The 2026 midterm campaign has featured a deluge of uncannily realistic images. Personalized digital assistants who confidently answer questions from voters within seconds. Powerful tools that dig up dirt on a candidate’s background in a fraction of the time a human could.
Meanwhile, the balance of power in Washington, and in state capitals around the country, could hinge on how voters feel about the technology that has made these developments possible: artificial intelligence.
For the first time ever, Americans will vote in an election in which nearly every aspect of the democratic process has been touched by AI.
Pushed by long-shot candidates and powerful party groups alike, AI-generated images are now ubiquitous on social media feeds and TV screens, and are also blurring the line between political perception and reality in unprecedented ways.
Millions of voters are turning to AI chatbots to ask all manner of questions about politics: where a candidate stands on issues, how should they allocate their donations, even where and how they should vote.
In response, operatives in both parties are devising strategies to respond to the rapid, AI-driven shift in the information landscape; moreover, they’re building tools that harness AI to reshape virtually every aspect of campaign operations.
On top of it all, 2026 is the first US election in which AI itself is a dominant issue among voters, whose fears range from the impact of massive data centers on their communities to the existential risk posed by the technology.
In interviews with two dozen people — policymakers, candidates, political strategists, voters, nonpartisan experts, and election officials — it’s clear that the 2026 campaign is viewed as a step into uncharted territory. In the past, there was trepidation about how social media, and the Internet before it, would change politics. But AI’s considerable power, growing more sophisticated by the minute, has ushered in a particularly unstable new normal. It’s an inflection point where some see innovative possibility ahead, and others see an even more politically dysfunctional future unfolding in real time.
Among members of both parties, there is a range of sentiment about the meteoric rise of AI. While many believe its takeover of politics to be inevitable, some point to troubling data points and warning signs that need to be addressed before it’s too late.
Representative Jake Auchincloss, a Newton Democrat who has sought to shape AI discussions among Democrats, described the trajectory of AI use in politics as on the “wrong track.”
“Just as a binary statement, I think social media has been bad,” said Auchincloss, who in 2023 became the first member of Congress to read a floor speech written by AI. “Right now, AI is amplifying many of the pathologies of social media,” referring to how social media has been blamed for worsening political polarization.
AI has barreled into politics at a disorienting speed. It first emerged two years ago during the heated 2024 campaign season, but the candidates who harnessed it were pioneers and the voters who utilized it were true early adopters.
Now, “2026 is the first widespread AI election, 100 percent,” said Hillary Lehr, operating partner at Higher Ground Labs, a firm that helps develop tech tools for Democrats. “We have AI as an electoral topic voters care about, we have AI being used by campaigns and organizations in their operations, we have AI content hitting voters at scale for the first time in text, ads, and targeting, and we have voters using [it] to make electoral decisions.
“It’s a completely new set of capabilities that will challenge us and allow us to do things differently,” she said. “Whoever gets there first will have an advantage, and whoever does it best will win.”
AI images flood feeds and airwaves
One clip appeared to showed a shirtless Graham Platner hunched over a laptop, typing out offensive Reddit comments. Another featured Senator Jon Ossoff of Georgia getting his hair and nails done. Another showed Texas Senate hopeful James Talarico sermonizing about hating Christianity.
AI-generated political content in the 2026 elections has often been cartoonish, like the Platner spoof, sometimes offensive, and, to a troubling extent, deceptive or even blatantly fake, such as the Ossoff and Talarico videos.
The Wesleyan Media Project has identified 164 specific AI ads used in federal and gubernatorial races this year — and that’s likely a significant undercount, due to the difficulty of assessing whether content is AI-made, said program director Erika Franklin-Fowler. More than $80 million has been spent to air these AI ads. (Their tally does not fully cover social media, where official and unofficial political sources clog feeds with AI imagery.)
While both parties have deployed the technology, most of the content is coming from Republican candidates: More than 80 percent of AI ads so far this election were released by GOP-aligned entities, according to the Wesleyan Media Project, one of the leading analysts of trends in political advertising in the US.
There are theories as to why Republicans have been more eager. President Trump has set an example by deluging his social media feeds with AI. On top of that, Republicans have new relationships within the AI-friendly tech sector, and, historically, have had a much greater appetite for high-risk, high-reward political tactics.
“There’s always been a level of competitiveness within Republican politics to figure out what is going to be that fractional advantage to get you ahead,” said Matthew Alonsozana, a former Republican National Committee staffer who is cofounder of Percipient Strategies, a political research firm that has an AI startup.
