**
In a revelation that could have profound implications for electoral integrity, a recent study has found that AI chatbots, when queried for voting advice, provide inconsistent and often inaccurate recommendations. Conducted during Hungary’s parliamentary elections in April, the research led by the civil liberties organisation Liberties highlights alarming deficiencies in the advice generated by popular AI models like ChatGPT and Gemini. With the nation witnessing a pivotal political shift, the findings raise critical questions about the role of artificial intelligence in democratic processes.
Misleading Recommendations: The Research Findings
The study’s findings paint a troubling picture of the reliability of general-purpose AI systems in electoral contexts. It was discovered that these chatbots frequently suggested parties that were either not contesting the election or entirely misrepresented the political landscape. For instance, when voters with profiles aligned to the opposition Tisza party sought guidance, the AI failed to recommend this party in a staggering 90% of cases. Instead, users were often directed towards smaller parties unlikely to secure parliamentary representation.
Péter Magyar’s Tisza party achieved a remarkable victory, ending Viktor Orbán’s long-standing tenure as Prime Minister. However, the research suggests that in a more competitive electoral environment, the inaccuracies presented by these AI systems could significantly influence voter behaviour and potentially sway election outcomes.
The Methodology Behind the Findings
To arrive at these conclusions, Liberties devised five distinct voter profiles, reflective of the major political parties contesting in the elections. These profiles were tested against both ChatGPT and Gemini, utilising two distinct queries: one soliciting direct advice on which party to support and another seeking a percentage match with the various parties. The results were disconcerting; ChatGPT identified Tisza as the preferred party for only 2% of the profiles aligned with its platform. In contrast, profiles aligned with Fidesz, Orbán’s party, were identified as suitable candidates in about half of the queries.
The inconsistency in responses was another significant concern. Identical voter profiles yielded contradictory recommendations across successive queries, showcasing the volatility of the AI’s outputs. Furthermore, a staggering 96% of responses included suggestions for parties that were not even on the ballot, further complicating the decision-making process for potential voters.
The Implications for Democracy
The report also emphasises the lack of transparency surrounding how these AI systems formulate their recommendations. Unlike dedicated voting advice platforms or independent media outlets, general-purpose AI does not disclose the rationale behind its political guidance. While the AI might respond with an air of authority, the underlying methods remain obscure, leading to a worrying potential for misinformation.
The researchers attribute these discrepancies to inadequacies in the AI’s training data, filters, and natural language processing capabilities. As Tisza emerged as a significant political force only after 2024, the static nature of AI training models rendered it virtually invisible in the context of the election.
Regulatory Gaps and the Need for Oversight
Liberties advocates for stringent regulatory measures surrounding AI systems that provide political advice. The EU’s AI Act and the Digital Services Act present frameworks intended to mitigate systemic risks within electoral processes; however, AI chatbots currently escape adequate regulation. The organisation is calling for the establishment of comprehensive safeguards to ensure that AI-driven political advice is transparent, accurate, and accountable.
Eva Simon, head of Liberties’s tech and rights programme, stated, “General-purpose AIs should not present opaque and unstable political matching as if it were reliable electoral guidance for voters. Democracy cannot rely on opaque systems that claim neutrality while delivering advice they cannot explain, reproduce, or guarantee to be accurate.”
Why it Matters
As the intersection of technology and politics continues to evolve, the findings from this study underscore a crucial necessity: ensuring that the tools designed to inform voters do not inadvertently mislead them. With an increasing number of citizens turning to AI for guidance in decision-making processes, it is imperative that these systems are held to rigorous standards of reliability and accountability. Failure to do so not only undermines the electoral process but poses a direct threat to the foundations of democratic engagement.
