The Netherlands fails the AI language test. Conclusion assessed the AI knowledge of 1,000 Dutch citizens using twenty statements on AI terminology and literacy. The average score was 5.1 out of 10. On AI terminology specifically, the country scored even lower: 4.9 out of 10. This is striking, as most Dutch people rate their own knowledge far more highly. More than two-thirds describe themselves as having at least a basic level of AI proficiency. However, the test shows that this self-assessment does not match reality. Nearly half believe generative AI searches for and summarises existing information, while only 20 per cent know what an LLM is.
More than two-thirds of Dutch people describe themselves as having at least a basic level of AI proficiency. Men tend to rate themselves more highly than women (76% versus 66%). However, only 17 per cent say they use AI regularly and know a great deal about it. That figure also declines sharply with age: from 34 per cent among those aged 16 to 29, to 14 per cent among 50 to 59-year-olds, and just 3 per cent among people aged 70 and over.
Knowledge of AI terminology varies considerably. Deepfake is by far the most widely recognised AI-related term, with 81 per cent of Dutch people knowing what it means. However, familiarity drops sharply when it comes to concepts related to how AI tools actually work. Only 20 per cent know what a large language model (LLM) is, despite it being the technology behind tools such as ChatGPT and Claude. Just 14 per cent understand that generative AI creates new content rather than searching for and summarising existing information. This misconception is held by almost half of all respondents (47%).
People who are unaware that AI can hallucinate are less likely to verify its responses.
What is an AI hallucination?
The largest gap between men and women appears in understanding the concept of an AI hallucination: when AI generates information that sounds convincing but is factually incorrect. Fifty-one per cent of men recognise the definition, compared with 32 per cent of women. Overall, 42 per cent of Dutch people know what the term means. This is concerning, as people who are unaware that AI can hallucinate are less likely to verify its responses.
Daniëlle Graat, director of Data & AI at Conclusion: "People think they understand AI because they use it. But using AI and understanding it are two very different things. If nearly half of the Netherlands believes that ChatGPT retrieves information in the same way Google does, then we are missing a crucial step in understanding what AI does and, just as importantly, what it does not do. That distinction is essential for using AI responsibly."
Search assistant and writing aid most popular use cases
Dutch people primarily use AI as a smart search assistant (69%) and as a writing aid (42%). Usage patterns, however, vary significantly by age group. Among people aged 16 to 29, only 4 per cent have never used an AI tool. Among 40 to 49-year-olds this rises to 13 per cent, among 60 to 69-year-olds to 31 per cent, and among those aged 70 and over to half of respondents (50%). The way AI is used also differs between generations. Forty-two per cent of younger respondents use AI to help inform decisions about purchases, travel and health. Among people aged 40 to 49, this falls to 30 per cent, and among those aged 70 and over to 10 per cent.
The research shows that the Netherlands is embracing AI, but has yet to master its language. Confidence is high, yet understanding of how AI works is lagging behind. For a country increasingly using AI in work, education and daily life, that represents a vulnerable position.
Graat added: "The AI language test shows that there is still work to be done. The Netherlands does not need to become a nation of AI experts. But people who use AI without understanding what it can and cannot do face risks they may not see coming. That understanding starts with the basics: knowing that AI does not look up answers, but generates them. It may sound like a small distinction, but it fundamentally changes how you evaluate the responses it produces."
Always up-to-date
Newsletter