No need, or no category? Recording refusal of generative AI in European official statistics
This paper argues that European official statistics erroneously attribute the non-use of generative AI to a lack of need because survey design limitations fail to capture critical barriers like disability and cost, thereby mistaking measurement artifacts for genuine public indifference.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
When governments want to understand how people live, they often turn to official surveys. These are not casual conversations; they are carefully designed tools that turn complex human lives into numbers the state can count and manage. To make this possible, survey designers must create categories—boxes into which people can fit their answers. If a person's experience does not fit neatly into a box, the system often forces them to choose the closest option or place them in a leftover pile labeled "other." This process is not just about organization; it shapes what the government can see. If a survey does not offer a box for a specific reason, that reason effectively disappears from the official record, making the people who hold that view invisible to policymakers. This is the core problem explored in a recent study of how Europe measures its relationship with artificial intelligence.
The study focuses on a specific moment in 2025, when the European Union released its first official statistics on why people do not use generative AI tools. The prevailing story from these numbers was one of indifference. The data suggested that the vast majority of people who do not use these tools simply do not feel the need to. This conclusion implied that the divide between users and non-users was not a deep social fracture, but rather a collection of individual choices that required little government intervention. However, a researcher named Jonas A. Mandalunes, working with data from the European Union's official survey on technology use, found that this reassuring story was an illusion created by the survey itself. The illusion was not a mistake in calculation, but a flaw in the design of the questions asked.
To understand the flaw, one must look at how the survey was built. The same questionnaire asked two very similar questions to two different groups of people. One group, those who did not use the internet, was asked why they stayed offline. They were given a long list of reasons and told they could tick as many as applied. This list included options for physical or mental disabilities, the high cost of equipment, and a general opposition to the internet. The second group, those who used the internet but had not tried generative AI, was asked why they avoided it. They were given a much shorter list and told to pick only one main reason. Crucially, this short list did not include options for disability, cost, or principled opposition. A person who could not use AI because of a visual impairment, or because they could not afford a subscription, or because they morally objected to the technology, had no box to check. They were forced to either pick a reason that did not describe them or fall into the "other" category.
The researcher compared the answers from nineteen European countries to see what this design choice cost the data. The findings were stark. In the survey about the internet, the three missing categories—disability, cost, and principled opposition—accounted for a median of 26 percent of the reasons given by non-users. In contrast, the "other" category for the AI survey, which was supposed to catch all the reasons that did not fit the main list, captured only a median of 4.5 percent. This gap existed in every single country studied. It was not a fluke of the data; it persisted even when the researcher adjusted the numbers to account for the fact that the internet survey allowed multiple answers while the AI survey did not. Even after this conservative adjustment, the gap remained large, suggesting that a significant portion of the reasons for avoiding AI were simply not being recorded.
The study also looked at who was being missed. One might assume that the people forced into the "other" category were the same elderly or less-educated groups who often struggle with technology. The data showed the opposite. The gap was actually larger among highly educated people. This makes sense when one considers that the missing category was "principled opposition." It takes resources and education to form a considered, ethical objection to a new technology. By failing to offer a box for this view, the survey system was effectively silencing the very people most likely to have a thoughtful reason for saying no. The system was not just missing data; it was misrepresenting the nature of the refusal, turning a thoughtful stance into a simple lack of interest.
The researcher tested whether these missing reasons might have been hiding inside the "no need" answer. Perhaps people who objected on principle were just ticking "no need" because they had no other choice. The data did not support this. The number of people saying "no need" did not rise in countries where the missing reasons were most common. Nor did the "other" category swell in those countries. The missing reasons did not seem to go anywhere; they simply vanished from the record. This led to a troubling conclusion: the statistical apparatus was unable to see disability, cost, or ethical objection as valid grounds for refusing AI. When a government cannot see a problem, it cannot fix it. If the data says people just don't need the technology, the policy response is to do nothing or to run awareness campaigns. If the data showed that people were excluded by cost or disability, the response would be to provide access or support.
The story does not end with a static critique, however. The researcher noted that the European statistical system has already begun to fix the most obvious part of this error. In the 2026 version of the survey, the rule requiring people to pick only one reason has been changed to allow multiple answers, and a new category for "ethical concerns" has been added. This addition acknowledges that people might refuse AI because of issues like bias or lack of accountability. However, the system has not yet added boxes for disability or cost in the AI section, even though it keeps those boxes for the internet section. This means that while the survey is learning to hear about ethical objections, it still cannot hear about the material barriers that prevent people from using the technology.
The final lesson of this study is that the way we ask questions determines what we can know about the world. The 2025 survey produced a picture of Europe where most people simply did not want AI. But that picture was a reflection of the survey's limited vocabulary, not necessarily the reality of European life. The data showed that the system was built to hear only a narrow range of reasons for saying no. Until the survey offers boxes for disability, cost, and principled objection, the official statistics will continue to mistake a measurement error for public indifference, leaving the real barriers to technology use invisible to the policymakers who need to see them.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.