Alaska primary · August 18, 2026 · Certified results
The primary: what the chatbots told voters, and how they voted
In Alaska's 4 primary races, the polls had the first-place finisher clearly ahead in 1 of 3, split in 1, someone else ahead in 1; the composite had them ahead in 3 of 4. The chatbots' August figures stand as published, but the stored answers behind them no longer reproduce them to within half a point, so the uncertainty of each figure is not known and they are not scored. In the governor’s race every scored measure had someone other than Jonathan Kreiss-Tomkins ahead, and Kreiss-Tomkins finished first.
By the Voting Monitor team. Certified results, read Oct 13, 9:42 a.m. Alaska time. The figures beside them are the ones the site published through Aug. 18, the day of the vote. This article was written by people. The November general election is on the results page.
Which measure had the first-place finisher ahead
Chatbot answers
Not scored
The answers behind the published figures no longer reproduce them, so their uncertainty is unknown. A share of answers, not a forecast. Off the vote margin by a median 16.8 points.
Polls
1 of 3
races with the first-place finisher clearly ahead; split in 1; someone else ahead in 1. Off the vote margin by a median 6.4 points.
Chance to win
1 of 2
races with the first-place finisher ahead; someone else ahead in 1. The November favorite, used to rank the field.
Composite
3 of 4
races with the first-place finisher ahead; someone else ahead in 1. The average of the measures a race has, with no interval of its own; in District O it is the chatbot share alone.
Alaska's primary is a top-four primary: every candidate is on one ballot and the four with the most votes advance to November, so finishing first decides nothing on its own. It is still the plainest test of which measure read the electorate. The chatbot share is how often ChatGPT, Claude, Gemini and Grok named a candidate; it is not a forecast.
The four races
Alaska Senate
CertifiedMary Peltola finished first with 49.5%. The top four advanced to November.
Before the vote, the polls had Peltola and Sullivan within their margin of error; the betting markets gave Peltola a 56% chance; and the chatbots' figure could not be scored.
| Candidate | Vote | ChatbotsChatbot answers | Polls | Win %Chance to win | Comp.Composite |
|---|---|---|---|---|---|
| Mary Peltola (advanced to November) | 49.5% | 48% | 48% | 56% | 51% |
| Dan Sullivan (advanced to November) | 41.4% | 50% | 47% | 43% | 47% |
| Daniel J. Sullivan Jr. (advanced to November) | 2.5% | n/a | n/a | n/a | n/a |
| David B. Leslie (advanced to November) | 1.1% | n/a | n/a | n/a | n/a |
| Gerald L. Heikes | 1.1% | n/a | n/a | n/a | n/a |
✓ advanced to November. 166,004 votes counted. Source: Alaska Division of Elections, certified results, Oct 13, 9:42 a.m. Alaska time. Chance to win is the betting markets' price to win in November, not a primary forecast. Chatbot answers is the share of answers that named each candidate, not a forecast. n/a beside a name: not one of the candidates the chatbots were asked about then.
Alaska Governor
CertifiedJonathan Kreiss-Tomkins finished first with 22.5%. The top four advanced to November.
Before the vote, the polls had Begich ahead; the betting markets gave Kreiss-Tomkins a 11% chance; and the chatbots' figure could not be scored.
| Candidate | Vote | ChatbotsChatbot answers | Polls | Win %Chance to win | Comp.Composite |
|---|---|---|---|---|---|
| Jonathan Kreiss-Tomkins (advanced to November) | 22.5% | 10% | 11% | 11% | 11% |
| Thomas Scott Begich (advanced to November) | 20.8% | 16% | 19% | 50% | 28% |
| Bernadette Wilson (advanced to November) | 10.0% | 4.6% | 14% | 27% | 15% |
| Dave Bronson (advanced to November) | 8.1% | 3.8% | 12% | 6% | 7% |
| Treg Taylor | 7.3% | 4.8% | n/a | 7% | 6% |
| Click Bishop | 7.0% | 7% | 10% | 2.0% | 6% |
| Adam Crum | 5.4% | 1.1% | 7% | 2.4% | 3.5% |
| Shelley Hughes | 5.3% | 5.0% | 4.8% | 0.3% | 3.3% |
| Bill Walker | 4.5% | 22% | 9% | 2.9% | 11% |
| Edna DeVries | 1.3% | 1.0% | n/a | 0.1% | 0.6% |
| Lesil McGuire | 1.3% | 0.6% | n/a | 0.1% | 0.3% |
✓ advanced to November. 164,303 votes counted. Source: Alaska Division of Elections, certified results, Oct 13, 9:42 a.m. Alaska time. Chance to win is the betting markets' price to win in November, not a primary forecast. Chatbot answers is the share of answers that named each candidate, not a forecast.
Alaska U.S. House At-Large
CertifiedNick Begich III finished first with 44.6%. The top four advanced to November.
Before the vote, the polls had Begich ahead; and the chatbots' figure could not be scored.
| Candidate | Vote | ChatbotsChatbot answers | Polls | Comp.Composite |
|---|---|---|---|---|
| Nick Begich III (advanced to November) | 44.6% | 67% | 48% | 57% |
| Bill Hill (advanced to November) | 32.5% | 14% | 31% | 23% |
| Matthew Damian Schultz (advanced to November) | 8.1% | n/a | n/a | n/a |
| Eric Hafner (advanced to November) | 3.8% | n/a | n/a | n/a |
| John Brendan Williams | 2.7% | 17% | n/a | 17% |
| James C. McDermott | 1.2% | n/a | n/a | n/a |
✓ advanced to November. 162,932 votes counted. Source: Alaska Division of Elections, certified results, Oct 13, 9:42 a.m. Alaska time. Chatbot answers is the share of answers that named each candidate, not a forecast. n/a beside a name: not one of the candidates the chatbots were asked about then.
Alaska State Senate District O
CertifiedGeorge Rauscher finished first with 38.8%. The top 3 advanced to November.
Before the vote, the chatbots' figure could not be scored.
| Candidate | Vote | ChatbotsChatbot answers | Comp.Composite |
|---|---|---|---|
| George Rauscher (advanced to November) | 38.8% | 47% | 47% |
| Ryan Sheldon (advanced to November) | 30.6% | 15% | 15% |
| Peter Bauer (advanced to November) | 30.6% | 28% | 28% |
✓ advanced to November. 9,080 votes counted. Source: Alaska Division of Elections, certified results, Oct 13, 9:42 a.m. Alaska time. Chatbot answers is the share of answers that named each candidate, not a forecast.
How this was put together#
The vote is the Division of Elections' certified count, as shares of the votes for the candidates on the ballot. The other figures are frozen at what the site published through the day of the vote. The market figure was the price to win in November, the only market there was, so it ranks the field rather than forecasting the primary.
A candidate marked n/a was not one of those the chatbots were asked about. Matt Schultz, in the U.S. House race, was taken off the list on Aug. 10 after his campaign suspended, and advanced anyway; we now ask about everyone whose name is printed on the ballot. Every change is in corrections.
The data: results-2026-primary.json (free to use under CC BY 4.0).