AI chatbots deliver inaccurate election advice, raising concerns about electoral interference

AI Chatbots Fail to Deliver Reliable Election Advice

AI chatbots are delivering inaccurate information to voters seeking electoral guidance, according to research conducted during Hungary's parliamentary elections. The study, published by the Civil Liberties Union for Europe (Liberties), tested ChatGPT and Gemini and found both systems provided inconsistent, unreliable, and often factually incorrect recommendations to voters asking which party to support1

1

. The findings expose how AI chatbots providing political advice can misclassify voter profiles, omit relevant parties, and include parties not even running in elections2

2

.

Researchers created five distinct voter profiles based on party positions outlined on Voksmonitor, a respected Hungarian voting advice app, with each profile aligned to one of the five parties registered on national lists. Each profile was tested 10 times in both ChatGPT and Gemini using two different request types: direct advice on which party to back and percentage-based matching with each party1

1

. The outcomes were highly volatile, with materially different answers given to identical prompts, raising serious concerns about the reliability of general-purpose AI systems in electoral contexts.

Striking Political Bias Against Opposition Party

The most concerning finding revealed an overwhelming bias in party visibility. In 90% of cases when ChatGPT received a detailed Tisza-aligned voter profile, it failed to recommend the opposition Tisza party that ultimately won the election convincingly1

1

. During percentage-matching tests, ChatGPT assigned Tisza a score in just 2% of cases. Instead, users with Tisza-aligned views were routinely recommended smaller parties unlikely to cross the 5% threshold for parliament, or parties not even on the national ballot1

1

.

By contrast, Fidesz-aligned voter profiles were recognized far more consistently. In direct advice prompts, ChatGPT identified Fidesz as the single party users should vote for in approximately 50% of cases, presenting it as a primary option in the remainder1

1

. This disparity highlights how AI is bad for elections when systems cannot accurately represent the full political landscape, particularly newer or rapidly emerging parties.

Widespread Inclusion of Non-Existent Candidates

Both AI systems regularly included parties that were not on the 2026 ballot. In 96% of responses from ChatGPT and Gemini, at least one party that was absent from the election appeared in recommendations2

2

. This represents a fundamental failure in providing accurate electoral guidance, as voters could be directed toward options that simply don't exist. The same voter profile was also matched with radically different parties in successive tests, demonstrating the unstable nature of AI-generated political recommendations1

1

.

Both AI models regularly began responses with polite disclaimers stating they "cannot give political advice" before providing several paragraphs of dense, highly persuasive party recommendations1

1

. The answers appeared well-argued, precise, and authoritative, creating a significant risk that users may treat the outputs as reliable even though the underlying method remains opaque and results are unstable.

Pattern of AI-Generated Misinformation Across Multiple Elections

The Hungarian findings join a growing body of evidence showing AI poses risks to elections globally. During Scotland's 2026 election, research organization Demos tested five AI services with 75 election-related questions and found factual errors in 34.1% of responses. ChatGPT produced errors 46.2% of the time, including incorrect election dates and eligibility rules, invented candidates, and fabricated political scandals2

2

. Almost half of those responses provided no citation or supporting link.

A Dutch regulator reached another troubling conclusion in 2025, finding that despite the country's broad multiparty system, four tested chatbots directed voters toward only two prominent parties in 56% of interactions2

2

. Separate academic investigations have also found recurring political preference patterns across ChatGPT and Gemini, though the direction and severity can change between models, prompts, languages, and elections.

Campaigns Already Learning to Manipulate AI Responses

The information feeding these systems can be deliberately shaped, creating new vulnerabilities. Missouri political candidate Dustin Lloyd published a carefully structured question-and-answer page on his campaign website, and subsequent chatbot responses began connecting his personal background with policy goals, demonstrating how quickly campaigns can alter AI-generated candidate profiles2

2

.

A BBC investigation showed how easily misinformation spreads through these systems. A journalist spent roughly 20 minutes publishing a fabricated blog post with false claims. Within 24 hours, ChatGPT, Gemini, and Google's AI Overviews were repeating parts of the invented story2

2

. Researchers have also documented that AI-enhanced search engines remain susceptible to specially developed manipulation techniques, with strategies including rewritten-query stuffing doubling the manipulation rate compared with baseline attacks in one 2026 study.

Critical Regulatory Gaps Leave Voters Unprotected

The report identifies serious regulatory gaps in current frameworks. Unlike voting advice sites or independent media explainers, general-purpose AI systems do not disclose how they generate political advice, do not yield reproducible results, and are not subject to election-related public oversight1

1

. With 29.8% of Hungary's population identified as AI users, the scale of potential impact is significant.

The EU AI Act obliges providers of general-purpose AI models to assess systemic risks, and the Digital Services Act covers systemic risks to electoral processes, but AI chatbots fall between the two frameworks1

1

. Eva Simon, head of Liberties's 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"1

1

.

The underlying causes likely stem from training data gaps, filters, and language-processing limitations. Tisza surged to prominence only after 2024, meaning static AI training models struggled to place it accurately1

1

. While researchers found no evidence the outcome of Hungary's election was affected—Tisza still won decisively—they warn that in more competitive elections or where voters are less certain, such outputs could potentially impact results. Liberties recommends that AI providers should stop offering personalized voting recommendations unless they can guarantee transparency, accuracy, consistency, and accountability.