËMER is in the data and in every headline figure; the sensitivity table above shows the results without it.
- Analyst-coded labels. Whether a brand is an independent designer, and the type of each Google result, were coded by one person. We verified 52 designers from public sources and corrected seven Google results; about 220 AI-named designers were not individually checked. Ownership can change.
- Prompts and models. The questions are analyst-written. The platforms show interface labels only (Gemini Auto, ChatGPT Instant), so exact model versions are unknown. We condensed the tails of some later raw answers; ordered brand lists and links are complete.
- Not measured: brand mentions inside third-party Google results, Google AI Overviews and People Also Ask, full Perplexity source panels, ChatGPT citation chips, statistical significance, answer accuracy, clicks and sales.
Conclusion
Being findable used to mean ranking. For an independent jewelry brand it now means two separate jobs, and success at one says little about the other. Google, in our test, gave about a quarter of its results to designers’ own sites and the rest to retailers, magazines and large brands. The AI assistants assembled their own lists, differently each time they were asked, and Perplexity, the one that shows its sources, drew mostly on other people’s pages.
That points to a habit more than a tactic: look at both shortlists for the questions your customers ask, look at who describes your work in the places both of them read, and check again next week. What a US shopper sees, whether an AI mention changes what someone buys, and how stable any of this is over months are the questions Found & Form will test next.
Research package and sources
Data and methods: the prompts, 179 Google results, 1,464 AI brand mentions, brand-level matching, raw answer text, coding scripts, the independence verification and a script that recomputes every headline figure are in research/insights/ai-recommendations-vs-google-rankings-2026/ in the Found & Form repository.
Disclosure: Found & Form has previously researched and communicated with ARTËMER, a brand in this data. ARTËMER is not a client or advisory partner. All results are descriptive and non-causal. Related Found & Form research: why AI recommends some jewelry designers, same question, different jewelry brands and the ChatGPT independent-jewelry test.