When clients ask AI for a Swiss private bank, who gets named?
We put 48 real buyer questions to ChatGPT, Gemini and Perplexity and checked which of 23 public Swiss private banks each one named. The answers diverge sharply by engine — and one whole category of questions goes to almost no bank at all. Run the check for your firm →
Hover or tap a dot for the question.
Why we ran this
A growing share of “which private bank should I use” research now begins with an AI assistant. The assistant assembles its answer from a handful of sources — a bank’s own site, the regulator, comparison platforms, forums — and it either names specific institutions or it does not. This benchmark reads which Swiss private banks currently surface, engine by engine, and where the openings are.
How we measured it
We wrote 48 questions in the language a prospective client, adviser or journalist actually uses, across six topic clusters (all 48 are explorable in full below). Each question was sampled once per engine, and every answer was checked for a mention of any of 23 public, Switzerland-active private banks — UBS, Julius Bär, Pictet, Lombard Odier and the mid-tier, the digital-asset banks Sygnum and AMINA, and the big international private-banking arms (HSBC, Deutsche Bank, Citi, Barclays, BNP Paribas, Edmond de Rothschild) — using the same name matcher a live Inserviss scan uses.
The tracked list covers the 23 names a prospective client would shortlist, whether or not any engine named them. Seven were added after the first run and re-matched against the same stored answers — no new sampling — so every figure rests on identical evidence. Credit Suisse is included on purpose: it was absorbed by UBS in 2023, and whether AI still names it is itself a result.
A single pass per engine establishes direction and priority; it is not a trend line, and category rollups (a bank named in k of 48 answers) carry more weight than any single number. Citations are reported as measured; query volumes are labelled estimates, shown as ranges. Engines are never blended. More in about this study.
What we found, at a glance
- The engine changes the answer. Gemini named a tracked bank in 42 of 48 answers, ChatGPT in 33, Perplexity in 32.
- Same firm, very different visibility. Julius Bär was named 17 / 30 / 8 times (ChatGPT / Gemini / Perplexity).
- ChatGPT is top-heavy. UBS dominates its answers; 9 tracked banks get zero mentions.
- AI still banks with Credit Suisse. Wound into UBS in 2023, it is still named 20 times across the 144 answers — fifth of 23, ahead of every mid-tier Swiss name and every international private-banking arm.
- The global banks barely register. HSBC, Deutsche Bank, Citi, Barclays, BNP Paribas and Edmond de Rothschild together drew fewer mentions than Pictet alone; only Barclays Private Bank drew zero on every engine.
- One cluster is soft on every engine: regulatory & tax — cross-border eligibility, lump-sum taxation, US-person questions — 3 / 4 / 3 out of 8.
- 5 questions were answered without a single tracked bank on any engine — the clearest openings.
Does the engine you ask change whether you get named?
Yes, materially. Below is the share of all 48 answers that named at least one tracked bank, per engine — then the full table, every bank against every engine, sortable by any column.
Share of 48 answers naming at least one of the 23 tracked banks. Per engine, never blended.
| Bank | ChatGPT | Gemini | Perplexity |
|---|---|---|---|
| UBS | 26 | 34 | 22 |
| Pictet | 14 | 34 | 13 |
| Julius Baer | 17 | 30 | 8 |
| Lombard Odier | 10 | 33 | 10 |
| Credit Suisse | 4 | 13 | 3 |
| Union Bancaire Privée | 2 | 10 | 6 |
| Vontobel | 4 | 9 | 2 |
| Sygnumdigital | 1 | 8 | 6 |
| LGT | 0 | 11 | 3 |
| AMINA Bankdigital | 2 | 7 | 5 |
| EFG International | 0 | 8 | 3 |
| Mirabaud | 0 | 7 | 4 |
| Citi Private Bank | 2 | 5 | 4 |
| J. Safra Sarasin | 0 | 7 | 2 |
| Bordier & Cie | 1 | 7 | 0 |
| Rothschild & Co | 1 | 4 | 1 |
| REYL Intesa Sanpaolo | 0 | 5 | 0 |
| Edmond de Rothschild | 0 | 3 | 2 |
| HSBC Private Bank (Suisse) | 1 | 1 | 3 |
| Deutsche Bank | 0 | 3 | 2 |
| Bank Syz | 1 | 0 | 3 |
| BNP Paribas Wealth Management | 0 | 2 | 1 |
| Barclays Private Bank | 0 | 0 | 0 |
Mentions across 48 answers per engine. Select a column heading to re-sort. Scroll the table sideways on a narrow screen.
