What AI assistants answer when people ask in Dutch.

The latest report examines 18,206 answers to 1,538 Dutch questions, covering 35 businesses and 5 main AI systems from 1 July to 30 September 2026. It examines search queries, business mentions and cited sources. The observations come from Klusio’s own measurement archive, not a representative sample of every Dutch business or AI user.

ChatGPT search queries

For cost questions with visible search queries, ChatGPT added a year in 82% of measured answers. This finding concerns that subset, not every AI answer.

Questions containing a place name

The tracked business appeared in 12.3% of answers to questions containing a place name versus 3.7% without one, giving each question equal weight. This is an association, not proof that adding a place causes a mention.

Being mentioned again

Following an earlier mention, the repeat-mention rate was 55.8% with equal weight per question, compared with 67.6% per measured transition. More frequently measured questions affect the second figure more.

• Read the complete Q3 2026 report →
• Download the shareable research charts →

We preserve the measured answers, cited sources and any search queries supplied by the system. Repeated measurements show how answers differ by system, date and question. Each quarterly report states which data was available, which analyses were performed and what was added later.

1. Answers as the research record

Questions, answers, available sources and search queries are archived. A query that was not supplied remains unknown. Derived assessments can be extended later; the report version records those changes.

2. Customer questions and a fixed question set

Aggregated customer measurements are combined with Klusio’s own questions. Q3 includes thirty-two questions from a fixed set started in late September. Quarter-to-quarter comparisons need to follow that series separately, using a fixed schedule and recording changes.

3. Dutch questions, different measurement routes

The archive combines web-interface observations, API responses and Google search results. Each report records the route, stored model labels, period and failed or empty responses for each system. Unknown historical settings remain explicitly unknown.

4. What “mentioned” means

The tracked business’s name appears in the answer text to a question that does not itself name that business. A name appearing only in a source list is counted separately. A mention is neither a recommendation nor a position; those require separate analysis. Recognition is checked on a sample each quarter.

5. Like-for-like comparisons

We compare within the same business and AI system using Mantel–Haenszel estimates, and show how many businesses have an association in the same direction. The 95% uncertainty interval accounts for repeated observations of a question. Raw figures and a sensitivity check excluding Klusio’s own questions are also reported. Associations do not establish causes.

6. Giving questions equal weight

The main findings show question-weighted results alongside answer-weighted results. Q3 also includes a check using at most one measurement per question, system and week. This is a retrospective selection, not a pre-planned measurement schedule.

7. Providers and positions

From Q3 version 1.2, a supplementary language-model analysis classifies named organisations and their order. It was applied retrospectively to the quarterly archive. Its method and limitations accompany those results.

8. Minimum group sizes

Substantive group findings require at least five businesses and one hundred answers; the fixed question set requires fifty answers. Smaller counts may be shown to diagnose recognition or missing data, but are not presented as market findings.

9. Versions and downloadable figures

Each publication has a version and change log. The CSV contains aggregated evidence, definitions and, where available, numerators, denominators and group sizes. Confidential answers remain protected, which limits independent reconstruction of all analyses.

10. Privacy

Customer names, websites and questions are not published in the research. Public examples come from Klusio’s own fixed question set. Platforms and public sources cited by the systems may be named.

The five main systems use different access routes. These are the routes recorded for the research, not a claim that every route reproduces every user’s personal answer.

Klusio develops software to help businesses improve their visibility in Google and AI. These reports are our own measurements. That commercial interest is why the methods and limitations accompany each report. The research does not measure the effect of using Klusio on customer growth.

• Read separately measured customer stories →
• Explore the AMI case evidence →

You may reproduce figures with attribution and a link to the report. Include its version, the measurement period and the relevant denominator or weighting. Do not turn a measured association into a guaranteed outcome.

Klusio (2026). When does AI mention a business? AI research report Q3 2026, version 1.3, source updated 4 October 2026. English edition prepared 21 October 2026. https://klusio.nl/en/ai-research/q3-2026/

If you are a journalist, researcher or editor and would like to discuss a figure or method, contact us. Sicco Dijkman is the contact for questions about the research.

• Ask about the research →