AI Visibility Benchmark — Luxembourg Fiduciary & Accounting Firms
When a prospective client asks ChatGPT or Perplexity for a Luxembourg fiduciary, which firms get named? We measured 405 AI answers to 80 questions across two engines and three languages, on 8 July 2026.
- 405
- AI answers analysed
- 80
- questions
- 2
- engines (ChatGPT, Perplexity)
- 3
- languages (FR, DE, EN)
- 65
- firms tracked (44 independent, 21 large/network)
Measured 8 July 2026 (7 answers re-measured 28 July 2026). 65 firms tracked: 44 independent Luxembourg fiduciaires and 21 large / network firms.
The two findings that matter
The language gap
Ask in French and independent Luxembourg firms compete; ask in English and the large networks take the answer. An English-speaking founder asking an AI for an accountant is more than twice as likely to be pointed at a Big Four firm than a French-speaking one — a 74% large-network hit rate in English against 42% in French. Only 14 distinct independents surfaced in English at all, against 30 in French.
The category is winnable by an independent
The single most-named firm in the entire benchmark is not a Big Four firm. It is an independent Luxembourg fiduciaire — Fiduciaire LPG, named in 66 of 405 answers, ahead of BDO (58), PwC (53), Deloitte (44) and KPMG (44). Its own website was cited as a source in 15% of all answers — more often than most national directories. A single firm’s own site can become a primary source for the whole category.
Headline finding
The same question, three languages, three different answers
The share of answers in each language that named at least one independent firm, at least one large / network firm, and the count of distinct independents that surfaced.
| Question language | Answers | An independent firm named | A large / network firm named | Distinct independents surfaced |
|---|---|---|---|---|
| French | 201 | 61% | 42% | 30 |
| German | 106 | 40% | 16% | 20 |
| English | 98 | 34% | 74% | 14 |
Plain-language reading: ask in French and independent Luxembourg firms compete; ask in English and the large networks take the answer.
Concentration
70% of answers named at least one tracked firm. The five most-named firms account for 34% of all firm appearances.
Invisibility
11 of the 44 independent firms tracked were never named in any of the 405 answers, and never had their website cited as a source. We report this as a count, never as a list of names.
By engine
ChatGPT and Perplexity, side by side
| Engine | Answers | Independent named | Large / network named | Avg. sources cited |
|---|---|---|---|---|
| ChatGPT (via API, native web search) | 195 | 52% | 42% | 5.3 |
| Perplexity Sonar | 210 | 46% | 45% | 8.2 |
The sources the engines actually read
Which third-party domains the answers cited
Share of the 405 answers that cited each domain as a source. One independent firm’s own website — Fiduciaire LPG — was cited in 15% of all answers, more often than most of the national directories below.
| Source | Share of answers |
|---|---|
| editus.lu | 27% |
| financialservices.lu | 23% |
| guichet.public.lu | 17% |
| fiduciaire-expert.lu | 16% |
| atlasfiduciaire.lu | 14% |
| lu.kompass.com | 12% |
| bookkeeper.lu | 9% |
| tyls.lu | 8% |
| luxtoday.lu | 7% |
Most-named firms
Firms named in five or more of the 405 answers
22 independents and 14 large / network firms cleared the five-answer threshold. Independents are not absent from AI answers as a class — the ones that appear, appear substantially. The gap is between independents, not between independents and networks.
| Fiduciaire LPG | 66 |
| PCG Luxembourg | 30 |
| Fiduciaire des Classes Moyennes SA | 22 |
| Global Compta Services | 22 |
| GT Fiduciaires | 21 |
| B.A. Fiduciaire Sarl | 15 |
| Laffargue Fiduciaire | 15 |
| Fiduciaire Pletschette, Meisch & Associés SA | 14 |
| Fiduo | 13 |
| Groupe Fiduciaire Luxembourgeoise | 13 |
| LPL | 13 |
| Europe Fiduciaire (Luxembourg) S.A. | 12 |
| Fiduciaire de Luxembourg SA | 12 |
| GSL Fiduciaire S.à r.l. | 12 |
| Fiduciaire Omnitrust | 10 |
| Le Cabinet Fiduciaire | 10 |
| Custom SA | 9 |
| Devaux et Associés | 9 |
| W-Conseil | 8 |
| Fiduciaire GL SA | 6 |
| Fideco, Fiduciaire de l'Economie | 5 |
| Fiduciaire des P.M.E. SA | 5 |
| BDO Luxembourg | 58 |
| PwC Luxembourg | 53 |
| Deloitte Luxembourg | 44 |
| KPMG Luxembourg | 44 |
| Grant Thornton Luxembourg | 39 |
| EY Luxembourg | 36 |
| Baker Tilly Luxembourg | 35 |
| ATOZ Services | 16 |
| Forvis Mazars Luxembourg | 12 |
| Alter Domus | 9 |
| RSM Luxembourg | 9 |
| Arendt (Arendt Services) | 8 |
| IQ-EQ | 7 |
| TMF Group Luxembourg | 5 |
Counts are the number of the 405 answers in which the firm was named. A firm named several times within one answer is counted once. Firms named in fewer than five answers are not listed.
Limitations
What this study cannot tell you
A study that hides its weaknesses is marketing. One that states them is research. Read these before you read the figures.
- 1
The sample is not random, and its bias runs in a known direction.
Firms were collected from public directories. Directory presence is also one of the things that drives AI visibility — editus.lu is read in 27% of answers. So firms with weak directory presence are underrepresented in the sample, and the true rate of invisibility across the sector is likely higher than reported here.
- 2
One measurement point.
All figures are from 8 July 2026. Repeated runs on that day were stable, but that says nothing about how these numbers move over weeks. Edition 2 will address this.
- 3
API proxies, not the consumer apps.
Measured through the OpenAI and Perplexity APIs, not the ChatGPT or Perplexity apps, which add their own prompts, memory and ranking. The figures are a close proxy, not the product.
- 4
Commercial interest, stated plainly.
20 More sells AI-visibility audits to firms in this sector. We publish the method and the prompt set so anyone can reproduce or challenge the findings.
No agency can guarantee placement in AI-generated answers. This study measures what the engines answered on a given date; it does not establish what causes those answers.
Method
How the numbers were produced
Identical questions were sent to each engine and the full raw answers were stored. Mentions and citations were then detected deterministically, with no AI judging the result — a firm is counted as named when its name appears in the answer text, and cited when its website appears as a source. The prompt set is versioned so later editions stay comparable.
Mentions and citations are measured separately and never added together. A mention is the firm being named in the answer a client reads; a citation is the firm’s website being used as a source the engine drew on. They answer different questions — “does the engine recommend you?” versus “does the engine read you?” — and a firm can lead on one while absent from the other. Collapsing them into a single score would hide exactly the distinction a firm needs to act on.
The full 80-question prompt set is available on request — email touschlaurent@gmail.com.
A recurring study
This is Edition 1 (July 2026). Later editions will repeat the same versioned prompt set so the figures stay comparable, and a compare-editions view will track how answers move over time — the one thing a single measurement point cannot show.
Want to know how the engines answer about your firm?
Any Luxembourg firm in this sector can request its own free measurement — the same method, run for your name. Email touschlaurent@gmail.com.
Curious how the measurement and the fixes work? See the AI-visibility service.