AI Moves From Experiment to Everyday Tool in the 2026 Midterms

AI-generated ads grow sharply

The most visible change is in political advertising. The Wesleyan Media Project, which tracks political ads, identified at least 220 ads produced or enhanced with AI by candidates or political committees this cycle, at a cost of just under $137 million through Oct. 11, according to reporting by NBC News and Axios. NBC reported that AI ad spending during the 2024 presidential election was under $50,000, and that spending on AI ads had already reached $20 million by May of this year.

"We really saw the growth in AI ads over the summer," Travis Ridout, a Washington State University professor and one of the project's directors, told Noticias Telemundo. "A lot of them were connected with closely fought primary races."

Reporting by Axios, CNN and The Washington Post describes ads that place AI-generated likenesses of opponents in fabricated scenes, sometimes with clear disclosures and sometimes with none. Researchers say most of the AI ads they have tracked were sponsored by Republican candidates and groups, and The Washington Post reported that some Democratic campaigns are betting the tactic will backfire with voters.

Campaigns use AI for texting, translation and outreach

Most campaign AI use is less visible than an attack ad. NPR reported in July that campaigns are using generative AI to run two-way text conversations with voters. Typically, a human writes and sends the first message, and an AI agent takes over when the recipient replies. One vendor told NPR it had sent 2.5 million texts this year and held tens of thousands of conversations. Another said its tools can respond "within 30 seconds, in any language." Experts quoted by NPR said Republicans have adopted these tools faster than Democrats, though it is hard to measure how many campaigns use them.

Vendors also advertise AI features for drafting fundraising appeals, testing message variants, translating scripts and web pages for multilingual outreach, and analyzing voter sentiment. The Center for Democracy & Technology has warned that these conversational tools are spreading faster than the rules that govern them. It noted that the Telephone Consumer Protection Act was written for prerecorded calls and mass texts rather than adaptive AI conversations, that voters may not know they are talking to a bot, and that the exchanges can be used to build detailed voter profiles.

A patchwork of state laws

No federal law specifically regulates AI-generated political ads, Axios reported. States have filled the gap unevenly. According to the National Conference of State Legislatures, 31 states have enacted laws on AI and deepfakes in campaigns and elections. Most of them require disclosure rather than banning the content outright.

A September analysis by the law firm Covington & Burling counted 29 states with election deepfake laws in effect. It noted that federal courts have blocked laws in California and Hawaii. Minnesota and Texas prohibit specified AI-generated media during defined periods before an election, and a Maryland law enacted in June targets deceptive media that spreads materially false election information. Florida requires a prominent disclaimer, "Created in whole or in part with the use of generative artificial intelligence (AI)," on qualifying deceptive content. Arizona requires disclosures on AI-manipulated content posted within 90 days of an election.

Enforcement has started. Covington noted lawsuits by candidates in Michigan and Arizona over allegedly unlawful AI content. In September, California's Fair Political Practices Commission issued a warning letter to a statewide campaign over a missing AI disclaimer on a video posted to X, and closed the case without a fine after the disclaimer was corrected.

Federal agencies and platforms

At the federal level, the Associated Press reported on Oct. 7 that the FCC is considering a petition to loosen rules on political robocalls, including calls that use AI-generated voices, before November. In July, a group of members of Congress wrote to CISA, the Department of Justice, the FEC and the Department of Homeland Security urging coordination on AI risks to the midterms.

Platform policies vary. Meta requires advertisers to disclose AI use in certain political and social-issue ads, labels detected AI-generated organic content, and blocks new political ads during the final week of the campaign. OpenAI says it prohibits using its products to create or distribute scaled campaign messaging. A Princeton Center for Information Technology Policy review found that Google requires election advertisers to complete verification and disclose realistic synthetic content, and that Anthropic prohibits automated campaign messaging and synthetic media. It also noted that X uses civic-integrity labels. Several AI companies say they direct voting-procedure questions to sources such as Democracy Works or provide results from the Associated Press.

Election offices and security

For election administrators, AI is mostly a security concern. In an Oct. 16 public service announcement, the FBI and CISA warned that attackers are using AI to make phishing and smishing campaigns faster, larger and more convincing, including with audio and video content. They noted that election officials, candidates and political organizations are attractive targets because of their access to voter registration data. The U.S. Election Assistance Commission has cautioned that AI-generated content may be used to impersonate election officials. It also warns that AI tools often lack the accuracy needed for critical information such as voting dates, hours and locations.

What to watch before Election Day

In the final stretch, observers will watch whether more undisclosed AI ads appear, how states enforce their disclosure laws, whether the FCC acts on the robocall petition, and how platforms apply their late-campaign ad restrictions. Election officials and security experts have the same advice for voters: confirm when, where and how to vote with the local election office and the official sample ballot, not a chatbot or a viral clip.