AI in politics
Last reviewed
Sources
No citations yet
Review status
Needs citations
Revision
v2 ยท 3,244 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
No citations yet
Review status
Needs citations
Revision
v2 ยท 3,244 words
Add missing citations, update stale details, or suggest a clearer explanation.
Artificial intelligence has become an established part of modern political work, used by campaigns, governments, journalists, and ordinary citizens. Its applications range from mundane back-office tasks, such as drafting fundraising emails and sorting voter records, to the production of synthetic media capable of imitating a politician's face or voice. The same technologies that help a campaign translate a speech into a dozen languages can also be used to fabricate a recording of a candidate saying something they never said.
This dual character has made AI a focus of debate about the health of democracy. Data analytics and microtargeting shaped election campaigns well before the arrival of consumer generative AI tools, but the public release of systems such as large language models and image and voice generators in the early 2020s lowered the cost of producing persuasive political content to almost nothing. Researchers, election officials, and regulators have since concentrated on a set of overlapping risks: deepfakes of candidates, AI-assisted disinformation at scale, automated accounts that amplify messages, and the erosion of public trust that can follow even when a forgery is quickly debunked.
The year 2024 was widely described as the largest election year in history, with national votes held in more than 60 countries that together are home to about half the world's population.[1] It became an informal test of how far AI would disrupt elections. Several studies published during and after that year found that the most extreme fears, of AI-driven content swinging major election results, did not materialize, while also documenting real harms to individual candidates and to the broader information environment.[2] This article surveys how AI is used across politics, the documented threats it poses, and the policy responses that have emerged as of 2026.
The use of data and statistical modeling to target voters predates modern machine learning. United States campaigns in the 2000s built large voter files and used them to segment the electorate, and the 2008 and 2012 Barack Obama presidential campaigns were noted for their data analytics operations, which combined voter records, consumer data, and field experiments to decide which voters to contact and how. These methods relied on conventional statistics and database work rather than on the neural networks later associated with the term AI, but they established the practice of treating individual voters as targets for tailored messaging.
The most prominent early controversy over data-driven targeting involved Cambridge Analytica, a political consultancy that worked for the 2016 campaigns of Ted Cruz and Donald Trump. The firm obtained data on Facebook users through an app called "This Is Your Digital Life," built in 2013 by researcher Aleksandr Kogan, which harvested information not only from people who used it but also from their Facebook friends. The data of up to 87 million Facebook profiles was collected this way.[3] Cambridge Analytica said it used such data to build "psychographic" personality profiles for targeting, although the firm later denied using the Kogan data or psychographic models in its work for the Trump campaign, and independent analysts have questioned how effective the techniques actually were.[3] The episode prompted regulatory action against Facebook, including a 5 billion dollar penalty imposed by the United States Federal Trade Commission in 2019, and it became a reference point in debates about privacy, microtargeting, and manipulation.[3]
The arrival of widely available generative tools marked a second phase. From around 2022, large language models able to write fluent text, along with image generators and voice-cloning systems, became accessible to anyone with an internet connection. This shifted the concern from how campaigns mine data toward how anyone, including foreign actors and individual pranksters, can manufacture convincing political content. By the 2024 election cycle, both the constructive and the malicious uses of generative AI had moved from speculation to documented practice.
AI is used across the day-to-day operations of campaigns and governments. Most of these uses are routine and attract little attention, in contrast to the deepfakes that dominate news coverage.
| Application area | How AI is used |
|---|---|
| Campaign messaging and microtargeting | Drafting social media copy, ad variants, and tailored messages for different audience segments, and analyzing how a candidate is perceived across groups. |
| Fundraising | Writing and testing fundraising emails and text messages; early testing reported that AI-aided emails raised several times more money per work hour than emails written solely by people.[4] |
| Voter analytics and opposition research | Summarizing news coverage, classifying and modeling voter data, and conducting first-pass opposition research. |
| Chatbots and constituent services | Conversational systems that answer voter questions, with field experiments suggesting informative chatbot conversations can modestly increase turnout; government chatbots that help residents with payments, registrations, and scheduling.[4][5] |
| Speechwriting and content | Producing drafts of speeches, talking points, and debate preparation, and translating speeches into other languages. |
| Polling and sentiment analysis | Analyzing social media and open data to gauge public sentiment and track how issues and candidates are discussed. |
| Government and public-sector use | Assisting legislative drafting and research, and powering chatbots that guide citizens through public services. |
Campaign messaging and microtargeting. Generative tools cut the cost of producing communications, letting campaigns generate many versions of an advertisement or message and tailor them to particular audiences. Researchers have noted that AI can trawl the web to assess how a candidate is perceived in different communities, summarize large volumes of news, and write copy adapted to specific segments.[4]
Fundraising. Routine fundraising appeals are among the most common uses, because they are repetitive and easy to test. One group working to help Democratic campaigns reported that AI-aided fundraising emails generated roughly three to four times more dollars per work hour than emails written entirely by people.[4]
Chatbots and constituent services. Conversational AI is used both to reach voters and to serve constituents. Randomized trials in the United States have indicated that simple chatbots that discuss practical matters such as how to vote can increase turnout, particularly when the conversation is informative.[4] In government, chatbots have become a common tool: a 2024 survey of United States cities found that close to half already used chatbots, with many more planning to adopt them, often to handle payments, registrations, and appointment scheduling.[5]
Government and public-sector use. Beyond service chatbots, several legislatures have experimented with AI to support lawmaking. The Italian Chamber of Deputies has supported a project to help summarize amendments and check bills against drafting standards, and the Brazilian Chamber of Deputies has expanded an internal system, Ulysses, that classifies legislative material.[6] These uses raise their own questions about accuracy and accountability, but they are distinct from the disinformation concerns that dominate election coverage.
