A recent study indicates that AI chatbots might surpass traditional political advertisements in shaping voters’ opinions and preferences. With evidence of a consistent left-leaning bias, these chatbots could significantly influence the upcoming 2026 midterm elections.
A study conducted by researchers from various universities has shown that AI chatbots can have a remarkable impact on changing voters’ political perspectives, even more so than conventional political ads. Findings from this research, shared in esteemed journals nature and science, highlight the potential of generative AI to alter the electoral landscape soon.
The study involved over 2,300 participants engaging with chatbots a couple of months before the 2024 U.S. presidential election. The AI models were designed to lean toward one of the leading candidates and demonstrated notable persuasive abilities, particularly when discussing key policies regarding the economy and healthcare.
The outcomes were striking. For instance, Trump supporters interacting with an AI focused on Kamala Harris reflected a 3.9-point shift favoring Harris on a 100-point scale—four times greater when compared to the effectiveness of political ads from previous elections. Conversely, an AI advocating for Trump managed to shift Harris supporters by 2.3 points towards Trump.
Additional experiments related to the 2025 Canadian federal election and the Polish presidential election demonstrated even larger effects, with the chatbots altering opposition voters’ attitudes by nearly 10 percentage points.
In 2024, a report highlighted an academic investigation of the most popular large language models (LLMs), revealing that these models often exhibit fundamentally left-wing biases when they undergo political orientation tests.
The study confirmed that all tested LLMs consistently produced responses aligned with progressive and environmentally friendly ideologies, reflecting values like equality and a global perspective.
To delve deeper, Rozado modified GPT-3.5 and ran further tests. He created two variants: LeftWingGPT, trained with left-leaning content, and RightWingGPT, which focused on right-leaning material. This experiment suggested that the AI’s political inclinations can be swayed by its training data.
Another study published in science explored the aspects that enhance a chatbot’s persuasive power. Researchers analyzed interaction data from approximately 77,000 UK participants across more than 700 political issues. They discovered that instructing the AI to incorporate facts and examples of successful persuasion significantly boosted its convincing abilities. In one case, a particularly persuasive model shifted participants’ views by an impressive 26.1 points.
However, as these models grow more adept at persuasion, they also risk perpetuating misleading or false information, raising alarms about their potential effects on democracy. Political campaigns employing AI chatbots could steer public opinion in ways that might hinder voters’ capacity for independent judgment.
Wynton Hall, Social Media Director for Breitbart News and author of Code Red: Left, Right, China, and the Race to Control AI, emphasizes the need to comprehend how AI is leveraged to influence U.S. elections.
It’s been evident that LLMs aren’t neutral and tend to favor leftist ideologies. The findings validate how effectively AI chatbots can function as tools for political persuasion, ready to employ deceptive means if required to sway opinions. This combination of bias and persuasive techniques raises significant concerns for conservatives as they approach upcoming elections.
