How many Dutch voters might seek AI voting advice?

Ahead of the Netherlands' 2025 national election, the AlgoSoc research consortium — a collaboration between the University of Amsterdam, Utrecht University, TU Delft, Tilburg University, and Erasmus University Rotterdam — surveyed 1,220 Dutch adults drawn from Centerdata's LISS panel on how likely they would be to consult ChatGPT or another AI chatbot for advice on political parties, candidates, or election issues. Scaled to the national population, the authors report this implies "naar schatting 1,8 miljoen Nederlanders" — an estimated 1.8 million Dutch people — could consult AI when deciding how to vote. The willingness to do so was strongly age-graded: it fell sharply as respondents got older.

The authors add a specific measurement caveat rather than treating the figure as a ceiling: because survey respondents tend to under-report behavior seen as "sociaal onwenselijk" (socially undesirable), "kan het werkelijke aantal nog hoger liggen" — the real number could be higher still. The same survey found that 69.2% of respondents — almost seven in ten — said they worried about other Dutch citizens receiving incorrect political information through AI tools, a concern the authors read as evidence that public awareness of the risk is already high even where personal admitted use is likely undercounted.

This is a self-report of stated intent, not of what chatbots actually tell the people who consult them — that is supplied by the companion finding in the same AlgoSoc/AP report, Do chatbots steer Dutch voters toward the same parties?. Read together, the two findings describe a sizeable and disproportionately young population opting into a channel whose political output, per that companion test, skews toward the same two parties regardless of what the user actually believes. The shape — AI uptake paired with a documented quality or accuracy cost in a public-facing information context — also echoes Do AI writing tools improve online discussion or degrade it?, though that study measured discussion quality rather than voting advice specifically.

The survey does not establish how many of the 1.8 million estimated would-be users actually consult AI before voting, whether consulting it changes anyone's vote, or how large the under-reporting gap really is — only that stated willingness is high and age-correlated, and that the authors believe it understates true use. Given the size of the population at stake and the fact that younger voters, who are forming voting habits for the first time, report the highest willingness, the finding points toward a cohort encountering electoral information through a channel that was not built, licensed, or audited to provide it.