Make your information reliable
Describe your services, prices or terms, service areas and contact details accurately. Give a date and source for verifiable claims.
Kanexio Observatory
AI search in Belgium
Statbel data and research findings, with context for your online visibility.
AI use in Belgium, in figures
42.01% of people aged 16 to 74.
Of Belgian residents aged 16 to 74 used generative AI in the three months before the 2025 survey.
32.66% of people aged 16 to 74.
in the European Union, for the same indicator.
[11] Statbel, IND_AI1B, cell B5
[1] European comparison, FPS Economy, page 58
AI use differs by age and region
Statbel’s table lets you look beyond the national average. Explore the figures by age or region.
Generative AI use by age
Share of residents in each age group who used generative AI in the three months before the survey, in 2025.
| Group observed | Chart on a scale of 100 | Share of the group |
|---|---|---|
| Aged 16 to 24 | 74.4% of the group concerned | |
| Aged 25 to 34 | 60.0% of the group concerned | |
| Aged 35 to 44 | 47.8% of the group concerned | |
| Aged 45 to 54 | 40.6% of the group concerned | |
| Aged 55 to 64 | 21.7% of the group concerned | |
| Aged 65 to 74 | 11.7% of the group concerned |
Statbel, IND_AI1B, cells B12 to B17. Scale from 0 to 100%, values rounded to one decimal place.
Generative AI use by region
Residents aged 16 to 74, using the same period and indicator. Regions follow the order of the source table.
| Group observed | Chart on a scale of 100 | Share of the group |
|---|---|---|
| Brussels | 47.1% of the group concerned | |
| Flanders | 43.5% of the group concerned | |
| Wallonia | 37.5% of the group concerned |
Statbel, IND_AI1B, cells B7 to B9. Descriptive differences; no significance test has been performed here.
Adapt your content to your audiences
The differences between age groups suggest adapting your content and contact channels to your customers. They cannot predict an individual’s behaviour.
Regional differences are not language differences. The table does not separately measure searches in French, Dutch and English.
Our recommendation: build visibility in AI answers while maintaining clear pages, visibility in traditional search and accessible ways to get in touch.
AI use varies with business size
Statbel’s 2025 ICT survey covers several AI technologies. The indicator is not limited to generative tools or online search.
34.5% of the enterprises concerned.
of enterprises with 10 or more persons employed use at least one AI technology.
This total excludes enterprises with 2 to 9 persons employed. They appear separately in the chart, using Statbel’s published breakdown.
AI use by business size
Share of enterprises in each size group in Belgium, within the scope of the 2025 ICT survey.
| Group observed | Chart on a scale of 100 | Share of the group |
|---|---|---|
| 2 to 9 persons employed | 21.1% of the group concerned | |
| 10 to 49 persons employed | 28.8% of the group concerned | |
| 50 to 249 persons employed | 54.5% of the group concerned | |
| 250 or more persons employed | 76.4% of the group concerned |
Statbel, ClasseEmploi, E_AI_TANY: AC171, X171, S171 and I171. The total for 10 or more persons employed is in Total10Plus, D171.
For a small business, start with a specific need: answering recurring questions, documenting an offer or tracking enquiries. This adoption rate is not evidence of productivity gains or commercial returns.
Personal, professional and educational uses
What is generative AI used for?
Among residents aged 16 to 74 who used generative AI in the three months before the 2025 survey.
| Group observed | Chart on a scale of 100 | Share of the group |
|---|---|---|
| Personal use | 79.8% of the group concerned | |
| Professional use | 49.6% of the group concerned | |
| Formal education | 24.4% of the group concerned |
Statbel, IND_AI2, cells B5 to D5. Purposes can overlap: the percentages do not add up to 100%.
A different denominator
This chart covers only people who use generative AI. It does not describe the entire Belgian population.
Personal use might involve writing, translating or planning an activity. Professional use might concern an internal task. These categories do not isolate searches for a product or service.
Our recommendation: start with your customers’ specific questions, then check whether your content helps answer them.
On Google, AI summaries are associated with fewer clicks
United States, March 2025. These are not Belgian rates and do not describe all AI assistants.
Visits followed by a click on a traditional search result
Google, US panel from March 2025. Share of visits to search results.
| Group observed | Chart on a scale of 100 | Share of the group |
|---|---|---|
| With an AI summary | 8.0% of the group concerned | |
| Without an AI summary | 15.0% of the group concerned |
Common scale from 0 to 100%. Source: Pew Research Center, published on 22 July 2025.
