Artificial intelligence has already shown that it can imitate political figures. The more serious election risk may be that it can also persuade voters, adapt its message, and do so cheaply at large scale.
From fake voices to persuasive conversations
In January 2024, voters in New Hampshire received a call that sounded like Joe Biden telling Democrats to “save your vote” by skipping the primary. The voice was not real. It was generated by artificial intelligence.
That incident made the danger of synthetic media easy to understand. AI can create fake audio, fake video, fabricated messages from politicians and celebrities, and even convincing news-style clips. Tools like OpenAI’s Sora have made realistic synthetic video easier to produce.
But imitation is only part of the problem. The larger concern is that AI systems can hold conversations, adjust their tone, respond to emotion, and personalize political arguments. That makes AI persuasion different from a static ad or a robocall.
The source points to two large peer-reviewed studies showing that AI chatbots shifted voter views by a substantial margin. In real election contexts in the United States, Canada, Poland, and the United Kingdom, brief chatbot conversations moved attitudes by up to 10 percentage points. Among US participants, the opinion shift was nearly four times larger than the effect of tested 2016 and 2020 political ads.
When models were explicitly optimized for persuasion, the shift reached 25 percentage points. That finding is central to the concern: if political influence can be automated, personalized, and tested in real time, the campaign environment changes.
Why scale changes the threat
Online influence operations once required large teams running fake accounts, meme operations, and coordinated posting. AI can automate much of that work. One system could write political messages, another could generate images or video, and another could distribute content while measuring what works.
The same technology behind customer service bots, tutoring apps, social media tools, language learning apps, dating platforms, and voice assistants could be used to shape political views. The persuasion may not always look like an ad. It could be embedded into ordinary digital interactions.
The source emphasizes the low cost. For less than a million dollars, anyone could generate personalized conversational messages for every registered voter in America, using an assumption of 10 brief exchanges per person, around 2,700 tokens of text, and current rates for ChatGPT’s API. The 174 million registered voters in America could be reached for under $1 million under that calculation.
The source also notes that the 80,000 swing voters who decided the 2016 election could be targeted for less than $3,000. That cost structure makes AI persuasion available not only to major campaigns, but also to PACs, advocacy groups, opportunists, grassroots collectives, and foreign actors.
The risk is not limited to the United States, but the source argues that the US is especially exposed because of the scale of its elections and the attention they attract from foreign actors. It warns that the next presidential election in 2028, or even the midterms in 2026, could be affected by whoever automates persuasion first.
Open models and foreign influence make control harder
Major AI providers including OpenAI, Anthropic, and Google use policies, automated safety filters, account-level monitoring, and suspensions for violations. Those controls matter, but they only apply to activity that runs through those platforms.
Open-source and open-weight models create a separate challenge. They can be downloaded by anyone with an internet connection. Although such models are usually smaller and less capable than commercial systems, research cited in the source shows that careful prompting and fine-tuning can allow them to match leading commercial systems.
That means a determined actor may not need access to a major AI platform. They may use open models and off-platform infrastructure to avoid restrictions.
The source describes early examples outside the United States. In India’s 2024 general election, tens of millions of dollars were reportedly spent on AI for voter segmentation, swing voter identification, personalized robocalls, chatbots, and related work. In Taiwan, officials and researchers documented China-linked operations using generative AI for subtle disinformation, including deepfakes and language model outputs biased toward messaging approved by the Chinese Communist Party.
The source names China, Russia, Iran, and others as foreign adversaries that already maintain networks of troll farms, bot accounts, and covert influence operators. With open-source language models, those networks could generate fluent and localized political content without human operators who understand the language or local context.
The US policy gap
US policymakers have focused heavily on deepfakes, but the source argues that they have not addressed the broader persuasive threat. The difference matters. A fake video may be visible and public. A private chatbot conversation can be harder to detect, measure, or regulate.
The European Union’s 2024 AI Act classifies election-related persuasion as a “high-risk” use case. Systems designed to influence voting behavior face strict requirements, while administrative campaign tools used for planning events or logistics are exempt.
The United States has not drawn comparable lines. The source says there are no binding rules defining a political influence operation, no external standards for enforcement, and no shared infrastructure for tracking AI-generated persuasion across platforms.
Existing US efforts are described as piecemeal. The Federal Election Commission is applying old fraud provisions, the Federal Communications Commission has proposed narrow disclosure rules for broadcast ads, and a handful of states have passed deepfake laws. Most digital campaigning remains outside those efforts.
Platform policies are also limited. Google and Meta require disclosure when political ads are generated using AI. X has remained largely silent on this, while TikTok bans all paid political advertising. These rules mainly address paid and public content, not unpaid or private persuasion campaigns.
What a stronger response could include
The source does not call for banning AI from political life. It notes that some uses may help voters, such as candidate chatbots that explain positions, answer questions, or translate complex policy into plain language. It also cites research showing that AI can reduce belief in conspiracy theories.
But it argues that the United States needs a strategy for AI persuasion. That strategy would include guarding against foreign-made political technology with built-in persuasion features. Examples in the source include a foreign-produced video game with political talking points, a social media platform whose recommendation algorithm favors certain narratives, or a language learning app that inserts subtle messages into lessons.
The source also calls for clearer rules around AI-driven persuasion, including access to computing power for large-scale foreign efforts, technical standards for systems that generate political content, and disclosure decisions for AI-generated political messaging while navigating First Amendment concerns.
Finally, it argues for a foreign policy response. Multilateral election integrity agreements could establish that states using AI systems to manipulate another country’s electorate risk coordinated sanctions and public exposure. Shared monitoring, aligned disclosure and provenance standards, coordinated takedowns, and forums like the G7 and OECD are presented as part of the needed response.
The core warning is straightforward: AI persuasion is becoming cheap, effective, and difficult to see. If election defenses remain focused only on fake media, they may miss the quieter campaigns designed to sound reasonable, familiar, and persuasive enough to change minds.