Voters photograph the ballot, get refused, and ask again
About 16 million Americans will use a chatbot to help decide at least one contest in November. The dominant way they do it is not “who should I vote for.” It is photographing the ballot, being declined, and rephrasing the question until the model answers. Here is what is measured, what is modeled, and exactly how it happens.
Projected to Election Day, November 3, 2026
Range 9–25M
Candidates, measures, logistics
Mostly summaries nobody opted into
Estimated A Civly model. It re-weights Pew’s February 2026 age crosstabs from adults onto the midterm electorate, then applies the conversion rates measured after Scotland’s May 2026 election. Every step is shown below.
From a switched-off feature to sixteen million voters, in two cycles
In 2024 the AI layer of an American election barely existed, and where it did exist the platforms turned it off. Two years later it is a mass behavior with a monitoring industry built around it.
-
Mar 2024 · Pew2% of US adults had used ChatGPT to find information about the presidential election. Just 2% trusted its election information a great deal. Measured
-
Mar 2024 · GoogleGoogle restricted Gemini from answering election questions entirely — including “how do I register to vote” — and extended the block to AI Overviews. The 2024 cycle’s AI layer was deliberately switched off. Measured
-
Nov 2024 · PerplexityThe first real US deployment: an Election Information Hub built on the Democracy Works API and AP results. Address in, ballot and polling place out. Measured
-
Feb 2026 · PewChatbot use reaches 49% of US adults, up from 33% in 2024. 42% use them to search for information; 60% read AI search summaries. Measured
-
Mar 2026 · Quinnipiac51% have used AI to research topics they were curious about, up from 37% eleven months earlier. Only 27% say they have never used an AI tool. Measured
-
Jul 2026 · Caucus AIAn industry tracker estimates at least 16 million voters have already received election information from AI this cycle, through chatbots or AI search results — before the general election began. Estimated
Four patterns, taken from published reporting
Voter prompts below are quoted from published reporting. Model responses are described rather than reproduced.
01Photograph the ballot
Rikki Powers, in Baltimore, photographed her ballot and asked Claude for bullet points on each candidate, then checked the links it returned before she voted. She put her research time at 20 hours down to one.
02The refusal-and-reframe loop
Mia Taylor, in Los Angeles County, photographed her ballot and asked Claude outright. It declined, citing political bias. She reframed, and got what she wanted anyway.
Chris Johnson, in Atlanta — a registered Republican with a 40-year voting history who describes himself as libertarian — hit the same refusal before Georgia’s May primary, and routed around it by naming a source of truth.
Both voters got an answer on the second ask. The guardrail did not stop either of them; it changed the shape of the question, and the answer came back looking like sourced research rather than a recommendation.
03Values-matching against a long ballot
Robert Siebelink, in Corona, California, faced 61 candidates for governor plus a slate of lower-profile races. He uploaded the ballot, asked Claude to suggest candidates aligned with his values, narrowed the governor’s race to two, then asked it to reason about strategy. Ballot complete in under 30 minutes.
Jeremiah Hain, in Los Angeles, used ChatGPT for mayor and other offices — “I don’t have the time, nor did I want to do the same kind of research” — then posted a TikTok showing others how to repeat it. The method propagates socially.
04The tier nobody opts into
60% of US adults read AI-generated summaries at the top of search results. A voter who searches sample ballot [county] or a judge’s name receives a synthesized answer whether or not they wanted one. Pew found that 10% were unsure whether they had ever read an AI summary, which means self-reported AI use will always undercount the real figure.
