What happened

In the weeks leading up to the Dutch national elections, the Dutch Data Protection Authority (AP) tested four popular AI chatbots and found that they disproportionately recommended parties at the far right and far left edges of the political spectrum, such as the far-right Freedom Party (PVV) and the left-wing GroenLinks-PvdA, regardless of voter profile.

Centrist parties were rarely mentioned, even when user queries matched their positions. 

This led to concerns that voters could be misled, with the advice received distorting public opinion and potentially affecting the integrity and fairness of elections. 

"AI chatbots give a highly distorted and polarised image of the Dutch political landscape," the regulator warned.

Why it happened

The core issue stems from the way AI chatbots process and present voting advice: their algorithms favour parties with stronger policy stances, resulting in polarised and oversimplified guidance.

Furthermore, chatbot makers have not put in sufficient safeguards to prevent misleading political advice. 

In addition, users cannot see how chatbot recommendations are generated, making bias hard to detect or correct, and meaningful accountability is undermined by the proprietary nature of these systems.