AI research reports.
How AI assistants answer Dutch questions.
New insights every quarter for business owners, marketers and researchers.
Receive every new report in your inboxWhen does AI mention a business?
What 18,206 Dutch AI answers reveal about search, business mentions and sources.
- ChatGPT adds a year on its own in 82% of measured answers to cost questions
- With a place name: 12.3% mentioned versus 3.7% without, weighting each question equally
- Mentioned again: 55.8% with equal weighting per question, 67.6% per transition
How we measure
Klusio studies how AI answers Dutch customer questions. We retain the measured answers, cited sources and available search queries. Each quarter, we share the patterns that emerge, with practical insights for business owners and supporting evidence that can be checked.
Repeated measurements show how answers differ by system, day and question. Each report states which data was available, which analyses were performed and what was added later.
Answers as the research basis
We archive questions, answers, and available sources and search queries. A search query that was not provided remains unknown. Derived assessments may be expanded later; the report version makes this visible.
Customer questions and a fixed panel
We combine aggregated customer measurements with our own test questions. Q3 includes 32 questions from a fixed set started in late September. For quarterly comparisons, that series must be tracked separately, with a fixed measurement schedule and recorded changes.
Dutch questions, different setups
We use web-interface capture, API responses and Google search results. For each system, the report states the collection method, recorded models, period and failed measurements. Unknown historical settings remain explicitly unknown.
What "mentioned" means
The tracked business's name appears in the answer text in response to a question that does not itself contain that name. A mention only in the source list is counted separately. A mention is not a recommendation or a ranking; we measure those separately. We check recognition on a sample each quarter.
Fair comparisons
We compare within the same business and AI system using Mantel–Haenszel, alongside the number of businesses for which the association points in the same direction. The 95% interval accounts for repeated measurements of the same question. We also show raw figures and check whether a finding holds when Klusio's own questions are excluded. These remain associations, not causes.
Equal weight for every question
The main findings show equally weighted results per question alongside results per answer. Q3 also includes a check using at most one measurement per question, system and week. That retrospective selection is not a measurement schedule defined in advance.
Providers and positions
Since Q3 version 1.2, an additional analysis has been included: a language model classifies mentioned names and their order. This analysis was performed retrospectively on the quarterly archive. The method and limitations accompany those results.
A clear minimum
Substantive group findings require at least 5 businesses and 100 answers; for the fixed question set, the minimum is 50 answers. Smaller counts may still be shown to check recognition or failed measurements. They are not presented as market findings.
Versions and downloadable figures
Every publication has a version and change log. The CSV download contains aggregated supporting evidence, definitions and, where available, numerators, denominators and group sizes. Confidential answers remain protected, so not all analyses can be fully recalculated externally.
Privacy
We never publish customers' names, websites or questions. Examples come from our own fixed question set. We do name platforms and public sources cited by AI itself.
Which systems we measure
Citation
You may reuse figures from the reports with attribution and a link to the report.
Questions about the data
Are you a journalist, researcher or editor who wants to know more about a figure or the methodology? Get in touch and we will explain.