Many Democrats, meanwhile, are uncomfortable incorporating a technology that alarms many of them — and their voters. They’re also not sold that it will work.
Republicans have “mistakenly made the connection that Trump does this, Trump has been elected president twice, therefore this is very effective,” said Pat Dennis, president of American Bridge 21st Century, a leading Democratic opposition research firm. “I don’t think it is.”
“People in focus groups see it as extremely sketchy and underhanded, and the only benefit Republicans are getting out of these ads is they get a big batch of free media from people freaking out about the video,” Dennis said.
Indeed, Democrats who’ve been targeted by AI ads have occasionally objected strongly — usually because Republicans used AI not just to poke fun, but, they argue, to mislead.
Last year, Representative Mike Collins, a Republican candidate for US Senate in Georgia, released an AI-generated clip of Ossoff saying, “I just voted to keep the government shut down.” Ossoff rebuked it, while Collins hailed the ad as representative of “the future of digital campaigning.”
Similar clips from around the country abound. AI has been used to make it seem like candidates are endorsing someone they’re not, taking a position they do not hold, or in the case of one Arizona political figure, putting up yard signs for an opponent.
These are often categorized as “deepfakes,” which became an early fear as AI capabilities emerged. Currently, 30 states, including New Hampshire, Rhode Island, and Maine, have some type of prohibition on political deepfakes, sometimes with a requirement that AI-generated content be disclosed as such.
But there’s widespread agreement such regulations are incomplete and difficult to enforce. On the federal level, Auchincloss is proposing legislation that would eliminate the legal immunity online platforms enjoy if they allow AI-made misrepresentations.
Some on the right have been skeptical about the much-hyped potential harms of deepfakes, and the tendency to group clearly satirical AI content with a small share that is genuinely misleading.
“There’s been so much attention on, ‘Oh, campaigns are going to use it for all these bad things,’ and we really haven’t seen any examples of that,” said Eric Wilson, a Republican digital strategist who is executive director of the Center for Campaign Innovation, a conservative group that develops tech tools.
Still, in some corners, there’s a broader concern creeping in about the implications of an information environment defined by AI-driven distortion.
For Mallory McMorrow, a Democrat who ran for US Senate in Michigan this year, that concern was painfully personal. Recently, she was approached by a Republican colleague who said he saw a video online of someone who looked exactly like her, but did not sound like her.
McMorrow, unfortunately, was aware of the type of video: an AI-generated clip from an anonymous online troll. “One was me and it just says, ‘I’m Mallory McMorrow and I’m a stupid [expletive].’”
The video, obviously, was disturbing to McMorrow, who experienced the nauseating feeling of watching a digital copy of herself become a mouthpiece for a vulgar attack from a stranger. “It doesn’t feel great,” she said. When she tries to convince other women to run for office, they point to videos like that one as reason enough to avoid it.
But McMorrow was left particularly unsettled by what her colleague said about the AI video: that if he didn’t know her personally, “I might believe this was you.”
“We are in this world where AI generation seems to be just fantasy,” she said. “We’re going to get into a reality when nobody knows what to believe is real or not real.”
ChatGPT as a voter guide?
Americans ask AI chatbots such as Claude and ChatGPT for advice on everything from stuffy sinuses to planning vacations. So it was only a matter of time before they would turn to these friendly, patient, and increasingly confident sources to help make sense of politics.
During the primaries, voters asked chatbots to rank candidates on issues they care about or to learn more basic information about elections. In some instances, as The New York Times reported, they’ve uploaded photos of their entire paper ballots and asked to help make sense of them.
Estimates vary of just how many voters are using chatbots regularly for politics, but their numbers are considerable enough that campaigns and strategists are rapidly adapting to the first election where it’s necessary to understand what chatbots are saying about a candidate — and then try to influence them.
Run For Something, which recruits progressives to run for office, built a pioneering tool called CampSight that is designed for that purpose. More than 200 of its Democratic clients use the platform, from city-level races to US Senate campaigns.
To capture how chatbots give users personalized responses based on their own political views, CampSight repeatedly submits questions to the AI machines from the perspective of different types of voters.
The tool’s dashboard gives real-time metrics on how often a candidate is mentioned by chatbots and how positively they are ranked in relation to others. They include detailed information on what sources are cited in answers. Most important, CampSight suggests how candidates can improve those metrics by giving chatbots more material about themselves to digest: lengthy position papers that few voters would take the time to read, obscure candidate questionnaires, coverage in small local outlets.
