A comparative benchmark of how brands and organisations appear in AI-generated answers across discovery and evaluation scenarios in Europe and Bangkok.
Edition 2 records which organisations surface on discovery, evaluation and comparison queries. The core European panel uses ChatGPT, Gemini and Perplexity. Paris & Lyon law uses Gemini, Meta AI and Perplexity; Bangkok schools use the core panel; Bangkok health adds Meta AI to the three core engines. We queried these engines across 394 queries, extracted 5,347 citations covering 1,057 players and 8 sectors, from Europe to Bangkok, then computed a proprietary score: the KVS.
Mistral AI, Doctolib and Revolut reach the normalised maximum of their sector panel. Their shared observed feature is presence across every engine used for that sector.
Across the European queries in this sample, 83% of the 30 highest-ranked players are European. This describes the measured panel; it is not a global market-share estimate.
91% of the top 40 is cited by all three engines in the European panel. Because KVS itself includes panel coverage, this is both an observed pattern and a property of the index.
Paris accounts for 40% of the SaaS leaders observed in the panel (Mistral, Doctolib, Contentsquare, Dataiku). London is strongly represented in fintech through Revolut, Wise and Checkout.com.
Several companies valued at more than €500M do not appear on any query in the European panel. The Barometer records this gap without assigning it a single cause.
Some brands with strong conventional search visibility are less present in the AI answers measured here. The study records that difference; it does not isolate the factors that caused it.
Bredin Prat reaches 100 versus 64 for Gide in this panel. The Anglo-Saxon firms studied are substantially less visible on the French-language queries.
SeLoger and Meilleurs Agents capture 71% of citations in the panel. Local agencies are scarcely represented on the measured “best agency in [city]” subset.
Doctolib records 21 rank-1 mentions out of 25. The rest of the panel separates the “user” layer (Doctolib, Kry, Qare) more clearly from the “market” layer (hospitals and clinical-AI tools).
Howard Avocats appears in 16 of 24 Paris employment-law answers. But 64% of the 232 firms cited appear only once, while price-related queries show low attribution to named firms.
NIST, Bangkok Patana and ISB account for most first-place positions in the panel. A data-quality anomaly also appears: Perplexity cites a non-existent “Dulwich College Bangkok” 21 times in the measured answers.
Bumrungrad records 64 first-place positions across the 83-query, four-engine panel. Specialised procedures have different leaders, while seven queries cite no clinic.
Extract from each sector ranking. KVS combines mention rank, frequency and panel coverage, then normalises the result to 100 within each sector. Scores from different sectors are not directly comparable.
| Sector | Cited leader | KVS | Followed by |
|---|---|---|---|
| SaaS / Tech | Mistral AI | 100 | Doctolib · Contentsquare · Dataiku |
| Finance / Fintech | Revolut | 100 | Adyen (96.0) · Stripe · Wise |
| Law · France (business) | Bredin Prat | 100 | Gide (64) · Magic Circle (<18) |
| Real estate | SeLoger | 100 | Meilleurs Agents (87.6) · Idealista · Notaires de France |
| Health · Europe | Doctolib | 100 | Kry · Qare · hospitals · clinical AI tools |
| Law · Paris & Lyon | Howard Avocats | 100 | Facchini (Lyon, 100) · Delsart (86.0) · Hoche (44.3) |
| International schools · Bangkok | NIST | 100 | Bangkok Patana (85.7) · ISB (82.0) · Shrewsbury (51.0) |
| Health & aesthetics · Bangkok | Bumrungrad | 100 | Yanhee (79.0) · Kamol (78.8) · Wansiri (59.8) |
Full ranking of the top 40 brands per sector, with the AI-by-AI breakdown, in the PDF report.
