The civil liberties organization Liberties has found that leading AI chatbots, including ChatGPT and Gemini, provided inaccurate, inconsistent, and unreliable voting advice during Hungary's 2026 parliamentary election. Researchers discovered that the chatbots frequently misclassified voter preferences, recommended parties that were not even on the ballot, and produced different answers when given identical prompts, raising concerns about their reliability for political guidance.
One of the study's most striking findings was that ChatGPT failed to recommend the opposition Tisza party in about 90% of tests using voter profiles aligned with the party, despite Tisza ultimately winning the election. Researchers attributed many of the errors to outdated training data, limited knowledge of newer political parties, and the probabilistic nature of large language models, which can generate varying responses to the same question.
The report also found that both chatbots frequently suggested parties that were not contesting the election and often presented their recommendations with unwarranted confidence, without explaining how they reached their conclusions. Researchers warned that with nearly a third of Hungary's population using AI tools, inaccurate political advice could influence voter decisions, particularly in closely contested elections.
The authors are calling for stronger safeguards to prevent general-purpose AI systems from acting as voting advisors unless they can provide transparent, consistent, and accountable recommendations. They argue that while AI can help users understand political issues, election-related decisions should rely on official election authorities, verified information sources, and multiple independent references rather than chatbot-generated advice alone.
