June 4, 2026AI and the Future of Political Campaigns
In February 2024, in the world's third-largest democracy, a 72-year-old former special forces general won the presidency partly on the strength of a doe-eyed cartoon of himself cuddling his cat.
That is not a joke. Prabowo Subianto had run for president of Indonesia twice before as a fiery nationalist and lost twice. For his third try, his team used generative AI to turn him into "gemoy" — Indonesian slang for cute and cuddly — a chubby-cheeked animated version of the candidate who threw Korean finger-hearts and danced on billboards, stickers, and TikTok posts that pulled in something like 19 billion views.1 About half of Indonesia's 205 million voters are under 40, and a lot of them decided the cuddly grandpa was their guy.2 He won in a landslide.
Now here's the part you can't skip if you want to talk about this honestly: the same general had spent decades dogged by allegations of human-rights abuses from Indonesia's dictatorship era, and the adorable cartoon did a remarkable job of papering over all of it.3 So in one campaign you get the whole argument about AI in politics, start to finish. It is astonishingly powerful. It can reach people nothing else reaches. And it can just as easily be used to make voters feel something that has nothing to do with the truth.
I am going to make a pro-AI case here, because the technology is already the most consequential thing to hit campaigning since the internet, and pretending otherwise is a terrific way to lose. But I am not going to sell you the clean version, because there is not one. Two things are true at once: AI is a genuine gift to the practice of democracy, and it carries real costs — to truth, to the people on the wrong side of the digital divide, and yes, to the actual physical planet the candidate is usually promising to protect. A campaign that uses this tool well must own all of it.
The genie is already on the ballot. This is not a forecast. Across the 2024 "year of elections," AI quietly moved from novelty to plumbing, and it did so on every side of every aisle.
In the United States, both parties built it into the machinery. On the Democratic side, a startup called Quiller — founded by strategist Mike Nellis — drafts fundraising emails and texts, and campaigns using it have reported cutting their drafting time roughly in half.4 The group Tech for Campaigns says it used AI in the 2024 cycle to cut the time spent writing fundraising solicitations by about a third, and put more than a third of its budget toward AI experiments.5 On the Republican side, Push Digital Group went, in its own words, "all in," using AI to spin up hundreds of ad variations automatically and to assist with targeting and data analysis, while a GOP startup called Numinar marketed voter modeling to roughly 300 political clients.46 The Republican National Committee put out an entirely AI-generated ad back in April 2023, billing it as a first.7
Abroad, it became bigger. In India's 2024 general election — nearly a billion voters, over a hundred languages — Prime Minister Modi's party used AI to translate and stream his speeches into multiple local languages on the fly, and reportedly made around 1,000 holographic appearances at roughly 900 rallies.8 In Japan, an independent long-shot named Takahiro Anno ran an AI avatar that fielded some 8,600 voter questions and finished fifth out of 56 candidates in the Tokyo governor's race — a genuinely startling result for someone with no machine behind him.9
The lesson is the same one every campaign eventually learns about every new tool: if it works and your opponent has it, you do not get to abstain from principle. The question was never whether AI belongs in campaigns. It is already there. The only question is whether you will use it with skill and a conscience, or clumsily and without one.
What it does well is more boring than the headlines. Quiller's founder has a line I love: AI is revolutionizing campaigns, but in the most boring way possible.ten He's right, and the boring part is the important part. The deepfakes get the headlines. The quiet, unglamorous uses are what actually change who can run and win.
Start with money because campaigns run on it. Writing fundraising appeals is a soul-grinding, repetitive job, and it is exactly the kind of task these tools are good at. One Democratic group reported that AI-drafted emails brought in three to four times more fundraising dollars per staff work-hour than human-written ones — not because the AI is a genius, but because it lets a small team test more, faster, and stop burning out their best people on copy.11 That's the real story: not robots replacing strategists, but a tired field organizer getting two hours of their night back.
Then there is the thing I find most genuinely hopeful: language. Nearly sixty-six million Americans speak a language other than English at home.twelve For decades, reaching them well meant money and rare translation skills that most campaigns — especially local ones — simply did not have. Now an advocacy group can get a $20,000 grant and build a chatbot that drafts campaign messages in Hindi, Tagalog, Chinese, Hmong, Korean, and Vietnamese, languages where good translators are genuinely hard to find.12 New York's mayoral race used live AI translation; San Francisco civic groups have tested real-time translation of city council meetings.13 Done carefully — and that word is load-bearing — this means voters who were functionally ignored for years can finally hear from a campaign in their own language, and even ask it questions and get answers.
