AI in Elections: How Votes Are Checked and Counted

Keeping the voter list accurate

A voter list is never finished. People move, change names, and die, so election offices must compare records from many sources — motor vehicle departments, death records, and change-of-address files — and correct the mistakes they find. AI tools help with the matching: they compare names and addresses across databases and flag possible duplicates for staff to check. International IDEA, which studies how election bodies use AI, lists voter-roll management as one of the core jobs where the technology is being tried — and one of the riskiest, because a wrong change can stop a real voter from voting.

The risk is not abstract. International IDEA warns that an AI system used in elections "may be biased, discriminatory, or misrepresent facts about the electoral process in ways that risks disenfranchising voters." That is why the offices that use these tools keep a person in the loop: the software suggests, and a member of staff decides. A list change should leave a record of who made it and why, so it can be checked later.

The US Election Assistance Commission makes a related point: AI-generated answers "may seem plausible, but it is often inaccurate," and voting dates, hours, and locations "require a higher degree of accuracy than many AI tools can currently provide."

Counting ballots with a second set of eyes

Machines have counted paper ballots for years. AI adds software that reads images of ballots and works out how each one was marked. Albania tested this in its 2025 partial mayoral elections. Its Central Electoral Commission ran an AI image-analysis tool for vote tabulation in 3% of polling locations, with the manual count still running in parallel. At first the tool performed well, but two problems appeared as the night went on. Tired poll workers held ballots up too briefly, so the system counted some votes twice. And the tool struggled to read ballots whose marks had grown faint from handling, so valid ballots were often not counted. The lesson, recorded by International IDEA: some problems only show up when real people use the tool under pressure.

Australia takes the opposite approach. The Australian Electoral Commission keeps core voting strictly analog and is exploring AI only for office work, communications, and voter services. It has appointed chief AI officers to lead and oversee its AI work, and Commonwealth agencies must publish transparency statements saying which AI tools they use and why.

Both examples point the same way. The count stays protected by paper and by people: most votes in the United States are recorded on paper, according to the Brennan Center, and states check the count with audits and reconciliation. AI can speed up the reading of a ballot, but the paper record is what lets officials check the machine's answer.

Answering voters, and checking the answers

The most common use of AI in election offices is also the least dramatic. In the Brennan Center's survey of 834 officials, most of those who used AI did so to write simple public materials, such as social media posts, or to schedule poll worker shifts — jobs where a mistake is easy to spot and cheap to fix.

Voters are turning to chatbots too: almost half of adults in the United States use them to search for information, and 24% use them every day, according to a February survey from the Pew Research Center.

The Brennan Center tested that quality directly. Between February and August 2026, its researchers asked six widely used chatbots — ChatGPT, Gemini, Grok, Claude, Perplexity, and DeepSeek — about five false claims about elections, including that voting machines are rigged. Every chatbot disputed every false claim, even when the researchers pushed back repeatedly. But accuracy was a separate problem: half of all the responses contained an inaccuracy or a bad citation, one in three had a factual error, and one in three contained a broken link or a misleading citation. In one test, a chatbot wrongly stated that ballots are only counted when observers from political parties are present; no state requires that.

The US Election Assistance Commission has published guidance for local officials on using AI. Its advice to voters is simple: get voting details from the official source — your local election office, or a nonpartisan tool such as Vote.org. A chatbot can be a starting point, but the date, time, and place of voting should be confirmed with the office that runs the election.

Campaign ads, deepfakes and the label rules

Campaigns are the other side of AI in elections. The Wesleyan Media Project, which tracks political advertising in the United States, found at least 164 ads generated or enhanced with AI in a snapshot released on September 4, 2026, with spending of just under $80 million. Of the ads it identified, 69% carried no disclaimer saying AI had been used, and those undisclosed ads accounted for 78% of the estimated spending. Its first report, in May 2026, had found 33 such ads and about $20 million in spending.

The rules for these ads are mostly written at state level. According to the National Conference of State Legislatures, 31 states have enacted laws regulating AI in election campaigns. Most focus on deepfakes — synthetic video or audio that shows a person saying or doing something they never said or did — and either prohibit them or require a label saying AI was used. There is no federal rule on AI in political advertising: the Federal Election Commission declined in 2024 to write a new rule, and the Federal Communications Commission banned AI-generated voices in robocalls in 2024.

