Liberties research warns against relying on AI for election guidance
Report calls for stronger safeguards for political AI.
The Civil Liberties Union for Europe (Liberties) has published research arguing that general-purpose AI systems should not be relied upon for personalised voting advice, after assessing ChatGPT and Gemini during Hungary’s 2026 parliamentary election campaign.
According to Liberties, the research found that the AI systems frequently failed to match users with the appropriate political party, produced inconsistent responses to identical prompts and often continued offering political guidance despite initially stating that they could not provide voting recommendations. The researchers also found that the systems did not explain a clear methodology for political matching or produce reproducible results.
The report argues that general-purpose AI systems differ from established voting advice applications, which typically operate under dedicated public oversight and transparent methodologies. It found that the AI systems gave little consideration to strategic voting, coalition dynamics or local electoral factors when generating recommendations.
Liberties said the findings expose governance gaps between the EU AI Act and the Digital Services Act, calling for stronger safeguards for AI systems that provide political information. The organisation recommends that AI providers direct users to trusted election information sources rather than offering personalised voting advice.
Why does it matter?
The report highlights the growing role of general-purpose AI systems as intermediaries for political information, even though they were not designed or regulated to function as voting advice services. Inaccurate, inconsistent or non-transparent recommendations could undermine informed electoral decision-making if users place undue trust in AI-generated guidance.
The findings also contribute to the broader debate over AI governance in democratic processes. As generative AI becomes more widely used during elections, policymakers may need to clarify how existing digital regulations apply to AI systems that influence political information without fitting neatly into established regulatory categories.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!