And it lowers the floor. AI ad and copy tools, voter-modeling startups, donor-research automation — the stuff that used to require a six-figure consulting contract is increasingly available to a school-board candidate or a shoestring nonprofit. That is a real shift, and mostly a good one.
The democratizing case (better know as the part I will defend) is key. For my entire adult life, the brutal truth of American politics has been that money and infrastructure decide who even gets to compete. The candidate with the big donor network hires the consultants, buys the data, runs the testing, and reaches the voters. The first-time candidate with a day job and a promising idea gets outspent into silence before anyone hears them. That is not a level field. It never was.
AI does not fix that. But it tilts it, a little, toward the people who never had the budget. It lets a small campaign produce what used to take a department. It lets an organizer who speaks one language reach a community that speaks five. It lets an underfunded down-ballot race — the school boards, the city councils, the state legislative seats where most of the decisions that actually touch your life get made — punch a few weight classes up. One Democratic technologist put it bluntly: the biggest risk she sees is fear dominating the conversation so completely that good people refuse a tool that could help them, and get left behind.11 I think she's right, and I think that caution is its own kind of harm.
So, when I say I am pro-AI in campaigns, that is what I mean. The leverage it gives the little guy. Two things are true. If the story stopped there, this would be a brochure. It does not stop there. The most obvious danger is deception, and 2024 gave us the textbook case. Two days before the New Hampshire primary, thousands of voters got a robocall in a cloned, AI-generated version of President Biden's voice, telling them not to vote.fourteen It was a deliberate voter-suppression effort. The consultant behind it — who, tellingly, was working adjacent to a rival Democrat, not the other party — was hit with a $6 million FCC fine and criminal charges, the telecom that carried the calls settled for a million, and the FCC moved to ban AI-generated robocalls outright.1415 The cross-party detail matters: this is not a problem that lives on one team. In the same cycle, U.S. intelligence reported that Russia used AI to doctor Kamala Harris speeches and Iran used deepfake audio targeting Donald Trump, and watchdogs counted hundreds of instances of viral AI-driven misinformation, including fake celebrity endorsements.sixteen Everybody is a target. Everybody is a potential abuser.
There is a subtler poison, too, and it has a name: the "liar's dividend." Once voters know any video or audio could be fake, the liar gets a gift — the ability to wave away anything real and inconvenient as "probably AI." The damage is not just the fakes people believe. It is the truths people now feel free to disbelieve. A democracy runs on a shared baseline of what happened, and that baseline is exactly what this technology can erode.
And the digital divide cuts both ways. The same translation tools that can finally include non-English voters can also be aimed at them — flooding the communities with the least third-party factchecking and the most untranslated rumor with targeted, persuasive, machine-made content. A tool that empowers the marginalized in the right hands can isolate them in the wrong ones. We do not get to claim only the upside.
Here is the part that I think separates a serious, conscientious campaign from a sloganeering one. Every one of those AI-drafted emails, every translated speech, every cartoon avatar, every "spin up 100 ad variants” all of it runs on physical buildings full of computers that drink electricity and water. The cloud is somebody's county. And if your campaign is out there promising to protect the climate and the communities living next to the infrastructure, you cannot pretend the tool you are using is weightless.
Let me give you the real numbers, honestly, in both directions.
The scale is genuinely large and growing fast. The International Energy Agency estimates that data centers used somewhere around 415 to 460 terawatt-hours of electricity in 2024 — roughly 1.5% of the world's power — and projects that to roughly double to about 945 TWh by 2030, which is close to the entire current electricity demand of Japan.1718 AI is the single biggest driver of that growth; its share of data-center power could climb from the recent 5–15% range to as much as 35–50% by 2030.19 These places are thirsty, too: the IEA tied global data centers to around 560 billion liters of water consumption in 2023, much of it for cooling.20 And the footprint shows up on the companies' own books — Google's data-center electricity use jumped 27% in a single year and its overall greenhouse emissions rose roughly 50% from 2019 to 2024, while Microsoft's energy use has climbed well over 100% since 2020, both citing AI and cloud growth.21
Now the honest other half because doom is as lazy as denial. Globally, data centers are still a small slice — around 1% of energy-related CO₂ emissions today, projected to stay under about 1.5% even as they grow.22 The energy efficiency of AI per task is improving at a rate the IEA calls essentially unprecedented in energy history, so each individual email or translation costs dramatically less than it did a year ago.23 Renewables already supply something like a quarter of data-center power, and AI itself is being used to make electricity grids cleaner and more efficient.24 The impact is also intensely local: in Ireland data centers eat over a fifth of national electricity, and in parts of Virginia it's around a quarter — so "small globally" can still mean "enormous in the town that hosts the server farm."25
So, what does an environmentally conscious campaign do with that? It does not swear off AI in some performative gestures (that just hands the advantage to a less scrupulous opponent). It uses the tool deliberately instead of carelessly: it does not generate ten thousand pieces of disposable slop because it can; it favors providers powered by cleaner energy; it treats computing as a real resource with a genuine cost; the same way a good campaign treats a donor's dollar. Caring about the planet and using AI are not opposites. But using AI thoughtlessly while campaigning on the environment — that is a hypocrisy voters can smell, and they should. Adapt — but adapt like you must live here.