Across the European SaaS/Tech queries in this panel, six of the ten highest-ranked players are French.
| # | Company | Country | KVS |
|---|---|---|---|
| 1 | Mistral AI | 🇫🇷 France | 100.0 |
| 2 | Doctolib | 🇫🇷 France | 90.8 |
| 3 | Odoo | 🇧🇪 Belgium | 74.2 |
| 4 | Contentsquare | 🇫🇷 France | 72.9 |
| 5 | Qonto | 🇫🇷 France | 66.8 |
| 6 | Dataiku | 🇫🇷 France | 65.1 |
Stripe, Typeform and Pipedrive — global SEO champions — cap between KVS 41 and 49 on European queries. The observed ranking differs materially from the conventional SEO hierarchy; the Barometer does not identify a single source or mechanism as the cause.
Every company above KVS 40 is cited by ChatGPT, Gemini and Perplexity. Since KVS rewards panel coverage, multi-engine presence contributes directly to the score; the observation should not be treated as proof of causality.
The head captures the essentials: beyond the top 12, dozens of players earn a single mention and vanish. A one-off mention is less stable than repeated presence across the panel.
The panel shows strong representation of French players among SaaS/Tech leaders. For a challenger, a more realistic hypothesis to test is a narrower query territory — by use case, workflow or category — followed by measurement of whether clearer entity signals, expert content and coherent third-party sources improve presence across multiple engines.
European fintech is polarised by the City. Adyen holds the Dutch seat; Stripe runs its European operations from Dublin.
| # | Company | Country | KVS |
|---|---|---|---|
| 1 | Revolut | 🇬🇧 United Kingdom | 100.0 |
| 2 | Adyen | 🇳🇱 Netherlands | 96.0 |
| 3 | Stripe | 🇮🇪 Ireland | 83.2 |
| 4 | Wise | 🇬🇧 United Kingdom | 75.4 |
| 5 | Qonto | 🇫🇷 France | 69.4 |
| 6 | Pennylane | 🇫🇷 France | 42.1 |
Revolut (100), Adyen (96.0), Stripe (83.2), Wise (75.4), Qonto (69.4) — number 6, Pennylane, already drops to 42.1, and the curve keeps falling.
Stripe takes rank 1 more often than anyone (17 times) yet stays third. Revolut is cited 41 times across the panel: being present everywhere weighs more than winning somewhere.
258 unique companies cited for 40 FR + EN queries. Hundreds of players receive a single mention; frequency and three-engine coverage both contribute to the score.
The panel surfaces several European hubs beyond London and Paris, including Amsterdam, Stockholm and Barcelona. For a fintech outside the top 5, one hypothesis to test is a narrower financial job-to-be-done — expense management, treasury or financing — and then track visibility at a constant perimeter rather than immediately targeting the broadest category queries.
Across 30 queries in two layers — 15 “consumer” and 15 “market” — a single firm ranks first across both. The Anglo-Saxon giants cap out on French-language queries.
| # | Company | Country | KVS |
|---|---|---|---|
| 1 | Bredin Prat | 🇫🇷 France | 100.0 |
| 2 | Gide Loyrette Nouel | 🇫🇷 France | 64.0 |
| 3 | Darrois Villey Maillot Brochier | 🇫🇷 France | 31.8 |
| 4 | Freshfields | 🇬🇧 United Kingdom | 29.9 |
| 5 | Captain Contrat (legaltech) | 🇫🇷 France | 20.5 |
Bredin Prat (100) leads Gide Loyrette Nouel (64.0) by 36 points, with Darrois Villey at 31.8. The fifth most-cited player is a legaltech — Captain Contrat (20.5).
Linklaters (17.8) and Clifford Chance (14.6) remain below 18 in this French-language panel despite strong offline and directory recognition. The Barometer does not identify the retrieval or source mix responsible for that gap.
Firms present across both the consumer and market layers record the highest scores in this sample. Firms visible on only one layer are concentrated lower in the ranking.
The fragmentation of the ranking suggests testing narrower territories by practice, industry or matter type. A defensible experiment can combine signed expertise content, verifiable credentials, structured data appropriate to the content and relevant third-party sources, then remeasure at a constant protocol. The Barometer does not support a promised KVS level or timeline.
