AI in Elections and Civic Life

Elections are the mechanism through which democratic societies translate citizen preferences into governing power. They depend on several conditions: voters receiving information adequate to make meaningful choices, the integrity of the vote-counting process, and the ability of citizens to participate without facing manipulation or coercion. AI is affecting each of these conditions in ways that democratic societies are only beginning to grapple with. This lesson examines how AI is used in political campaigns, how it is being used — and potentially misused — to influence public opinion, what it means for the mechanics of voting and election administration, and how citizens and institutions can respond.

AI-Powered Political Targeting

Political campaigns have used data analytics for decades. What AI has changed is the precision, scale, and personalization of voter targeting. Modern campaigns combine voter registration data, consumer data, social media behavior, and survey responses to build predictive models of individual voters: their likely party affiliation, their probability of turning out, their positions on specific issues, and their psychological profiles. The Cambridge Analytica case (2018) brought public attention to the scale of this practice. The firm claimed to have built psychographic profiles of 87 million Facebook users — characterizing voters on dimensions of personality (openness, conscientiousness, extraversion, agreeableness, neuroticism) — and used those profiles to tailor political advertising to individual psychological vulnerabilities. The extent of Cambridge Analytica's actual influence on election outcomes remains disputed among researchers; what is not disputed is that the data was harvested without users' meaningful consent and used for purposes they did not anticipate. AI has substantially increased the capability for this kind of micro-targeting. Generative AI can produce personalized political messages at scale — individually tailored versions of the same argument, each formulated to resonate with a specific recipient's known interests and emotional profile. This raises a question about authenticity: if every voter receives a different version of a campaign's message, optimized for their specific psychology, is that democratic communication or manipulation?

Coordinated inauthentic behavior — the operation of fake accounts, bot networks, and artificial amplification systems to manufacture the appearance of grassroots support — has become a documented feature of contemporary political influence operations. The Stanford Internet Observatory and similar research centers have documented hundreds of influence operations across dozens of countries, operated by state and non-state actors, using AI-generated content and automated accounts to flood public discourse with coordinated messaging. The intent of these operations is not primarily to persuade individual voters. It is to distort the perceived distribution of opinion — to make fringe positions appear mainstream, to make manufactured outrage appear organic, to make the public conversation look like something it is not. This is a form of epistemic pollution: corrupting the information about public opinion that citizens use to calibrate their own views and their sense of the possible. Generative AI dramatically lowers the cost of producing this kind of content. A single operator with access to language models can generate thousands of plausible posts, comments, and replies, in multiple languages, across multiple platforms, without any human writing each individual message. The scale advantage historically held by large, well-resourced influence operations is now accessible to much smaller actors.

Election Administration and AI

AI is also being explored in election administration — the mechanics of running elections rather than the politics of campaigns. Applications include: ballot scanning and tabulation software that uses computer vision to read marked ballots, anomaly detection systems that flag unusual patterns in vote reporting, and cybersecurity systems that protect election infrastructure from intrusion. These applications are largely beneficial when implemented transparently and subject to robust audit mechanisms. Automated ballot scanning has substantially reduced processing time and human error. Anomaly detection can identify genuine errors or intrusion attempts. However, election administration AI also introduces risks. Any software system can fail or be compromised. The opacity of algorithmic systems can make public verification of results more difficult, which matters enormously in a context where trust in election integrity is itself a civic resource. The response to these risks is not to avoid technology in election administration but to insist on transparency, paper trails, human oversight, and independent auditing — principles that apply to all voting infrastructure regardless of whether it involves AI.