The genie is out of the bottle. You cannot un-invent the cloned voice or the instant translation or the cartoon that nineteen billion people scrolled past. If you refuse the tool on principle, you do not stop it — you just lose to someone who did not. In campaigns, even more than in business, it really is adapt or die. But "adapt" is not a moral free pass, and this is the one place I would push back on the pure efficiency crowd. A business that adopts AI hopelessly loses money. A campaign that adopts AI badly can lose something a country does not get back easily — the basic trust that the voice on the phone is real, that the video happened, that the candidate asking for your vote in your own language is telling you the truth. That trust is the actual product of a democracy. It is the thing worth protecting even when protecting it is slower.
So use it. Use it to get a tired organizer their evening back, to reach the grandmother who has never once been spoken to in her first language, to let the underfunded innovative idea compete with the well-funded bad one. That is the promise, and it is real. But disclose when it is AI. Never use it to deceive a voter about what is real. And run the whole operation like someone who must live in the country — and on the planet — they are asking to lead. Prabowo's team proved the cartoon can win. The harder, better question, the one this whole technology is going to keep asking every campaign from now on, is whether you can win with it and still deserve to. I think you can. But only if you decide, on purpose, that the truth and the people and the place are not the parts you are willing to automate away.
Sources
Reuters, "How Generative AI is transforming Indonesia's election" — Prabowo's AI "gemoy" avatar (made with Midjourney), ~19 billion TikTok views, billboards/stickers. https://www.reuters.com/technology/how-generative-ai-is-transforming-indonesias-election-2024-02-08/ ↩
Reuters (as above) ~205 million voters, about half under 40; first-time voters citing "gemoy" as their reason. ↩
Columbia Political Review, "The Dark Side of Generative AI in Prabowo Subianto's Presidential Campaign" — the avatar rebrand against a backdrop of human-rights allegations from the Suharto era; an AI deepfake of the late dictator Suharto also circulated. http://www.cpreview.org/articles/2024/11/the-dark-side-of-generative-ai-in-prabowo-subiantos-presidential-campaign ↩
The American Prospect, "AI Is Changing How Politics Is Practiced in America" — Quiller (Democratic fundraising), Push Digital Group going "all in" (Republican), Chorus AI / BattlegroundAI, DonorAtlas, RivalMind. https://prospect.org/2025/10/10/ai-artificial-intelligence-campaigns-midterms/ ↩↩2
The American Prospect (as above) and Axios, "Exclusive: AI turbocharges campaign fundraising" — Tech for Campaigns cut fundraising-drafting time ~one-third; allocated 33%+ of budget to AI. https://www.axios.com/2024/01/30/ai-campaign-fundraising-democrats-chatgpt ↩
Axios, "AI becomes a political 'super-weapon'" — Numinar (Republican voter modeling, ~300 clients); Sterling Data; MissionWired's AdvantageAI. https://www.axios.com/2022/10/07/ai-becomes-a-political-super-weapon ↩
The Hill, "How a start-up is using AI to write fundraising emails” RNC's April 2023 AI-generated ad billed as a first; Quiller founding details. https://thehill.com/homenews/campaign/4146154-how-a-start-up-is-using-ai-to-write-fundraising-emails/ ↩
VOX ATL, "Global Politicians Who Used AI to Win their Election” Modi's ~1,000 holographic appearances at ~900 rallies, BJP AI translation of speeches. https://voxatl.org/vox-5-politicians-who-used-ai-to-win/ (See also Nikkei Asia: https://asia.nikkei.com/business/technology/artificial-intelligence/from-indonesia-to-india-to-u.s.-ai-creeps-into-elections) ↩
CB Insights / Quiller company profile, citing the Tokyo race — independent candidate Takahiro Anno's AI avatar answered ~8,600 voter questions and finished 5th of 56. https://www.cbinsights.com/company/quiller ↩
Digiday, "How political startups are helping small political campaigns scale content and ads with AI” Mike Nellis on AI "revolutionizing political campaigns, but in the most boring way." https://digiday.com/media/ai-briefing-how-political-startups-are-helping-small-political-campaigns-scale-content-and-ads-with-ai/ ↩