Despite the breadth of possible uses, adoption in the 2024 United States cycle was more cautious than expected. Reporting found that many campaigns remained wary of generative AI, in part because of negative public perceptions, and that several campaigns that did use it preferred not to advertise the fact.[4]
The sharpest concerns about AI in politics center on deception. The core worry is that cheap, realistic synthetic media and automated amplification can mislead voters, defame candidates, and degrade trust in the information that elections depend on.
Several concrete incidents have shaped the debate. In Slovakia, two days before the parliamentary election of 30 September 2023, an audio recording spread on social media that purported to capture Michal Simecka, leader of the liberal Progressive Slovakia party, and the journalist Monika Todova of the newspaper Dennik N discussing how to rig the vote. The recording was a fabrication, and fact-checkers identified signs of manipulation, but it circulated during a pre-election moratorium period when candidates and media faced restrictions on responding, which complicated efforts to rebut it.[7]
In the United States, in the days before the New Hampshire presidential primary of 23 January 2024, thousands of voters received a robocall using an AI-generated voice imitating President Joe Biden that discouraged them from voting in the primary. A political consultant, Steve Kramer, admitted commissioning the call, saying he paid 150 dollars for the recording and intended it as a warning about the dangers of AI.[8] The episode produced a significant legal and regulatory response, discussed below, including a proposed federal fine and state criminal charges; a New Hampshire jury acquitted Kramer of all charges in June 2025.[8][9]
The 2024 United States general election also saw lower-stakes but widely shared synthetic content. In August 2024, Donald Trump posted AI-generated images on his Truth Social account that depicted supporters in "Swifties for Trump" shirts and a fabricated image suggesting the musician Taylor Swift endorsed him, captioned "I accept!"[10] Swift subsequently endorsed Kamala Harris and cited the fake images as a reason for speaking publicly, writing that the episode "conjured up my fears around AI, and the dangers of spreading misinformation."[10]
Outside the United States, generative AI featured heavily in the 2024 Indian general election, where parties used voice and video tools to reach voters across many languages. Narendra Modi's Bharatiya Janata Party used an AI translation system to render his speeches into regional languages, and parties used cloned avatars to deliver personalized messages; one estimate put spending on authorized AI-generated campaign content at around 50 million dollars for the cycle.[11] In Indonesia, the campaign of Prabowo Subianto used a cartoon-style digital avatar to soften his public image among younger voters.[11] These examples show AI used openly as a campaigning tool rather than as covert deception.
| Incident | Year | Description |
|---|---|---|
| Slovakia audio deepfake | 2023 | Fabricated audio of Progressive Slovakia leader Michal Simecka and a journalist, released two days before the election during a moratorium period.[7] |
| New Hampshire Biden robocall | 2024 | AI-voice robocall imitating President Biden urging voters not to vote in the primary; consultant Steve Kramer admitted commissioning it and was later acquitted of criminal charges.[8][9] |
| Trump "Swifties for Trump" images | 2024 | AI-generated images falsely implying Taylor Swift's endorsement, posted on Truth Social.[10] |
| India campaign AI | 2024 | Widespread authorized use of AI translation and cloned avatars by Indian parties during the general election.[11] |
The New Hampshire robocall illustrated how voice cloning can be combined with mass-calling infrastructure to deliver a deceptive message directly to voters. More broadly, generative tools allow the rapid production of fake images, fabricated news stories, and large numbers of social media posts. Automated or semi-automated accounts, sometimes called bots, can then amplify such content, a practice studied under the heading of computational propaganda. The concern is not only that any single forgery will deceive, but that an abundance of synthetic material can pollute the information environment, make genuine content harder to trust, and provide cover for politicians to dismiss authentic recordings as fakes, an effect sometimes called the liar's dividend.
