A study based on Hungary's parliamentary elections has found that AI chatbots provide inaccurate, inconsistent, and unreliable guidance to voters, often recommending the wrong party or listing parties not even running. The research, conducted by the civil liberties group Liberties, tested ChatGPT and Gemini during the April 2026 elections, which saw opposition leader Péter Magyar's Tisza party win decisively, ending Viktor Orbán's 16-year rule.
Key findings on AI inaccuracy
The report, shared with the Guardian, concluded that AI systems misclassified voter profiles, omitted relevant parties, and included parties not on the ballot. In 90% of cases, when ChatGPT was fed a detailed Tisza-aligned voter profile, it failed to recommend Tisza. Instead, it often suggested smaller parties unlikely to cross the 5% parliamentary threshold or parties not running. By contrast, Fidesz-aligned profiles were recognized more consistently, with ChatGPT identifying Fidesz as the single party to vote for in about 50% of direct advice prompts.
Volatility and lack of transparency
The outcomes were highly volatile, with materially different answers given to identical prompts. In percentage-matching tests, ChatGPT assigned Tisza a score in just 2% of cases. The researchers noted that unlike voting advice sites, general-purpose AI systems do not disclose how they generate guidance, yield reproducible results, or are subject to election-related oversight. Despite this, their responses often appear confident and authoritative, increasing user trust. With 29.8% of Hungary's population identified as AI users, the study tested five voter profiles based on the Voksmonitor app, each aligned with one of the five parties on national lists.
Regulatory gaps and risks
The report argues that the errors likely stem from training data gaps, filters, and language-processing limitations, as Tisza surged to prominence only after 2024. The researchers warn that in a more competitive election, such outputs could potentially impact the outcome. They also exposed a regulatory gap: the EU's AI Act and Digital Services Act cover systemic risks, but AI chatbots fall between them. Liberties recommends developing safeguards for AI systems providing political advice and urges providers to stop offering personalized voting recommendations unless they can guarantee transparency, accuracy, consistency, and accountability. Eva Simon, head of Liberties' tech and rights programme, stated: 'General-purpose AIs should not present opaque and unstable political matching as if it were reliable electoral guidance. Democracy cannot rely on opaque systems that claim neutrality while delivering advice they cannot explain, reproduce or guarantee to be accurate.'
