By Timothy Jacqmin — Co-founder, Nexuro Digital · July 2026
TL;DR
- Across a panel of 44 Belgian queries (14 sectors, French + Dutch, 440 sources analysed in July 2026), 70% of the sources that engines retrieve to answer are Belgian (.be).
- But that average hides a huge sector gap: from 85% Belgian in finance and legal to just 20% in restaurants/tourism.
- Reddit is the single most-cited domain across all sectors, appearing in 20% of queries.
- Institutional sources (belgium.be, wallonie.be) and comparison/aggregator platforms (Test-Achats, Indeed, comparators) carry heavy weight: a comparator appears in 48% of queries.
- Wikipedia appears in none of these transactional/local queries — contradicting its reputation as a top AI source.
Why this study
When a Belgian asks ChatGPT, Claude, Perplexity, Gemini or Google's AI Overviews a question — "which health insurer should I pick?", "best restaurant in Brussels", "independent accountant" — which sources will the machine read before answering, and which will it cite?
That is the core question of GEO (Generative Engine Optimization): the art of being the source an AI reuses. Unlike classic SEO (ranking in a list of links), GEO is decided by what others say about you and by the sources the engine deems worthy of synthesis.
To bring hard data to the Belgian side of the question, we measured. Here are the results, the method, and what every player — whatever the sector — can take from it.
Methodology (in full transparency)
- Panel: 44 queries representative of Belgian demand, spread across 14 sectors (health, legal, real estate, construction, finance/insurance, restaurants/tourism, automotive, employment/HR, energy, retail, education, public administration, local services, business/IT).
- Languages: 39 queries in French, 5 in Dutch (to capture the country's bilingual dimension).
- Measurement: for each query, retrieval of the top 10 sources that engines surface, via the Firecrawl search API — a total of 440 sources classified by type and by geographic origin (.be domain vs foreign).
- Date: July 2026 (a point-in-time snapshot).
Limitations, acknowledged. This method measures the sources retrieved by engines (a reliable proxy for what LLMs read during retrieval), not the exact wording of a given ChatGPT's final answer. The Dutch-language sub-panel is small (50 sources). The "Media / Professional site" category is deliberately broad. The figures describe a trend, not carved-in-stone truth.
Finding 1 — 70% of sources are Belgian… on average
Across the whole panel, the origin of the 440 sources breaks down as follows:
| Origin | Share |
|---|---|
| Belgium (.be) | 70% |
| International (.com, etc.) | 27% |
| France (.fr) | 2% |
| Other EU | 1% |
Good news for the Belgian ecosystem: generative engines strongly favour local sources when the query is Belgian. But the average is a trap — the sector detail tells a very different story.
Finding 2 — The sector gap (the real takeaway)
The share of Belgian sources varies fourfold depending on the sector:
| Sector | Belgian sources | Reading |
|---|---|---|
| Finance / Insurance | 85% | Regulated trust, strong .be presence |
| Legal | 85% | Notaries, bar associations, national legal framework |
| Automotive | 83% | Local dealers and services |
| Energy | 82% | Suppliers and regional grants |
| Health | 80% | Local practitioners and insurers |
| Construction / Renovation | 80% | Regional grants, local contractors |
| Real estate | 72% | Strong Belgian platforms |
| Public administration | 70% | Public .be sources |
| Retail / E-commerce | 70% | Local / international mix |
| Local services | 67% | Directories + reviews |
| Business / IT / Marketing | 65% | International content competition |
| Employment / HR | 60% | Global job platforms |
| Education | 45% | International rankings and content |
| Restaurants / Tourism | 20% | Booking, TripAdvisor & co. dominate |
The pattern is clear: the more a sector rests on regulated trust or a local relationship (finance, law, health, energy, construction), the more AIs cite Belgian sources. The more a sector is commoditised or structured by international platforms (tourism, education, employment, tech/marketing), the more the citation leaks abroad.
In other words: in tourism, a Brussels restaurant relying on ChatGPT is, 80% of the time, at the mercy of foreign platforms. In finance, a Belgian institution plays at home.
Finding 3 — Which source types AIs retrieve
| Source type | Share |
|---|---|
| Media / Blog / Professional site | 77% |
| Institutional / Official / Public | 9% |
| Directory / Comparator / Platform | 8% |
| Forum / Reddit / Q&A | 3% |
| Social networks | 2% |
| YouTube / Video | 1% |
Editorial content (media, blogs, company sites) remains the bedrock: three quarters of sources. But two "short" categories carry disproportionate strategic weight.
Reddit accounts for only 3% of total volume, yet it is the single most-cited domain in the entire study (12 appearances), present in one query out of five. AIs go there for genuine reviews, real-world feedback and marketing-free comparisons.
Official sources and comparators act as gatekeepers. belgium.be is the 2nd most-cited domain (10 appearances); comparators and platforms (Test-Achats, Indeed, sector-specific comparators, Sortlist…) appear in 48% of queries, and in the top 3 for nearly one query out of four.
Finding 4 — The most-cited domains
The domains that recur most, across all sectors:
| Domain | Appearances | Type |
|---|---|---|
| reddit.com | 12 | Forum |
| belgium.be | 10 | Official |
| notaire.be | 8 | Professional site / federation |
| facebook.com | 7 | Social network |
| youtube.com | 6 | Video |
| indeed.com | 6 | Job platform |
| test-achats.be | 5 | Comparator |
| wallonie.be, doctoranytime.be, xerius.be, selectra.be | 4 | Official / platforms |
Finding 5 — Wikipedia, the great absentee
Globally, Wikipedia is often cited as one of the two most-reused sources by AIs. In our Belgian panel, it does not appear a single time. The reason lies in the nature of the queries: transactional and local ("a plumber", "an insurer", "a restaurant"), not encyclopedic. The lesson: the source to target depends on intent, not on a universal hierarchy.
Finding 6 — The FR / NL language divide
Dutch queries surface slightly more Belgian sources (74%) than French queries (69%). The most likely hypothesis: Belgian French-language content is competing with France's vast pool of sources (.fr), whereas Belgian Dutch-language content is more "insulated". The sample is small, so treat it with caution — but the implication is real: a French-speaking Belgian company has to work harder on its local anchoring to avoid being overtaken by French content.
What this means for you (whatever your sector)
If you are in a "home advantage" sector (finance, legal, health, energy, construction, automotive): AIs already cite mostly Belgian sources — the challenge is to be one of them. Build authority on a .be domain, publish dated and sourced expert content, and secure a presence in your profession's federations, orders and official references.
If you are in a "leaking" sector (tourism, education, employment, tech/marketing): you are competing against international platforms. The GEO counter-move: produce local reference content (data, comparisons, Belgium-anchored guides) that the engine will prefer over a generic foreign page — and polish your presence on the platforms that are being cited.
For everyone, four universal levers:
- Comparators & reviews — be present (and well rated) on the platforms cited in your sector (Test-Achats, Google, trade directories): one listing often equals one citation.
- Reddit, authentically — find the Belgian discussions in your sector and be genuinely useful there (never disguised promotion: it backfires and is easily detected).
- GEO-structured content — direct answers up top, hard figures, tables, FAQs, original data: what an AI can copy without rewriting.
- Entity consistency — the same description, name and contact details everywhere (website, Google Business, platforms, social) so AIs "understand" who you are.
FAQ
What is GEO (Generative Engine Optimization)? It is optimising your presence to be cited and reused by generative engines (ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews), where SEO aims for a rank in a list of links. GEO is decided as much on your own site as by what other sources say about you.
Do AIs mostly cite Belgian sources for Belgian queries? On average yes — 70% of retrieved sources are Belgian in our study. But it varies widely: up to 85% in finance and legal, but only 20% in restaurants/tourism, where international platforms dominate.
Is Reddit really important to be cited by AIs? Yes. In our Belgian panel, Reddit is the single most-cited domain, present in one query out of five. AIs find authentic reviews and real-world feedback there. Being useful (without aggressive marketing) is a powerful GEO lever.
Do you need a Wikipedia page to be cited? Not for transactional or local queries: Wikipedia appeared on none of the queries in this study. It remains useful for brand authority and informational queries, but it is not the priority lever for a local service provider.
How can an SME improve its visibility in AI answers? By combining: presence on the comparators and platforms cited in its sector, structured and sourced content on its own site, authentic presence on Reddit and forums, and consistency of its brand identity everywhere it appears.
Is French-language content at a disadvantage in Belgium? Slightly: Dutch-language queries surface a few more Belgian sources (74%) than French ones (69%), probably because Belgian French-language content competes with sources from France. Strong local anchoring (Belgium, city, region) helps you stand out.
Study conducted by Nexuro Digital (digital marketing agency & Official Odoo Partner, Belgium) — 44 queries, 440 sources, July 2026. Reproducible method: measurement of sources retrieved by engines via the Firecrawl API.
