**
In a striking revelation from recent research, AI chatbots tasked with providing electoral advice have been found to deliver inaccurate and unreliable recommendations to voters. The study, conducted during Hungary’s parliamentary elections, raises critical questions about the efficacy of general-purpose AI systems in guiding citizens through the tumultuous waters of political choice.
The Study’s Surprising Findings
This eye-opening report, released by the civil liberties organisation Liberties, scrutinised the advice dispensed by popular AI chatbots, specifically ChatGPT and Gemini, during the recent elections in Hungary. It found a shocking trend: these systems often suggested parties that were either not running or completely misaligned with voters’ profiles.
In particular, the Tisza Party, which emerged victorious in the elections, was overlooked in nearly 90% of the evaluations when users provided detailed profiles aligned with its platform. Instead, voters were frequently directed towards smaller parties unlikely to meet the 5% threshold required for parliamentary representation. This misalignment not only questions the chatbots’ reliability but also highlights a potential threat to informed voting.
The Lack of Transparency in AI Guidance
One of the most concerning aspects of the findings is the opaque nature of how AI systems generate their political recommendations. Unlike dedicated voting advice platforms, these chatbots do not disclose their decision-making processes. Their outputs often appear authoritative and well-reasoned, which can mislead users into believing that the information is trustworthy, despite its inherent inconsistencies.
The researchers established five distinct voter profiles, each corresponding to different political parties in Hungary. They tested these profiles by querying the chatbots multiple times, revealing alarming discrepancies in responses. For instance, ChatGPT identified Fidesz, the ruling party at the time, as the ideal choice for Fidesz-aligned profiles approximately 50% of the time, while Tisza was hardly acknowledged.
Implications for Future Elections
While the report does not claim that the election outcome was directly influenced by these AI-generated recommendations—Tisza triumphed decisively—it raises significant concerns about potential impacts in more competitive electoral environments. The researchers warned that in scenarios where voter uncertainty is heightened, the misleading guidance from AI could sway results in unforeseen ways.
The study also detailed the erratic nature of the chatbots’ responses, noting that identical queries could yield radically different answers. Furthermore, in a staggering 96% of cases, responses included parties that were not on the 2026 ballot.
The Call for Regulatory Oversight
Given these findings, the researchers advocate for stronger regulatory frameworks surrounding AI systems that offer political advice. Current EU legislation, including the AI Act and Digital Services Act, leaves significant gaps in accountability. As these chatbots do not fit neatly within existing regulations, the report calls for the development of safeguards to ensure that AI recommendations are transparent, accurate, and consistent.
Eva Simon, head of Liberties’ tech and rights programme, emphasised the necessity for AI systems to refrain from presenting opaque and unstable political recommendations as reliable guidance. “Democracy cannot rely on opaque systems that claim neutrality while delivering advice they cannot explain, reproduce, or guarantee to be accurate,” she stated.
Why it Matters
The implications of this study extend far beyond Hungary’s borders, posing critical questions about the role of AI in democratic processes worldwide. As more voters turn to technology for guidance in an increasingly complex political landscape, the need for accurate and transparent AI-driven advice becomes paramount. Without robust oversight, the risk of misinformation and manipulation could undermine the very foundations of democratic participation. As we venture deeper into the digital age, ensuring that technology serves to empower voters rather than mislead them is essential for the health of democracies everywhere.
