The Dutch data protection watchdog has warned voters not to ask artificial intelligence chatbots for voting advice ahead of the country’s general election next week.

“AI chatbots give a highly distorted and polarized image of the Dutch political landscape in a test,” the data protection watchdog warned in a study published on Tuesday.

“We warn not to use AI chatbots for voting advice, because their operations are not transparent and verifiable,” Monique Verdier, vice-chair of the authority, said in a statement. She called upon the chatbot developers to “prevent that their systems are being used for voting advice.”

Dutch voters elect a new parliament next Wednesday.

The Dutch data protection authority ran an experiment on how parties were portrayed in voting advice across four different chatbots, including OpenAI’s ChatGPT, Google’s Gemini, Elon Musk’s Grok and French Mistral AI’s Le Chat.

The authority set up profiles that matched different political parties (based on vetted Dutch voting-aid tools), after which it asked the chatbots to give voting advice for these profiles.

Voter profiles on the left and progressive side of the spectrum “were mostly directed to the GreenLeft-Labor” party led by former European Commission Executive Vice President Frans Timmermans, while voters on the right and conservative side “were mostly directed to the PVV,” the far-right party led by Geert Wilders that is currently leading in the polls.

Centrist parties were hardly represented in the voting advice, even though these parties were represented equally in the voter profiles fed to the chatbots.

OpenAI, Google and Mistral have all signed up to the EU’s code of practice for the most complex and advanced AI models, while Grok’s parent company xAI has signed up to parts of it. Under the code, these companies commit to address risks stemming from their models, including risks to fundamental rights and society.

The Dutch authority argued that chatbots giving voting advice could be classified as a high-risk system under the EU’s AI Act, for which a separate set of rules will start to apply from mid-next year.