Because so many elections fell in 2024, the year became a natural experiment in AI's electoral impact. Several research efforts concluded that the most catastrophic scenarios did not occur. The Centre for Emerging Technology and Security at the Alan Turing Institute analyzed major votes and reported finding no evidence that AI-enabled disinformation or deepfakes meaningfully changed the results of the United Kingdom, French, or European Parliament elections.[2] Its researchers identified a limited number of viral cases, on the order of 16 during the United Kingdom general election and 11 across France and the European Union, and found that much of the exposure tended to reinforce voters' existing beliefs rather than convert them.[2]
These assessments came with caveats. Researchers cautioned that a limited effect on outcomes did not mean the technology was harmless, pointing to damage to the integrity of public debate and to the targeting of individual candidates, with women and members of ethnic minorities disproportionately affected by abusive or sexualized deepfakes.[2] Commentators also noted that the absence of a decisive AI "October surprise" in 2024 offered no guarantee about future elections as the tools improve.[2]
Governments and platforms have responded to these risks with a patchwork of rules, many of them adopted in 2024. As of 2026 there is no comprehensive United States federal law governing AI in elections, and much of the activity has occurred at the level of individual agencies, states, and the European Union.
United States federal agencies. On 8 February 2024, the Federal Communications Commission issued a unanimous declaratory ruling stating that calls using AI-generated voices are covered by the Telephone Consumer Protection Act, effectively making unsolicited AI-voice robocalls illegal and giving the agency clearer authority to act.[12] The ruling followed the New Hampshire robocall. The FCC proposed a 6 million dollar fine against Steve Kramer, and Lingo Telecom, the carrier that transmitted the calls, agreed to pay 1 million dollars in a settlement in August 2024.[9] The FCC also proposed, in mid-2024, a rule that would require disclosure of AI use in political advertisements on broadcast radio and television, though that proposal remained contested.[12] The Federal Election Commission, by contrast, declined on 19 September 2024 to open a new rulemaking specifically on AI in campaign ads, instead adopting an interpretive rule clarifying that its existing ban on fraudulent misrepresentation is technology neutral and applies to AI-assisted media as it would to other means.[13]
United States state laws. States moved faster than Congress. By the end of 2024, roughly 20 states had enacted laws addressing deepfakes in elections, and related bills had been introduced in dozens more.[14] Most of these laws require that manipulated political content be disclosed or labeled, while a smaller number, including California, Minnesota, and Texas, took a more prohibitive approach for certain deceptive election content.[14] California enacted several measures in 2024 aimed at AI election content, including a disclosure requirement; some provisions were later challenged in court, and in 2025 a federal judge struck down parts of California's framework, in part on the ground that they conflicted with Section 230 of the Communications Decency Act.[14]
European Union. The EU AI Act, which entered into force in 2024, includes transparency obligations relevant to political deepfakes. Under Article 50, providers of systems that generate synthetic audio, image, video, or text must mark outputs as artificially generated in a machine-readable way, and deployers who create deepfakes must disclose that the content is artificially generated or manipulated, with lighter requirements where content is plainly artistic or satirical.[15] These transparency obligations are scheduled to apply from August 2026, and the European Commission published a draft code of practice on marking and labeling AI-generated content in December 2025 to guide compliance.[15]
Platform policies. Major technology and social media companies introduced their own measures, such as labeling AI-generated content, requiring disclosure in political advertising, and in some cases restricting the use of their tools to impersonate candidates. These policies vary between platforms and have been criticized as inconsistently enforced, but they form part of the practical governance of AI in elections alongside formal law.
The debate over AI in politics weighs efficiency and access against deception and inequality. On the benefit side, AI can lower the cost of communicating with voters, help campaigns and officials translate material into many languages, ease constituent services, and support routine government work such as summarizing documents. In countries with high linguistic diversity, translation and voice tools can let politicians reach communities they could not otherwise address directly, which some commentators have argued was a net positive in the 2024 Indian election despite the prevalence of deepfakes.[11]
The risks are equally concrete. Realistic forgeries can defame candidates and mislead voters, and even debunked fakes can corrode trust. Synthetic content disproportionately targets women and minority candidates, often in sexualized forms. The liar's dividend allows bad actors to dismiss genuine evidence as fabricated. Microtargeting and opaque data practices raise privacy and manipulation concerns that long predate generative AI. There are also accuracy and accountability problems when AI systems are used inside government, for example to draft legislation or to answer citizens' questions, where errors can have direct consequences.
As of 2026, the evidence from the 2024 election cycle suggests that AI did not, on its own, decide major elections, but that it has become a permanent feature of political life that carries real risks to individuals and to public trust. The trajectory of the technology, toward cheaper, more realistic, and more accessible synthetic media, means that the gap between what can be faked and what can be detected is likely to remain a central problem. Policy is still catching up: detection and provenance tools such as watermarking and content credentials are advancing but are not yet reliable at scale, transparency rules such as the EU AI Act's labeling obligations are only beginning to take effect, and United States federal law remains fragmented. Most analysts expect the focus to shift from the question of whether AI will swing a single election toward the slower, structural challenge of preserving a trustworthy information environment as synthetic content becomes ordinary.