AI Chatbots Unreliable for Voting Advice, Dutch Watchdog Warns

As voters increasingly turn to artificial intelligence for guidance in various aspects of life, Dutch authorities are raising alarms about using AI chatbots for electoral advice. The Dutch Data Protection Authority (AP) has cautioned against relying on popular AI tools such as ChatGPT and Gemini for voting decisions ahead of the country’s October 29 snap elections, citing reliability concerns and systematic bias.

In a comprehensive study, the AP tested four leading chatbots—ChatGPT, Gemini, Grok, and Le Chat—by inputting 200 typical voting profiles per political party currently represented in the Dutch parliament. The results revealed troubling patterns in how these AI systems approach political recommendations.

“What we saw was a kind of oversimplification of the Dutch political landscape, where one party on the left kind of sucked up all the votes in that corner of the political spectrum and the same with the one party on the right,” explained Joost van der Burgt, project manager at the AP.

The investigation found that regardless of user inputs or specific questions, the chatbots consistently recommended just two parties in 56% of cases: Geert Wilders’ right-wing Party for Freedom (PVV) or Frans Timmermans’ left-leaning Green Left-Labour Party (GL/PvdA)—coincidentally, the two parties forecast to win the most parliamentary seats.

Meanwhile, centrist viewpoints were notably underrepresented, and smaller parties like The Farmer-Citizen Movement (BBB) and Christian Democratic Appeal (CDA) were almost never suggested, even when user profiles directly aligned with these parties’ platforms. Oddly, the relatively new right-wing party JA21 received a disproportionately high number of recommendations despite having less extensive media coverage than established parties.

Van der Burgt attributes these biases to the fundamental way large language models operate. “They are basically statistical machines that predict missing words in a phrase or a certain output,” he told Euronews’ verification team. “If your political views position you towards one end of the political spectrum, it’s maybe not that surprising that generative AI picks a political party that fits that side and seems like a safe choice.”

The findings raise concerns about AI’s potential classification under the European Artificial Intelligence Act (AI Act), which came into force in August 2024. Researchers suggest chatbots providing voting advice could qualify as “high-risk systems” under rules scheduled for implementation in August 2026.

Van der Burgt believes safeguards should be implemented similar to those already in place for other sensitive topics. “It is already the case when it comes to questions about mental health or aids to create improvised weapons,” he noted. “In all these scenarios, a chatbot clearly states, ‘I’m sorry, we’re not allowed to help you with it.’ And we think that the same sort of mechanism should be in place when it comes to voting advice.”

The AP compared chatbot results to established Dutch voting advice tools like StemWijzer and Kieskompas, which rely on structured data rather than generative AI. These traditional tools ask voters to answer a series of 30 questions to determine their political alignment. Similar systems exist in other European countries, including Germany’s government-approved Wahl-O-Mat.

Experts highlight a critical advantage of traditional voting tools: transparency. “One fundamental problem with these chatbots is that the way they work is untransparent,” said van der Burgt. “We, nor the public, nor journalists can figure out why exactly they will produce a certain answer.”

Despite concerns, some researchers see potential in controlled AI systems designed specifically for electoral guidance. In Germany, researcher Michel Schimpf launched Wahl.Chat ahead of the country’s 2025 federal elections as an alternative to traditional tools. This specialized bot incorporates party manifestos and allows voters to ask specific policy questions.

“If you ask ChatGPT a question, its source could be a biased website, but we made sure that our bot relies on party manifestos,” Schimpf explained. His team incorporated fact-checking functionality and designed the system to reduce bias, though he acknowledges limitations: “It’s still AI, which is ultimately probabilistic.”

Other researchers are taking even more measured approaches. Naomi Kamoen, assistant professor at Tilburg University, is helping develop a limited chatbot with answers written by qualified researchers rather than generated by AI. “The goal is to inform, not necessarily to tell them what to vote for,” she said.

Experts generally agree that AI can play a constructive role if used properly. Jianlong Zhu, doctoral researcher at Saarland University, suggests AI can help educate voters by breaking down complex political concepts in real-time—something traditional voting tools cannot do.

“If you are using Wahl-O-Mat and there are concepts that you don’t understand, what you end up doing is randomly clicking stances,” Zhu explained. “But with a chatbot, you can ask what those terms mean, and the chatbot is effective in breaking down those terms to help you engage with the topic better.”

As AI continues to evolve, experts recommend users focus on asking chatbots about parties’ positions rather than requesting direct voting recommendations. Many also advocate for regulatory frameworks to ensure responsible AI use during election campaigns, highlighting that artificial intelligence should complement—not replace—traditional methods of political engagement like watching news, engaging in discussions, and reviewing party platforms.