Public attitudes are split in an interesting way. A peer-reviewed study in the journal Political Communication, published in February 2026, found that Americans strongly disapprove of deceptive AI in election campaigns, but that disapproval does not make them dislike the candidates who use it. Instead, it increases their support for stronger rules on AI. Legal scholars Bobby Chesney and Danielle Citron coined the term "the liar's dividend" in 2019 for a related problem: once the public expects fake video, a person caught doing something wrong on genuine video can claim it was made by AI — and be believed.

How to check a claim before you share it

The most useful habit in an election season is also the oldest one: check with the source that runs the election. Before sharing a clip, a post, or a chatbot answer, confirm the date, time, and place of voting with the local election office, and treat a claim from a chatbot or a social feed as unverified until an official source confirms it.

A second check is the label. If a video or ad says it was made with AI, take that at its word; if it does not, that proves nothing on its own — the Wesleyan Media Project found most AI ads it tracked carried no disclaimer. Look for the original source of a clip, and be careful with material that appears in the last days before voting. And remember the liar's dividend: deepfakes existing is not a reason to dismiss genuine evidence, just as a missing label is not a reason to trust a fake.

The offices using AI well share the same pattern: low-risk jobs first, testing in real conditions, a person in charge of the decisions that matter, and a public record of what the tools do. Voters can apply the same test: who made this, what is the original source, and does the official record agree?

Frequently asked questions

Does AI count votes on its own?

No. In the places studied, AI works as a helper, not as the decider. Albania's 2025 pilot kept the manual count running alongside the AI tool, and Australia keeps core voting analog, using AI only for office work and voter services. In the United States, most votes are recorded on paper, and officials check the count with audits.

Can I trust a chatbot's answer about where and when to vote?

Treat it as a starting point, not as the answer. The Brennan Center found that chatbots consistently disputed false election claims, but half of their answers contained an inaccuracy or a bad citation. Confirm voting dates, hours, and locations with your local election office.

Are AI political ads labelled?

It depends on the state. Thirty-one states have laws on AI in election campaigns, most of them about deepfakes, and they either prohibit deceptive AI or require a label. There is no federal rule on AI in political ads. The Wesleyan Media Project found that 69% of the AI ads it identified in 2026 carried no disclaimer, so the absence of a label does not prove an ad is genuine.

AI in elections is best understood as a set of office tools, not as a new way of voting. It tidies lists, reads ballot images, drafts notices, and answers questions — and it makes mistakes in each of those jobs. The offices and voters who get the most from it check its work: a manual count beside the machine, an audit after the count, and an official source behind every voting detail.

If your team needs clear, checked explainers like this one for your own audience, Eri-Tech Global Ltd can help — see our AI content writing service, or get in touch through our contact page.

What to watch next

Three things to watch: whether more states add disclosure rules for AI in campaigns; whether chatbot accuracy improves — in follow-up tests in October 2026, only ChatGPT had improved its safeguards against creating fake election content; and whether election offices publish more of their AI use. International IDEA is also collecting case studies from election bodies in countries including Nigeria, Estonia, and the Philippines.

Sources

1. International IDEA — From pilot to policy: how electoral bodies are responsibly adopting AI — date checked: 2026-11-06 — https://www.idea.int/news/pilot-policy-how-electoral-bodies-are-responsibly-adopting-ai
2. SciLine — Chatbots and deepfakes in the 2026 midterms: An evidence-based AI guide for journalists — date checked: 2026-11-06 — https://www.sciline.org/elections/ai-artificial-intelligence-chatbots/
3. Brennan Center for Justice — Does AI Fight or Fuel Election Disinformation? — date checked: 2026-11-06 — https://www.brennancenter.org/our-work/research-reports/does-ai-fight-or-fuel-election-disinformation
4. National Conference of State Legislatures — Artificial Intelligence (AI) in Elections and Campaigns — date checked: 2026-11-06 — https://www.ncsl.org/elections-and-campaigns/artificial-intelligence-ai-in-elections-and-campaigns
5. UMA Technology — Nearly $80 Million Spent on AI Political Ads, WMP Finds — date checked: 2026-11-06 — https://umatechnology.org/tracking-finds-nearly-80-million-spent-on-ai-political-ads-most-without-disclosure/
6. U.S. Election Assistance Commission — Artificial Intelligence (AI) and Election Administration — date checked: 2026-11-06 — https://www.eac.gov/AI