The effect runs inverse to how much attention a race gets
AI’s share of the decision is near zero at the top of the ticket and largest at the bottom. Senate and governor races come with a party cue, saturation advertising and news coverage. A nonpartisan county judgeship comes with none of those. That is where a voter has no substitute, and where the model’s answer can be the only input. Analysis
| Contest type | Party cue | Other information available | AI’s likely role |
|---|---|---|---|
| US Senate / Governor | Strong | Saturation ads, news, debates | Negligible |
| US House | Strong | Some coverage in competitive seats | Minor |
| State legislature | Present | Thin to none | Meaningful |
| Ballot measures | None | Legalese text, partisan mailers | Substantial |
| Judicial / county / school board | Often none | Effectively none | Can be decisive |
Cornell’s experiments found that chatbot conversations moved likely Trump voters 3.9 points toward Harris on a 100-point scale — roughly four times the effect of a political ad — and that a persuasion-optimized model shifted opposition voters 25 points. If even a tenth of the down-ballot consultations land on a genuinely open choice, that is on the order of one to two million votes shaped by model output, concentrated in races decided by hundreds. Colorado’s 3rd district was decided by 546 votes in 2022. Analysis
Campaigns are already optimizing for it
A monitoring-and-optimization industry formed around this behavior in 2026, which is itself evidence that the behavior is real at scale.
| Operation | Who runs it | What it does |
|---|---|---|
| CampSight | Run for Something Action Fund | Runs AI chats in real browser sessions to mimic voters, across ChatGPT, Claude and Gemini. 300,000+ campaign-focused conversations analyzed, with a waitlist of 60+ campaigns and roughly 400 candidates. |
| Caucus AI | Two chief analytics officers from the 2024 Harris campaign | Tracks what Gemini, Grok and ChatGPT say about candidates state by state, stores the responses, and exposes which sources the models drew on. |
| Research Books | American Bridge 21st Century | Gives chatbots direct access to opposition research files on 2026 House, Senate and Governor Republicans. |
| AI Narratives | CivlyOurs | Probes all seven engines — ChatGPT, Gemini, Claude, Perplexity, Grok, Google’s AI Overviews and Google’s AI Mode — on the questions voters type, grades every answer helps / neutral / hurts for the race, and returns the sources each engine leaned on plus the pages to publish to change them. |
Tom Steyer spent over $200 million in California’s gubernatorial primary and finished third. Run for Something’s analysis found that chatbots ranked him sixth on affordability, cost-of-living and education-funding questions. Causation is unproven — but it is the argument every campaign is now hearing. Estimated
Accuracy is improving fast. Completeness is not.
UK, May 2026
Six states
Six states
A model that states only true things but omits half the candidates still decides the race for whoever asked.
Five steps, each one reproducible
| # | Step | Result |
|---|---|---|
| 1 | Take Pew’s February 2026 chatbot use by age (18–29: 66% · 30–49: 61% · 50–64: 42% · 65+: 23%) and re-weight it from US adults onto the 2026 midterm electorate, using the 2022 CPS shape in which voters 65 and over are 30.4% of the electorate. | 49% → 44.1% |
| 2 | Correct back up for education and income. Voters are about 8pp more likely to hold a BA; Quinnipiac finds a 14-point education gradient (60% against 46%) and a steep income one (72% above $200k against 42% below $50k). | +2pp |
| 3 | Age the February data forward to November. Pew’s series runs 23% → 33% → 49%; Quinnipiac’s research-use figure ran 37% → 51% in eleven months. Blended, with saturation deceleration. | +6pp → 52% |
| 4 | Apply the conversion rates measured after the fact in Scotland (Survation / Strathclyde, n=1,042): 29% of adults used AI for election information and 7% used it to help decide — a 62% query rate among AI users, and a 24% decision rate among those who queried. Adjusted upward for US ballot length and vote-by-mail. | ×66% ×38% |
| 5 | Multiply by turnout: a voting-eligible population of roughly 247M at 47–52% turnout, central case 49%. | 121M ballots |
52% of voters using AI at all × 66% = the 34% who query about the election; × 38% = the 13% who use it to decide. The third tier carries Pew’s 60% AI-summary readership onto the same electorate.
It is the only place a real electorate has been measured after the fact. A Scottish voter casts two votes for parties they already know; a US midterm voter faces 15 to 40 contests including nonpartisan judges and legalese ballot measures — decisions with no party cue and no substitute source. Roughly a third of US ballots are also filled out at home with a phone in hand. Offsetting downward: lower midterm salience, and heavy straight-ticket voting.
