Artificial Intelligence update: Research during Hungary election shows recommended parties

Election voting advice from AI chatbots has come under fresh scrutiny after research linked to Hungary’s parliamentary election found general-purpose systems could be inconsistent, misleading and, in some cases, plainly wrong.

The study, published by the civil liberties group Liberties and shared with The Guardian, examined how chatbots responded to voter-profile prompts during this year’s Hungarian election. Its central conclusion was blunt: the systems did not provide reliable electoral guidance. According to the report, they could misclassify user profiles, omit relevant parties and even include parties that were not standing in the election.

What the study found about election voting advice from AI chatbots

One of the clearest concerns raised by the research was variability. The report said identical prompts could produce materially different answers, suggesting the advice offered to voters was not stable enough to be treated as dependable guidance.

The findings also pointed to party-selection errors. In some cases, chatbots recommended the wrong party. In others, they failed to mention the party that best matched the profile provided. The report also said the systems sometimes listed parties that were not running in the election at all.

That combination, the researchers argued, makes the tools poorly suited to electoral settings where accuracy and consistency are essential.

Why the Hungary case matters

The study was based on Hungary’s parliamentary election, which was won by Péter Magyar’s opposition Tisza party. The result ended Viktor Orbán’s 16-year hold on power.

That political backdrop matters because it shows how election-related AI advice can intersect with high-stakes democratic decisions. When a chatbot is asked which party a voter should support, the answer is not just informational. It can shape political understanding, influence comparisons between parties and potentially affect how people approach the ballot box.

The report’s most striking finding concerned Tisza. It said that when ChatGPT was given a detailed voter profile aligned with the party, it failed to recommend Tisza in 90 percent of cases. The researchers described this as evidence of a major visibility bias.

For the study’s authors, the problem was not only that a chatbot could be wrong once. It was that the output could be unreliable in a systematic way, producing different answers to the same question and mishandling the basic structure of an election.

Broader concerns for voters and election integrity

The report said the results raise serious concerns about the reliability of general-purpose AI systems in electoral contexts. That warning is significant because these tools are increasingly used for everyday information searches, including politically sensitive questions.

Unlike specialist election services, general-purpose chatbots are designed to respond broadly across topics. The research suggests that when they are asked to act as voting advisers, they may not consistently distinguish between verified election information and incomplete or inaccurate outputs.

For voters, that creates a risk of confusion. For campaigns and election watchdogs, it raises questions about how these systems should be tested, monitored or explained when used in democratic settings.

The study does not solve those policy questions, but it does add evidence to a growing concern: AI chatbots may be able to answer election questions quickly, yet speed does not guarantee reliability.

As elections continue to draw in more digital tools and more online information seekers, the findings from Hungary suggest caution is needed before treating chatbot-generated voting advice as trustworthy guidance.