Study finds AI chatbots give inaccurate voting advice in Hungary election

A recent study released by the civil liberties organization Liberties has found that general‑purpose AI chatbots failed to provide reliable guidance to voters during Hungary’s 2024 parliamentary election. Researchers tested the systems by posing identical prompts asking which party a voter should support, and recorded the responses. The chatbots frequently recommended parties that were not contesting the election, omitted the major parties that were actually on the ballot, or listed multiple parties with contradictory endorsements.

The analysis revealed a pattern of volatility and inconsistency across the AI platforms. In many cases, the same question produced different party suggestions from the same chatbot, while other systems offered entirely different answers to the same prompt. The study highlighted that such inaccuracies could mislead voters, undermining the integrity of the electoral decision‑making process. “The results raise serious concerns about the reliability of general‑purpose AI systems in electoral contexts,” the report concluded.

The findings underscore the need for stricter oversight of AI tools used in political contexts, particularly as their use becomes more widespread. While the study focused on the Hungarian election, its implications extend to any democratic process where voters may seek automated advice. Regulators and developers are urged to address these reliability gaps to safeguard informed voter choice.