SeLoger and Meilleurs Agents (AVIV group) together take nearly three quarters of AI mentions in French real estate. No brick-and-mortar agency appears before rank 14, with a KVS below 8.
| # | Company | Country | KVS |
|---|---|---|---|
| 1 | SeLoger | 🇫🇷 France | 100.0 |
| 2 | Meilleurs Agents | 🇫🇷 France | 87.6 |
| 3 | Idealista | 🇪🇸 Spain | 44.3 |
| 4 | Notaires de France | 🇫🇷 France | 43.1 |
| 5 | Rightmove | 🇬🇧 United Kingdom | 27.1 |
SeLoger (100) and Meilleurs Agents (87.6) — same owner — absorb nearly three quarters of mentions. No physical agency or network before rank 14, below KVS 8.
On the measured “best agency in [city]” queries, national platforms appear more often than local agencies. The Barometer records the gap but does not establish whether content, authority, retrieval or another factor causes it.
On REIT, crowdfunding and yield queries, different leaders emerge: Wemo One (19.8), Iroko Atlas (18.0) and EstateGuru (12.1), all absent from the consumer top 5.
Local agencies are under-represented in the geographic queries measured here. One hypothesis to test is stronger factual local pages, structured data appropriate to the activity, consistent profiles and independent local sources, followed by before/after measurement. These observations do not support a guaranteed KVS level or timeline.
With 21 rank-1 appearances out of 25, Doctolib stands out across both consumer and professional health-query layers. Below it, the actors surfaced in those two layers differ substantially.
| # | Company | Country | KVS |
|---|---|---|---|
| 1 | Doctolib | 🇫🇷 France | 100.0 |
| 2 | Kry / Livi | 🇸🇪 Sweden | 30.7 |
| 3 | Qare | 🇫🇷 France | 26.9 |
| 4 | Maiia | 🇫🇷 France | 14.5 |
| 5 | Medadom | 🇫🇷 France | 8.4 |
Doctolib records 21 rank-1 appearances out of 25 measured health answers and appears across both the consumer and professional layers.
User layer: Kry/Livi (30.7), Qare (26.9), Maiia (14.5). Market layer: Karolinska (21.0), DAX Copilot (15.6), Charité Berlin (15.0), Nabla (9.5). Almost no company in both.
Nabla, Owkin and Corti each appear more strongly on one or two engines than on the full panel. This dispersion reinforces the value of tracking the same query set across multiple engines.
The results separate several demand layers: consumer teleconsultation, clinical tools and healthcare providers. In a regulated sector, improvement hypotheses should be tested with medically reviewed content, properly sourced evidence and certifications, relevant structured data and strict editorial governance. The Barometer does not support a fixed progression timeline.
80 individual-client queries — 5 practice areas, 2 cities — put to Gemini, Meta AI and Perplexity: 640 citations across 232 firms. A firm is named in 75% of answers. The sector's problem isn't absence — it's volatility and concentration.
| # | Paris | KVS | Lyon | KVS |
|---|---|---|---|---|
| 1 | Howard Avocats — employment law | 100.0 | Facchini Avocat — real estate | 100.0 |
| 2 | Hoche Avocats — business | 44.3 | Delsart Avocats — IP | 86.0 |
| 3 | Kohen Avocats — employment | 39.1 | Bressand Avocat — IP | 81.3 |
| 4 | Cabinet Zenou — employment | 38.7 | Cabinet Bouchara — IP | 74.9 |
| 5 | Avi Bitton — employment | 37.2 | Erovic Avocats — employment | 57.1 |
KVS normalised to 100 per city, against the local leader.
When the question turns to price, named-firm citation falls to 17% in this panel, and Perplexity names no firm on the measured price questions.
232 firms are named, but most of the long tail appears only once. The data therefore distinguishes isolated mentions from repeated presence without attributing that difference to a specific tactic.
Gemini names 3.1 firms per answer versus 1.3 for Meta AI. In Lyon, referrals to the Bar appear 33 times versus 17 in Paris, illustrating how attribution patterns vary by geography and engine.
Howard in Paris employment law, Delsart in IP and Facchini in Lyon real estate illustrate strong concentration on particular practice × city combinations. Elsewhere the field remains fragmented. This supports measuring query territories separately rather than inferring broad visibility from a single ranking.
