Another study indicates that AI has negative implications for elections, and the situation continues to deteriorate.

      Another election, another study shows that seeking political advice from an AI chatbot is an exceptionally poor choice. Research conducted during Hungary’s 2026 parliamentary election revealed that ChatGPT and Google Gemini provided not only inaccurate but also inconsistent and unreliable voting recommendations. The chatbots miscategorized voter profiles, neglected relevant parties, suggested parties that were not on the ballot, and sometimes gave significantly different responses even when presented with the same information multiple times.

      AI continues to falter in the voter assessment

      The Civil Liberties Union for Europe created five fictitious voter profiles based on the stances of the five parties competing for Hungary’s national seats. Each profile underwent multiple tests using prompts asking for direct voting advice and percentage-based party correlations.

      In 90% of tests involving a detailed Tisza-aligned profile, ChatGPT failed to suggest the opposition Tisza party. During the percentage-matching evaluations, Tisza received a score in only 2% of cases. Views aligned with Fidesz were recognized much more consistently, while 96% of ChatGPT and Gemini responses included at least one party that was not listed on the 2026 ballot.

      The researchers found no indication that this imbalance was intentionally engineered, nor did they claim that chatbot responses influenced the election outcome. Gaps in training data, safety filters, limitations in language processing, and the quick rise of Tisza post-2024 might have all played a part. However, these explanations offer little reassurance to voters who receive polished and convincing recommendations from systems with opaque reasoning.

      This cautionary finding adds to a growing body of similar research. During Scotland’s 2026 election, Demos assessed five AI services with 75 election-related queries and discovered factual inaccuracies in 34.1% of the responses. ChatGPT made errors in 46.2% of cases, including incorrect election dates and eligibility criteria, nonexistent candidates, and fabricated political controversies. Nearly half of the responses lacked citations or supporting links.

      A Dutch regulator reached another concerning conclusion in 2025. Despite the nation’s diverse multiparty system, four tested chatbots directed voters toward only two major parties in 56% of interactions. Independent academic studies have also identified consistent political bias patterns in ChatGPT and Gemini, although the direction and severity can fluctuate based on models, prompts, languages, and elections.

      Politicians are already learning to manipulate responses

      Inaccuracy and ingrained bias are just part of the issue. The information that feeds these systems can also be intentionally manipulated. A recent article from The New York Times highlighted Missouri political candidate Dustin Lloyd, whose priorities were hardly represented when voters inquired about him through chatbots. Lloyd published a carefully curated question-and-answer section on his campaign website, and as a result, subsequent chatbot responses began linking his personal background to his policy objectives, demonstrating how swiftly a campaign can influence the AI-generated portrayal of its candidate.

      Keeping an accurate campaign website is not inherently wrong. However, this same method presents a clear opportunity for inflated claims, attack pages, fake organizations, and sites mainly designed to sway AI responses.

      A BBC investigation revealed how low the technical barriers can be. A journalist spent about 20 minutes creating a false blog post claiming to be the world’s greatest hot-dog-eating technology reporter. Within 24 hours, ChatGPT, Gemini, and Google’s AI Overviews were echoing parts of the fictitious narrative, while Claude did not fall for it.

      Similarly, another BBC report examined Google’s efforts to combat this emerging manipulation industry. The company has identified websites containing instructions aimed at hijacking browsing AI systems, influencing recommendations, promoting specific businesses, and potentially stealing data. Google anticipates that these indirect prompt-injection tactics will grow in scale and complexity.

      Researchers have also demonstrated that AI-enhanced search engines remain vulnerable to specially designed manipulation strategies. Techniques such as rewritten-query stuffing and segmenting promotional text doubled the manipulation rate compared to a baseline attack in one 2026 study. In response, Google has expanded its spam regulations to include attempts to distort answers generated by AI Overviews and AI Mode.

      An electoral hazard with unpredictable consequences

      AI chatbots possess several volatile characteristics. Their responses may be inaccurate, politically biased, and persuasive, while also being difficult to replicate and reliant on websites that campaigns or adversarial actors can modify. The potential risk extends far beyond a chatbot explicitly endorsing the incorrect candidate.

      Electoral harm could start with a missing party, outdated voting information, a fabricated scandal, or a strategically placed page that becomes sanitized into a seemingly authoritative answer. AI has evolved into an electoral bomb, with politicians, platforms, researchers, and opportunistic manipulators all poised to pull the trigger—and the extent of the impact remains uncertain.