A recent study highlights alarming deficiencies in the guidance offered by AI chatbots to voters during Hungary’s parliamentary elections earlier this year. The research, conducted by the civil liberties organisation Liberties, reveals that popular AI systems like ChatGPT and Gemini frequently delivered inconsistent and inaccurate advice regarding party recommendations. This inconsistency not only undermines the efficacy of these technologies in electoral contexts but also poses significant risks to the integrity of democratic processes.
Inaccurate Recommendations and Unreliable Guidance
The analysis, which focused on the Hungarian elections in April 2026, found that AI chatbots often failed to suggest the correct parties, instead recommending those not participating in the election or omitting relevant options altogether. Over 90% of the time, when users provided detailed profiles aligned with the Tisza party, the AI systems neglected to recommend it, opting instead for smaller parties unlikely to meet the electoral threshold.
The study’s findings raise critical questions about the reliability of general-purpose AI (GPAI) in political contexts, particularly given that these systems do not disclose their methodologies or provide reproducible outcomes. As Eva Simon, head of Liberties’ tech and rights programme, noted, “General-purpose AIs should not present opaque and unstable political matching as if it were reliable electoral guidance.”
Analysis of AI Performance During the Elections
Liberties employed a sophisticated approach to assess AI performance, creating five distinct voter profiles based on the positions outlined in Voksmonitor, a respected Hungarian voting advice platform. Each profile was tested multiple times using both ChatGPT and Gemini, which are among the most popular AI chatbots. The results were striking: ChatGPT assigned Tisza a match score in only 2% of cases, while Fidesz-aligned profiles received a recommendation approximately 50% of the time, highlighting a clear bias in AI responses.
The researchers noted that the outputs from these AI models were often presented with an air of authority, leading users to place undue trust in their recommendations. This creates a perilous situation where voters may act on misleading information, especially in closely contested elections.
The Implications of AI in Electoral Processes
While the Hungarian elections concluded with a decisive victory for Péter Magyar’s Tisza party, the researchers caution that the inaccuracies observed could have dire consequences in more competitive or uncertain electoral situations. The inconsistency of the AI outputs—where identical prompts yielded radically different results—highlights a systemic risk that could influence voter behaviour and outcomes.
Moreover, the study noted a significant regulatory gap concerning AI technologies. Current EU frameworks, such as the AI Act and the Digital Services Act, do not adequately address the unique challenges posed by AI chatbots in electoral contexts. Liberties advocates for stricter safeguards, asserting that AI systems should not offer personalised political advice unless they can ensure transparency, accuracy, and accountability.
Why it Matters
The findings from this study serve as a stark reminder of the potential pitfalls of integrating AI into the democratic process. As reliance on technology in electoral systems increases, ensuring the reliability of AI recommendations becomes paramount. If voters cannot trust the guidance provided by these systems, the very foundation of informed electoral participation is at risk. Establishing robust regulatory frameworks to govern the use of AI in politics is crucial to safeguarding democracy and maintaining public confidence in electoral outcomes.