Falsifiable predictions. If a US pollster finally asks before November, we expect “used AI to help decide” at 8–12%, “used AI for election information” at 25–35%, and under-40 decision use at 20–28%. Expect the first figure to land below our 13%: fewer than 1 in 10 Americans think it is even a good idea for voters to use chatbots to pick candidates, and 76% trust AI output “hardly ever” or only sometimes, so admitting to it is low-status. Treat that gap as the social-desirability discount, not as a failure of the model. If it comes in above 15% or below 5%, the ballot-complexity adjustment in step 4 is what broke.
Nobody has measured this directly. The three headline figures are modeled, not observed. The load-bearing assumption is that a Scottish voter choosing between parties they already know converts to an AI question at a rate that tells you anything about an American facing 40 contests and a nonpartisan judge. We adjusted that rate upward and said by how much, but the adjustment is a judgment, not a measurement.
The two sixteen-millions are not corroboration. Caucus AI counts voters who received election information from AI during the primaries, which is closer to our 42M tier. Ours counts voters who will use AI to help decide a race in November. Same number, different quantity, arrived at independently — and it would be a mistake to read either as confirming the other.
The four patterns are five voters in one newspaper account. They show the shape of the behavior, not its size. Nothing about how Powers, Taylor, Johnson, Siebelink or Hain filled out a ballot establishes how many people do the same thing, and the projection does not rest on them.
Find out what the models say about your race
Every voter in the 42M tier is asking an engine about a candidate, and the engine answers with whatever it can find. Find out what it currently says about yours — and what to publish to change it.
Every figure is traceable
- Pew Research Center, Americans and AI 2026: Chatbots, Smart Devices and Views on Impact — n=5,119, fielded 17–23 February 2026; and the age crosstabs.
- Pew Research Center, Americans’ use of ChatGPT is ticking up, but few trust its election information, 2024 — the 2% baseline.
- Quinnipiac University, The Age of Artificial Intelligence — n=1,397 adults, fielded 19–23 March 2026 (PDF).
- Survation / University of Strathclyde ParlView, How AI Featured in the 2026 Scottish Parliament Election — n=1,042, fielded 24–30 April 2026.
- Demos, One third of information provided by chatbots during the election was incorrect, May 2026.
- States United Democracy Center, AI and Elections: How Well Do AI Platforms Answer Voter Questions? — roughly 1,000 responses across six states.
- The Philadelphia Inquirer / The New York Times, ‘Who should I vote for?’ Voters turn to AI before casting their ballots, 4 July 2026 — the source for Taylor, Johnson, Hain, Siebelink and Powers.
- NOTUS, ChatGPT and Claude Will Be a Force in Elections; and A New AI Tool Lets Candidates See What Chatbots Are Telling Voters.
- Run for Something Action Fund, CampSight launch, July 2026.
- Cornell University, AI chatbots can effectively sway voters — in either direction, December 2025.
- CNBC, Google restricts election-related queries for its Gemini chatbot, March 2024; and Democracy Works, Democracy Works and Perplexity collaborate to empower voters, 2024.
- UChicago Harris / AP-NORC, Bipartisan concern about the use of artificial intelligence in elections; and Pew Research Center, Voting patterns in the 2022 elections, for electorate composition.
Civly Politics Research · prepared August 20, 2026. No US survey has yet asked voters whether they used AI to decide their vote — not Pew, AP-NORC, Quinnipiac, Gallup, Ipsos, Harvard IOP, Verasight or States United. Every figure on this page is therefore tagged Measured, Estimated or Analysis, and the November projection is modeled from adjacent measurements rather than observed directly.
Voter prompts are quoted from published reporting, principally the Philadelphia Inquirer’s July 2026 account. Model responses are described rather than reproduced. Estimated figures carry the stated method and its ranges; anything labeled Analysis is Civly’s interpretation and is not a measurement.
Findings describe measured and modeled behavior. Nothing here establishes that any chatbot answer caused any vote. Not legal, compliance, or investment advice.