86 queries from the parental decision journey — discovery, curriculum, budget, SEN, relocation — put to ChatGPT, Gemini and Perplexity: 1,769 citations across some sixty schools, 10 intent categories.
| # | School | Rank 1 | KVS |
|---|---|---|---|
| 1 | NIST International School | 68 | 100.0 |
| 2 | Bangkok Patana School | 28 | 85.7 |
| 3 | ISB — International School Bangkok | 51 | 82.0 |
| 4 | Shrewsbury International School | 16 | 51.0 |
| 5 | St Andrews International Schools | 18 | 40.9 |
| 6 | Harrow International School Bangkok | 10 | 31.5 |
NIST records 68 rank-1 citations and appears strongly in both the “parents” layer (discovery, relocation) and the “proof” layer (results, comparisons).
Several categories show clear leaders — IB, AP, SEN and boarding — while fee, neighbourhood and Thai-English bilingual queries are less consistently attributed to the largest schools.
Perplexity cites a “Dulwich College Bangkok” that doesn't exist 21 times — spread by SEO aggregators. And RIS racks up 76 mentions without a single rank 1: always cited, never recommended.
Budget, neighbourhood, bilingual and national-community queries are less concentrated than several generic queries in this panel. A school can test factual content around those needs, clearly maintained fees and admissions information, appropriate structured data, and consistency between official and third-party sources. King’s College’s gap between Gemini and ChatGPT mainly illustrates why measurement should cover more than one engine.
83 patient queries — generic, aesthetic, medical tourism, trust — put to four models: ChatGPT, Gemini, Perplexity and Meta AI. 1,068 citations across 47 players; 82 of 83 queries cite at least one clinic.
| # | Player | AIs | KVS |
|---|---|---|---|
| 1 | Bumrungrad International Hospital | 4/4 | 100.0 |
| 2 | Yanhee International Hospital | 4/4 | 79.0 |
| 3 | Kamol Cosmetic Hospital | 4/4 | 78.8 |
| 4 | Wansiri Hospital | 4/4 | 59.8 |
| 5 | ID Clinic Bangkok | 4/4 | 47.0 |
| 6 | Masterpiece Hospital | 3/4 | 28.1 |
17 of 47 players are cited by all four engines in the Bangkok health panel. Bumrungrad records 64 first-place positions across generic, trust and medical-tourism queries.
Masterpiece scores strongly on ChatGPT in the underlying grid but records no Gemini citation in this panel. The difference illustrates why engine-specific reporting matters.
Kamol ranks first on all four engines for gender-reassignment surgery. Conversely, seven measured queries on price, safety or comparison cite no clinic.
Hospitals are highly visible on generic queries, while some specialist clinics appear more strongly on specific procedures. Less-attributed queries can be tested with medically reviewed procedure pages, accurate safety and pricing information where publication is appropriate, relevant structured data and independent sources. Any optimisation must remain compliant with applicable medical and advertising rules.
35 international-patient questions — general, procedures, success rates, price, legal framework, source markets — put to ChatGPT, Gemini, Perplexity and Meta AI. Full grid: 140 answers coded and 34 clinics scored; no query has one stable leader across the full four-engine panel.
| # | Clinic | AI | KVS |
|---|---|---|---|
| 1 | Safe Fertility Center | 4/4 | 100.0 |
| 2 | Bumrungrad Fertility & IVF Clinic | 4/4 | 79.4 |
| 3 | Bangkok Hospital Fertility Center | 4/4 | 50.9 |
| 4 | Prime Fertility Center | 4/4 | 41.9 |
| 5 | Jetanin Institute | 4/4 | 41.3 |
| 6 | Superior A.R.T. | 3/4 | 25.4 |
On several procedure queries (ICSI, egg freezing, FET…), Prime Fertility ranks first in the measured answers. Procedure-level visibility therefore differs from the generic hospital ranking.
Across the 35-query fertility panel, no query has the same #1 on three engines or more. Rankings therefore vary materially by engine.