Fortune, "Democrats are in a fevered race with GOP to exploit AI in the 2024 campaign" — AI emails reportedly brought 3–4x more fundraising dollars per work-hour; Jessica Alter on fear dominating the conversation; Rep. Adam Schiff's campaign using Quiller. https://fortune.com/2024/05/06/ai-presidential-election-campaign-democrats-republicans-joe-biden-donald-trump/ ↩↩2
Rest of World, "Political campaigns use AI for multilingual voter outreach" — ~66 million U.S. residents speak a non-English language at home; AAPI Victory Alliance's $20,000 grant for a multilingual chatbot (Hindi, Tagalog, Chinese, Hmong, Korean, Vietnamese). https://restofworld.org/2024/aapi-victory-alliance-ai-voter-outreach/ ↩↩2
Harvard Ash Center, "The Role of AI in the 2024 Elections” live AI translation in the New York mayoral race; the AI candidate in the Tokyo governor's race. https://ash.harvard.edu/resources/the-role-of-ai-in-the-2024-elections/ ↩
FCC / NBC News — the AI-cloned "Biden" robocall before the NH primary urging people not to vote; consultant Steve Kramer fined $6 million and criminally charged; Lingo Telecom settled for $1 million. https://www.nbcnews.com/politics/2024-election/telecom-company-agrees-1-million-fine-biden-deepfake-rcna167564 ↩↩2
Perkins Coie, "FCC Fines Telecom That Transmitted AI-Generated Deepfake Robocalls" — FCC's $6M forfeiture order and the broader move to ban AI robocalls; the consultant was working adjacent to a rival campaign. https://perkinscoie.com/insights/update/fcc-fines-telecom-transmitted-ai-generated-deepfake-robocalls-impersonating ↩
VOX ATL (CNN-cited) — 500+ instances of viral election misinformation, including false celebrity endorsements; reporting on Russia altering Harris speeches and Iran targeting Trump with deepfake audio in 2024. https://voxatl.org/vox-5-politicians-who-used-ai-to-win/ ↩
Wikipedia, "Environmental impact of AI" (citing IEA 2025) data centers used ~415 TWh in 2024 (~1.5% of global electricity), projected to ~945 TWh by 2030, AI as the main driver. https://en.wikipedia.org/wiki/Environmental_impact_of_AI ↩
Carbon Brief, "Five charts that put data-center energy use into context" (IEA data) — ~945 TWh by 2030 is roughly equal to Japan's current electricity demand; sector more than doubles 2024–2030. https://www.carbonbrief.org/ai-five-charts-that-put-data-centre-energy-use-and-emissions-into-context/ ↩
Carbon Brief (as above) — AI responsible for ~5–15% of data-center power recently, potentially 35–50% by 2030. ↩
NIH/PMC, "The carbon and water footprints of data centers" (citing IEA) — ~560 billion liters of data-center water consumption in 2023; ~182 Mt CO₂ from data-center electricity in 2024. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12827721/ ↩
Brookings, "Global energy demands within the AI regulatory landscape" — Google's data-center electricity up 27% YoY and overall emissions up ~51% (2019–2024); Microsoft's energy up ~168% since 2020. https://www.brookings.edu/articles/global-energy-demands-within-the-ai-regulatory-landscape/ ↩
IEA, "Energy and AI — Energy supply for AI" — data centers rise from ~1% of global generation today to ~3% by 2030, at under 1% of global CO₂ emissions; data-center emissions remain below ~1.5% of energy-sector emissions through the period. https://www.iea.org/reports/energy-and-ai/energy-supply-for-ai ↩
IEA, "Key Questions on Energy and AI — Executive summary” per-task AI energy efficiency improving at a rate "unprecedented in energy history." https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary ↩
IEA, "Energy and AI” renewables currently supply ~27% of data-center electricity; AI applications can help balance grids and integrate renewables, potentially enabling emissions cuts larger than data centers' own footprint. https://www.iea.org/reports/energy-and-ai/executive-summary ↩
AI Multiple, "AI Energy Consumption Statistics" (IEA/Carbon Brief synthesis) — Ireland ~21% of national electricity for data centers (potentially 32% by 2026); Virginia ~26%. https://aimultiple.com/ai-energy-consumption ↩