How AI Is Reshaping Voter Demands
- The integration of generative artificial intelligence into political processes is shifting how voters access information and how candidates engage with the electorate.
- Research indicates that human-AI dialogues can meaningfully alter voter perspectives on policy and presidential candidates.
- These treatment effects were found to be larger than those typically observed from traditional video advertisements.
The integration of generative artificial intelligence into political processes is shifting how voters access information and how candidates engage with the electorate. As large language models (LLMs) become primary vessels for political data, their capacity to influence voter attitudes is demonstrating a measurable impact that exceeds traditional campaign methods.
Research indicates that human-AI dialogues can meaningfully alter voter perspectives on policy and presidential candidates. In experiments conducted during the 2024 US presidential election, the 2025 Canadian federal election, and the 2025 Polish presidential election, participants who interacted with AI models advocating for specific candidates showed significant shifts in preference.
These treatment effects were found to be larger than those typically observed from traditional video advertisements. The persuasion occurs primarily through the presentation of relevant facts and evidence rather than the use of sophisticated psychological techniques.
Accuracy and Political Bias in AI Models
While AI models use evidence to persuade, the accuracy of that information varies by political orientation. Data across three different countries revealed that AI models advocating for candidates on the political right produced more inaccurate claims than those advocating for the left.
This disparity highlights a critical challenge for the stability of democratic information ecosystems. Because LLMs often operate as black boxes, the methods used to train them and the opinions they reflect remain obscured from the end user.
Shifting Information Consumption Patterns
The adoption of AI for information gathering is occurring at a pace that surpasses the initial uptake of social media. This shift is coinciding with a decline in traffic to traditional search sites and news outlets.
As of the period leading into the 2026 midterms, more than half of Americans have access to AI tools to gather information on candidates and issues. This transition places LLMs in the role of arbiters of political information, capable of personalizing data in ways that can shape voter beliefs at scale.
Beyond direct voter persuasion, AI is being utilized by firms and researchers to simulate polling results and synthesize voter opinions, further altering the traditional landscape of political research.
Institutional Responses and Electoral Management
Electoral Management Bodies (EMBs) are currently grappling with the speed at which AI is entering electoral processes. According to the International Institute for Democracy and Electoral Assistance (IDEA), AI is often being deployed faster than institutions can govern or fully understand it.
The application of AI in elections is manifesting in several distinct ways:
- Controversial Applications: The use of AI-generated mayoral contenders, deepfakes used to incriminate candidates, and AI-facilitated speeches delivered by incarcerated politicians.
- Administrative Integration: The use of AI for day-to-day electoral management, including data analysis, information provision, planning, and oversight.
- Regulatory Efforts: Attempts by model providers to implement responsible approaches to election-related topics, as seen with Google’s public commitments prior to the 2024 US presidential election.
The way EMBs choose to design and regulate these tools will have far-reaching implications for trust in public institutions and the overall integrity of democratic systems.
Broader Implications for Democracy
The transition to AI-driven political information mirrors the earlier impact of social media, which contributed to the spread of misinformation and the creation of online echo chambers. However, the scale and invisibility of AI’s influence present new risks.
The potential for AI to influence specific ballot measures has already been documented; for example, large persuasion effects were observed among Massachusetts residents regarding a ballot measure to legalize psychedelics.
As these technologies continue to evolve, the focus remains on whether institutional capacities can advance quickly enough to mitigate disruptive risks while leveraging the potential of AI to strengthen democratic processes.