Across the five legal questions in this panel, two cite no clinic and some answers surface aggregators rather than providers. These results require careful legal interpretation and should not be treated as legal advice.
Safe Fertility and Bumrungrad are highly visible on generic queries, while several commercial and legal queries remain fragmented or mediated by aggregators. For a clinic, these topics require factual, medically and legally reviewed pages covering eligibility, logistics and destination comparisons. Legal information must be kept current and should never be inferred from the Barometer alone.
In the sample, high scores are strongly associated with presence across multiple engines. This relationship also follows from the KVS construction, which includes panel coverage, so it should be read as a property of the index as well as an observed market pattern.
A strategy measured on one engine alone cannot estimate coverage across the rest of the market. The differences observed between ChatGPT, Gemini and Perplexity support multi-engine monitoring at a constant scope. The geographic distributions shown here describe only the sectors and queries in this edition.
Several companies valued at over €500M appear in none of the 145 queries in the European panel. The Barometer cannot isolate a single cause. It does, however, identify several dimensions worth auditing and testing.
1. Public footprint. Check whether important expertise, offers and evidence exist in publicly accessible sources and are stated clearly enough to be retrieved.
2. Entity clarity. Check the consistency of the organisation name, profiles, executives, domains, offers and reference sources. Wikipedia is relevant only when an entity independently meets its notability requirements.
3. Retrieval infrastructure. Check accessibility, indexability, page clarity, internal linking, structured data appropriate to the content and how easily factual answers can be extracted.
1. Third-party authority. Test the effect of independent, editorially credible and sector-relevant sources rather than reasoning only in link volume.
2. Relevant structured data. Use Schema.org where it accurately describes the content and entity; markup can improve clarity but does not guarantee a citation.
3. Cross-source consistency. Reduce contradictions between the official website, professional profiles, media coverage and other relevant public sources.
4. Extractable content. Structure information so it answers real market questions clearly, without assuming that one page format or markup type will always be preferred.
The Barometer does not turn these dimensions into causal claims. They form an experimentation agenda: measure, change one identifiable lever, then measure again under the same protocol.
The scope, engines, repetition rules, limitations and calculation logic are documented so the index can be audited and compared over time under a constant protocol.
Questions typical of a decision-maker, a patient or a parent in the discovery phase, asked in a neutral session — no history, no personalisation. Protocol for edition 2: three runs per question per engine, to absorb the variability inherent to generative answers. Edition 1 (18 April 2026) used a single submission — the two editions are only comparable with that caveat in mind. Next measurement: 30 September 2026.
For each answer: players cited, rank in the list, mention context. 1,057 unique entities across 8 sectors, from Europe to Bangkok.
KVS combines mention position, frequency and coverage across the engines in the sector panel. The result is then normalised within each sector; it is a relative index, not market share.
KVS = Σ (CTRrank × weightAI) × (1 + coveragebonus)
Three signals are combined: mention rank, frequency across the query set and multi-engine coverage. Summing observed mentions incorporates frequency, after which the KVS coverage bonus is applied under the index rules. The result is normalised to 100 within each sector.
Traffic. The KVS measures citation, not the click through to your site.
Product quality. Only the models' perception is assessed.
The brand long tail. The scope is deliberately sector-based and reproducible.
The cause of a citation. Presence in an answer also depends on the engine queried, its retrieval layer and the context of the question — the KVS records an outcome, it does not isolate what produced it.
Core panel: ChatGPT, Gemini and Perplexity. Paris & Lyon law uses Gemini, Meta AI and Perplexity. Bangkok international schools use the core panel, while Bangkok health adds Meta AI. Cross-sector comparisons should remain descriptive when engine panels differ.