Download this chart as an image →
The cluster nobody owns
Coverage varies by question type. Below is the weakest engine’s coverage for each cluster, out of 8 questions. Five clusters are well covered on at least one engine. Regulatory & tax is thin on all three — the cross-border, lump-sum-taxation and US-person questions that decide whether a relationship can even open go to generic advice, the regulator, or a law firm.
Answers naming a bank on the weakest engine, per cluster. Amber = soft on all three engines.
How this is estimated
Estimate — a range, not a keyword-tool export. "Swiss private bank for a German-resident client", "lump-sum taxation Switzerland private bank", "US person Swiss account FATCA", "which Swiss banks still take EU residents", "Swiss deposit protection limit". Highest-intent, most cross-border. Head contribution ~300–800/mo; ~7 long-tail phrases at ~30–100/mo; ×2–4.
| Cluster | Coverage (worst engine) | Est. questions / mo |
|---|---|---|
| Liquidity Events & Onboarding | 4/8 | 1,100–3,600 |
| Investments & Advisory | 4/8 | 1,400–4,800 |
| Custody & Digital Assets | 6/8 | 800–3,000 |
| Wealth Structuring | 7/8 | 1,000–3,400 |
| ●Regulatory & Tax | 3/8 | 900–3,200 |
| Lending & Real Estate | 5/8 | 700–2,600 |
Volumes are an estimate, shown as a range. Illustrative estimate, not a keyword-tool export. Each cluster has ~8 seed buyer questions; long-tail intent phrases in this category run ~20–120 monthly searches each across Switzerland plus cross-border DE / UK / US-expat demand, one or two head terms add ~300–900/month, and an AI-assistant reformulation multiplier of ×2–4 accounts for the wider phrasing people use with an assistant. Ranges are rounded and widened for uncertainty. For a specific firm, the real number comes from the check, not this table. Under-owned clusters total roughly 900–3,200 questions/month.
The check runs this same 48-question universe against your own name, per engine, and returns your AI Visibility Score plus the exact questions where a competitor — or no one — is being named instead.
Check your firm’s AI visibility →Start here: the questions nobody owns
5 of the 48 questions were answered without a single tracked bank — on any engine. There is no incumbent to out-rank and no competitor to displace. These are the lightest-lift openings: publish the clearest, best-structured answer and a firm can become the one AI cites.
A firm doesn’t appear in these answers because no bank does — not because a rival is ahead. The full set, and the near-misses where only one engine names a bank, are in the explorer below.
The 5 questions no tracked bank owns on any engine, with the sources AI cites instead — a ready-made content brief.
One email. No list-selling. Unsubscribe any time.
Explore the 48 questions
This is the part a bank can act on directly: every question, and exactly what each engine did with it — which of your peers it named, and which sources it pulled from. Filter by cluster, search for your own name, or show only the questions no bank has claimed.
Where each engine gets its answers
The domains each engine actually pulled from. ChatGPT leans on bank-owned sites and the regulator; Perplexity’s top sources are SEO aggregators and Reddit, ahead of any bank’s own site; Gemini grounds on far fewer pages, mostly moneyland.ch and a long tail of individual Swiss bank sites.
Times a domain was cited across the 48 answers. Blue = a tracked bank’s own site. Gemini’s list is short because many of its answers ground on no sources at all.
About this study
- Single pass per engine. One sample per question per engine establishes direction; it does not average out run-to-run answer variance. Two runs a day apart moved Julius Bär from 11 / 31 / 18 to 17 / 30 / 8 — the direction held, the exact counts did not.
- Snapshot in time. Collected September 9, 2026. AI engines change their answers week to week; this is not a trend line.
- Conservative name matching. Short one-token names (UBS, LGT, Syz, Vontobel…) require an exact whole-word match, so a slight undercount versus a human reader is possible.
- Three engines. Microsoft Copilot is added next; it draws on Bing, and enterprise buyers use it heavily. The measurement framework is identical when it joins.
How to cite this study
To request the underlying data or discuss a firm-specific analysis, contact igor@inserviss.app. The methodology matches Inserviss’s own product — the same questions, the same per-engine measurement.
The same measurement, run against your name — per engine, question by question — with your score and the gaps a competitor is filling.
Go to the Private Banking page →