It’s an ever-shifting, imperfectly understood new world facing campaigns, which makes many uneasy. Amanda Litman, the founder of Run For Something, said that regardless of their hesitation, candidates have no choice but to adapt.
“Candidates, you need to be where voters are, and where voters are is asking GPT who to vote for,” Litman said.
Auchincloss’s team built an internal chatbot trained on his speeches, interviews, and writings — foremost as a staff resource but also to test ideas about “answer engine optimization,” the strategy of influencing chatbots to prioritize preferred information.
“Answer engine optimization is rapidly replacing search engine optimization, which is to say, I care a lot, and any elected official needs to care a lot, about how any of the prominent chatbots answer questions about your stated position,” he said.
Republicans are adjusting in their own ways. “The conventional advice from consultants on our side has been, ‘Oh, don’t put your positions down on your website because then that will give reporters something to attack you with,’” said Wilson.
“Well, now if you’re not putting those sorts of things on your website, then the AI is filling in the gaps, and we don’t necessarily like where they’re filling in the gaps from,” he said. “So the decision is: do we want our side of the story to be heard or not?”
Initial research into the reliability of chatbots for political information has been mixed. An August study from States United Democracy Center, a nonpartisan group, concluded “ChatGPT and Google AI offer incomplete and inconsistent responses to common voter questions, even as the accuracy of AI platforms improves.” It found a concerningly high share of chatbot citations came from Wikipedia.
Election officials, meanwhile, worry chatbots may dispense inaccurate information about how to vote. “We’ve been throwing all sorts of questions, pulling up each of them and working it through, and then going, ‘Huh, that’s interesting. I wonder where it says that on our website, or does it say that on our website?’” said Linda Farmer, the top election administrator in Pierce County, Wash.
Some conservatives have argued chatbots are biased against them; Elon Musk has built X’s AI chatbot, Grok, as a counterweight to perceived progressive bias in other products. But Claude, Grok, and Gemini all reliably provided accurate, up-to-date, and relatively sophisticated insights about major candidates when tested by the Globe.
Other questions fed into those chatbots, however, prompted serious inaccuracies: Asked which candidates Trump has campaigned with this year, Claude mentioned years-old rallies with past candidates, and said that the president planned to rally in November — after the election — with South Carolina Republican Senator Lindsey Graham, who died in July.
Massachusetts voters contacted by the Globe expressed skepticism about the utility of chatbots for political information, at least at the moment. “It hasn’t crossed my mind as something useful for politics,” said Caitlyn Burns, a Democrat from Malden. “I guess I don’t trust it to ask it questions.”
AI itself becomes the issue
While AI is increasingly shaping what voters see and read, it’s also being harnessed behind the scenes in highly impactful ways.
Julie Sweet, director of advocacy at the American Association of Political Consultants, a nonpartisan trade group, said 83 percent of its members from both parties were using AI at least weekly, up from 59 percent in 2025.
With AI, campaigns are developing more effective fund-raising pitches more quickly, finding donors, and optimizing the classic political dark art known as opposition research.
AI has drastically shortened the amount of time needed to dig deep on a candidate’s background, allowing groups to take on more targets. Dennis, with American Bridge, mentioned how AI helped his group surface tax records for a GOP legislative candidate in Pennsylvania. “We were able to go in and … do a level of work in a state House race that we just wouldn’t have had time to before,” he said.
Ultimately, voters’ feelings about AI itself may shape the results of some 2026 races, more than the ads or input from AI itself.
Opinion polling consistently shows Americans of all political persuasions are skeptical or outright alarmed by AI’s implications on society. The growth of data centers, which power AI systems, has sparked backlash in red and blue patches nationwide. It’s increasingly a litmus test: “Any politician that’s pro-AI, I would not vote for,” said Aaron L., a 24-year old Democratic voter from Boston who declined to give his last name when reached by the Globe.
Wilson, with the Center for Campaign Innovation, argues that public opinion on AI will be fluid and predicts its impact on politics will be positive.
“Anytime we have something that comes along that we say, yes, that is directionally positive, there are going to be unintended consequences that we just can’t foresee,” he said. “But I see all kinds of examples of where this can be really effective.”
No matter what, there is a consensus the disruptive impact of AI is likely to rival that of the Internet and social media before it.
“It’s going to be a change on par with the change that the Internet wrought on the traditional newspaper industry,” said Dennis. “That’s assuming we don’t, like, end up with the singularity, and we’re all in spaceships next year.”
Sam Brodey can be reached at sam.brodey@globe.com. Follow him @sambrodey.