| Sector | Queries | Language | Country | Engines | Date |
|---|---|---|---|---|---|
| SaaS / Tech | 15 | FR / EN | Europe | ChatGPT · Gemini · Perplexity | 30 June 2026 |
| Finance / Fintech | 40 | FR (20) · EN (20) | Europe | ChatGPT · Gemini · Perplexity | 30 June 2026 |
| Law · business law | 30 | FR (15) · EN (15) | France · Europe | ChatGPT · Gemini · Perplexity | 30 May 2026 |
| Real estate | 30 | FR (15) · EN (15) | France · Europe | ChatGPT · Gemini · Perplexity | 30 May 2026 |
| Health · Europe | 30 | FR (15) · EN (15) | Europe | ChatGPT · Gemini · Perplexity | 30 April 2026 |
| Law · Paris & Lyon | 80 | FR | France | Gemini · Meta AI · Perplexity | 30 June 2026 |
| International schools · Bangkok | 86 | EN | Thailand | ChatGPT · Gemini · Perplexity | 30 June 2026 |
| Health & aesthetics · Bangkok | 83 | EN | Thailand | ChatGPT · Gemini · Perplexity · Meta AI | 30 June 2026 |
The five European sectors total 145 queries. The Bangkok Fertility & IVF study (35 questions, ChatGPT · Gemini · Perplexity · Meta AI, 15 July 2026) is published separately and sits outside the eight sectors above.
Edition 2 — June 2026 (current edition). 394 queries, 5,347 citations, 1,057 players, 8 sectors. Extends the European panel (145 queries) to the Bangkok studies and the Paris & Lyon law panel. Core panel: 3 engines; Paris & Lyon law uses Gemini, Meta AI and Perplexity; Bangkok health adds Meta AI to the three core engines. Protocol raised to 3 runs per question per engine. Next measurement: 30 September 2026.
Edition 1 — 18 April 2026, European panel only. 55 queries (SaaS/Tech and Finance/Fintech), 3 engines, a single submission per engine, 1,240 citations across 444 companies. Archived, not overwritten: view the archived edition 1
Edition 3 — upcoming. Panel widened to 5 engines (Gemini, Perplexity, ChatGPT, Meta AI, DeepSeek), varied session locations, and a minimum of three runs per question with the variance published. Part of any shift between editions will come from that panel change: comparisons only hold at an identical perimeter.
The sector rankings, engine-by-engine detail, Europe–Bangkok analyses, methodology, interpretation limits and strategic implications of edition 2. Free, in exchange for your work email.
Methodology, limits, and how to read — then improve — your position.
KVS is a proprietary index normalised to 100 that measures an organisation’s relative visibility across a defined set of queries and engines. It combines mention rank, frequency and panel coverage. In edition 2, the engine panel varies by sector between three and four engines; the score is normalised within each sector and should not be compared directly across different sectors.
Edition 2 was built in sector waves. The core panel uses ChatGPT, Gemini and Perplexity. The Paris–Lyon law panel replaces ChatGPT with Meta AI, while Bangkok health adds Meta AI to the three core engines. The scope table publishes the exact panel for every sector. Edition 3 is planned with a standardised five-engine panel.
The 394 queries were designed to represent sector-specific discovery, evaluation and comparison scenarios. Language, country, query count, engines and collection date are published in the scope table. Sessions are run without personal history and under a constant protocol within each panel.
No. KVS measures a different layer: whether a brand appears in AI-generated answers across a defined question set. It complements SEO, traffic, conversion and reputation metrics; it does not replace them.
It means only that no citation for your brand was observed within the tested scope and on the stated collection dates. It is not a judgement on company quality and does not prove invisibility across every possible query. A dedicated diagnostic can investigate technical, editorial, entity and authority factors associated with the gap, without guaranteeing a correction timeline.
No. The Barometer is independent of KAIS commercial relationships. Rankings are based on answers collected under the published protocol; a client cannot buy a position or have a competitor removed.
Factors to test include entity clarity, factual and extractable content, structured data appropriate to the content, consistency across public sources, third-party authority and coverage across multiple engines. The Barometer is observational: these factors should be measured before and after changes, and none guarantees a citation on its own.
Yes, with attribution: “Source: KAIS AI Search Visibility Barometer 2026, edition 2 — KaiZen AI Strategy” and a link to this page. When a figure is sector-specific, also state the sector and relevant engine panel.
Start with one free signal, then deepen the analysis with a Snapshot or Diagnostic built around your market, language and actual competitors.
Test one question for free View KAIS diagnostics