[3] Findings published on 22 July 2025
The panel comprises 900 US adults. The analysis covers 68,879 distinct Google searches; search results were collected from 7 to 17 April 2025, after the browsing activity observed in March. This delay limits how closely they match what each person actually saw.
The rates concern visits to search results, not a proportion of people or a measured loss of traffic to your site. The comparison is observational: on its own, it does not isolate the causal effect of the AI summary.
[4] Panel recruitment, weighting and limitationsAn August 2026 preprint extends the analysis with a statistical model accounting for differences between queries and participants. It uses the same panel, rather than new Belgian data. We distinguish it from a peer-reviewed journal article.
[5] Additional analysis, August 2026Practical priorities for your visibility
Kanexio recommendations, to adapt to your customers and assess through measurement.
Maintain SEO fundamentals
Check indexability, internal links, mobile readability and whether structured data matches the page. Google does not require an AI file or special schema for its AI features.
Track genuine enquiries
Compare identifiable visits and qualified enquiries by page and language. A citation, a visit and a customer are three different outcomes.
In Search Console, links from Google’s AI features are included in the Web search type. That total is not a separate measure of AI search. Identifiable referral visits in Analytics remain incomplete; a mention without a click does not become a visit.
[6] Google’s documentation on AI featuresA protocol for collecting Belgian observations
This protocol is proposed for future data collection. This edition publishes no Kanexio citation rate.
Define the questions before testing
Set the services, intentions, Belgian locations and languages to be covered. Publish the exact questions and how they were selected. Separate branded from unbranded questions; a selected set of questions does not represent Belgian internet users.
Fix the observation conditions
Record the platform, the displayed model or its absence, date and time, language, location, sign-in status and conversation state. Plan repetitions and their dates before collecting data.
Keep every attempt
Archive the response and its references, including missing answers, refusals and errors. Define a usable response in advance. Report raw counts, denominators and exclusions for each platform and language.
Code the findings so others can check them
Distinguish a brand mention, a cited URL, an accessible page and an accurate claim. Have a second person check ambiguous cases and record disagreements. Repeated versions of the same question are not independent observations.
Compare comparable rounds
Keep a stable core of questions and disclose any change to the model or protocol. Keep incompatible series separate. A before-and-after measurement alone does not establish an intervention’s causal effect.
Publish useful evidence
Include the protocol, negative findings, calculation rules and limitations. Remove personal information, check the lawful basis and follow platform terms. No survey margin of error will be assigned to a non-probability sample without a justified model.
Clear methods and explicit responsibilities
Publisher and commercial interest
Kanexio publishes this review and sells digital visibility services. Recommendations are identified as such. The organisations cited neither commissioned nor approved this page; their publications do not endorse Kanexio.
Review methods and limitations
The search covered official Belgian and European sources, primary research on search behaviour and methodological references available at the edition date. Figures without an identifiable population, period or method were excluded. No composite score or average across dissimilar studies was calculated.
Ethical references
The principles of transparency and separating findings from recommendations draw on the 2025 ICC/ESOMAR Code and AAPOR guidance. ISO 20252:2026 is the relevant service standard for market and opinion research. This page claims no ISO certification, membership or approval from these organisations.
Data and writing assistance
AI assisted the desk research, structuring and writing. The figures come from the cited sources; no respondents or results were simulated. This working version has not undergone independent scientific review. No personal client data or private GSC or GA4 data is published.
For future data collection involving people, the purpose, lawful basis, participant information, data minimisation and retention period must be documented.
[10] GDPR, Articles 5 and 6Correction history
- 30 July 2026: internal percentages were withdrawn because the dataset was not published and totals were inconsistent. Those figures remain withdrawn.
- 24 September 2026: the evidence review was expanded with Statbel’s 2025 tables, charts by audience and business size, and a real photograph under a CC0 licence. The original URL is retained.
Cite this edition or suggest a correction
Kanexio, AI search in Belgium: evidence review, edition of 24 September 2026. Belgian generative AI usage data: 2025 ICT survey. Click behaviour: Pew’s US panel from March 2025. These scopes are not interchangeable. This edition contains no original measurement of the Belgian AI search market. Source: https://kanexio.com/en/resources/ai-search-belgium-barometer/
A newer source or a point to correct? Write to ask@kanexio.com.
Measure what matters to your business
Let’s define your questions, audiences and enquiries to track, with goals suited to your business.
