AI Chatbots Offer Unreliable Voting Advice – Lessons from Hungary’s 2026 Election

A research project by the civil- liberties organization Liberties examined how two leading general-purpose AI chatbots—ChatGPT and Gemini—responded to five fabricated voter profiles aligned with each of Hungary’s five parties on the national list for the 2026 parliamentary election. Each profile was queried ten times for direct party recommendation and ten times for a percentage-match score.

The Guardian

• Tisza-aligned profiles received a recommendation for the Tisza party in only 2 % of percentage-matching tests, and the party was omitted in 90 % of direct-advice prompts.

The Guardian

• Fidesz-aligned profiles were identified correctly far more often, with ChatGPT naming Fidesz as the sole recommendation in roughly 50 % of direct-advice queries.

The Guardian

• In 96 % of all responses, the models suggested parties that were not on the 2026 ballot.

The Guardian

• Identical prompts produced “materially different” answers across repetitions, indicating high volatility.

The Guardian

The study concluded that the chatbots’ guidance was “inaccurate, inconsistent and unreliable,” raising “serious concerns about the reliability of general-purpose AI systems in electoral contexts.”

The Guardian

AI-driven voting advice is emerging globally. Prior to the United Kingdom’s May 2026 local elections, one in seven voters said they were open to using chatbots for voting decisions, with 5 % already doing so. In the Netherlands, the data-protection authority warned against chatbot advice after a study found “vastly over-recommended two out of 15 parties.” Brazil’s Superior Electoral Tribunal (TSE) responded by imposing a blanket ban on chatbot voting advice in March 2026.

Chatbot Voting Advice Could Help and Harm Voters. Let’s Regulate Accordingly.

• 29.8 % of Hungary’s population were identified as AI users during the election period.

The Guardian

• ChatGPT’s percentage-matching assigned a Tisza score in 2 % of cases; Fidesz appeared in ?50 % of direct-advice prompts.

The Guardian

• 96 % of chatbot replies included at least one non-ballot party.

Chatbot Voting Advice Could Help and Harm Voters. Let’s Regulate Accordingly.

Liberties’ report highlights a regulatory gap: the EU’s AI Act obliges providers to assess systemic risks, while the Digital Services Act addresses “systemic risks” to electoral processes, yet chatbots fall between the two frameworks. The organization calls for safeguards that would require AI providers to guarantee transparency, accuracy, consistency and accountability before offering personalized voting recommendations.

The Guardian

Brazil’s TSE ban demonstrates a decisive, albeit broad, governmental reaction, while the EU’s Regulation on the Transparency and Targeting of Political Advertising mandates disclosure of who authorized political ads and how they were targeted. California’s attempt to require independent AI audits was vetoed by Governor Gavin Newsom in September 2024.

Chatbot Voting Advice Could Help and Harm Voters. Let’s Regulate Accordingly.

Experts warn that chatbot advice can be shaped by advertiser influence, hallucinated facts, and entrenched societal biases (e.g., sexism, racism). The lack of up-to-date training data—illustrated by Tisza’s emergence after 2024—means static models may omit newer parties entirely. Potential threats also include deliberate data-poisoning attacks and the possibility of owners promoting specific parties, as seen with the “Grok” model’s refusal to avoid “politically incorrect” content.

Chatbot Voting Advice Could Help and Harm Voters. Let’s Regulate Accordingly.

The Hungarian study does not claim that chatbot advice altered the election outcome; Tisza still won decisively. However, researchers caution that in tighter races or among less-certain voters, such outputs could “potentially impact the outcome.” The study also lacked insight into how many voters actually consulted chatbots, leaving a gap in assessing real-world influence.

The Guardian

• “The results raise serious concerns about the reliability of general-purpose AI systems in electoral contexts,” — Liberties report

The Guardian

• “They misclassified profiles, omitted relevant parties and included parties not running in the election,” — Liberties report

The Guardian

• “In a more competitive election or where voters are less certain, such outputs could … potentially impact the outcome.” — Liberties report

The Guardian

• “General-purpose AIs should not present opaque and unstable political matching as if it were reliable electoral guidance” — Eva Simon, head of Liberties’s tech and rights programme

The Guardian

Liberties recommends that AI providers stop offering personalized voting recommendations unless they can ensure transparent, reproducible, and accurate guidance. It also urges the EU and national regulators to close the gap between the AI Act and the Digital Services Act, and for independent audits to become a standard accountability mechanism for generative AI systems that influence democratic participation.

The Guardian