A study by civil liberties organization Liberties examining AI chatbot voting recommendations during Hungary’s parliamentary elections has revealed significant accuracy and consistency problems. Researchers tested ChatGPT and Gemini using voter profiles aligned with five registered parties, running each profile through both systems 10 times using different prompting approaches. The findings indicated that general-purpose AI systems frequently misclassified voter preferences, failed to recommend appropriate parties, and in 96% of cases included parties not even competing in the 2026 ballot.
The most striking results involved bias toward established parties. When presented with voter profiles aligned with Péter Magyar’s opposition Tisza party, ChatGPT failed to recommend it in 90% of cases where direct advice was requested. By contrast, profiles aligned with Viktor Orbán’s Fidesz party were recognized and recommended far more consistently, with ChatGPT identifying it as the primary option in approximately 50% of direct advice prompts. Researchers attributed this disparity partly to training data limitations, as Tisza rose to prominence after 2024 and static training models struggled to incorporate the newer party.
The research highlighted additional concerns about transparency and consistency. Both AI systems provided responses preceded by disclaimers stating they could not offer political advice, yet followed these disclaimers with detailed, authoritative-sounding party recommendations. Identical voter profiles yielded materially different results across successive tests, undermining reliability. Unlike dedicated voting advice websites or independent media analysis, these AI systems disclosed neither their recommendation methodologies nor underlying data sources, and faced no election-specific regulatory oversight.
Researchers noted that while AI chatbots appear confident and precise to users, their opaque processes and unstable outputs create significant risks. The study warned that in closer electoral contests where voters lack strong preferences, such inaccurate guidance could potentially influence outcomes. Liberties called for regulatory action and recommended that AI providers cease offering personalized voting recommendations unless they could guarantee transparency, accuracy, consistency and accountability. The organization argued that current EU regulations create a gap where AI chatbot political advice falls outside existing oversight frameworks.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI