Dutch regulator warns AI chatbots ignore local parties ahead of municipal elections
A new investigation by the Dutch Data Protection Authority (AP) has exposed a significant flaw in how AI chatbots handle municipal elections. The research found that when users ask for voting guidance, AI tools systematically overlook local parties, raising concerns about the impartiality and accuracy of AI-driven political advice.
The Dutch Data Protection Authority studied four widely used chatbots, ChatGPT (OpenAI), Gemini (Google), Mistral, and Grok (xAI), and compared their results with traditional voting aids like Kieskompas and StemWijzer. The report comes in preparation for the municipal elections on March 18.
The AP research found that fewer than 1 percent of voting recommendations mentioned a specific local party, despite the fact that such parties typically win about 30 percent of the vote in municipal elections.
Some reasons mentioned for this by the AP are that AI chatbots are trained on vast amounts of online data, where national parties are far more visible than local ones. This means local parties are usually omitted from answers. Additionally, local parties often provide less detailed and less current information online compared with major national parties.
The Dutch Data Protection Authority cautions that using these systems as voting tools can give a skewed picture of the political landscape, as voters may not be properly informed about key choices, including local parties.
The authority also stressed that AI chatbots are not appropriate for providing voting advice, both due to this bias and the lack of transparency in how their responses are produced.
Previous research from October 2025 found that chatbots frequently recommended the same two parties in national elections, no matter the question. GroenLinks-PvdA or the PVV appeared in over 56 percent of recommendations, and for one chatbot, this figure rose to 80 percent. In the tests, parties like SGP, DENK, BBB, and CDA were rarely included, even when users provided their precise policy positions.
Pieter Heerma, Minister of the Interior, has voiced concern over voters relying on AI chatbots for election guidance. He reminded citizens that advice from artificial intelligence cannot be considered trustworthy. “It vividly demonstrates that the advice provided is partly arbitrary,” the minister said in response to the reports. He concluded that the results are “unreliable, opaque, and not easily understood.”
A spokesperson for the AP told NL Times that the report is meant to clearly explain their stance regarding AI, large language models, and chatbots, especially when providing advice on important issues, like elections. While the office does not yet have regulatory authority on the matter, the AP believes it could become the watchdog over the issue as early as this summer if the Cabinet quickly implements policies supported by their predecessors.
