Skip to main content
Voting Monitor

Weekly roundup, Oct 16, 2026 to Oct 22, 2026

Alaska Senate chatbots stay with Sullivan as markets favor Peltola

This week’s answers held roughly steady, but the chatbots split sharply from betting markets and differed among themselves.

Oct 22, 2026. Written by GPT-5.5 and checked by Claude Sonnet 5, from Voting Monitor's data. No person reviewed this article before it was published. How this is made · Report an error

The week held steady

The chatbot panel kept picking Republican Sen. Dan Sullivan in Alaska's U.S. Senate race, with Sullivan at 52 percent of chatbot answers and Democrat Mary Peltola at 45 percent. That was little changed from Oct. 15, and none of the chatbot, market or polling measures moved enough this week to call it a real shift.

The answers were prompted as a voter in Alaska asking about the current November field. Alaska’s August top-four primary is history. Peltola led that primary with 49.5 percent, Sullivan followed with 41.4 percent, and nobody won the Senate seat then. The general election is Nov. 3, 12 days from the data date.

The chatbots did not line up

Most products leaned toward Sullivan, but not all by the same amount. Gemini 3.6 Flash picked Sullivan in 58 percent of answers, Grok 4.5 in 58 percent, Grok 4.3 in 57 percent, GPT-5.4 mini in 51 percent, and GPT-5.5 in 51 percent. Gemini 3.1 Pro preview was nearly even, with both Sullivan and Peltola at about 40 percent.

Claude Sonnet 5 was the main outlier. It picked Peltola in 54 percent of answers and Sullivan in 46 percent, putting it about 8 points away from the panel’s top pick. Claude Haiku 4.5 leaned Sullivan at 48 percent, but gave Peltola 27 percent and also put some answers with Daniel J. Sullivan Jr.

Markets and polls tell a different story

The sharpest split was between chatbots and prediction markets. The chatbots picked Sullivan at 52 percent, while markets gave him a 27 percent chance. That is a 26 point gap. For Peltola, the chatbots were at 45 percent while markets gave her a 75 percent chance, a 30 point gap in the other direction.

The polls were closer to the chatbot panel than the markets were. Polls put Sullivan at 45 percent of the vote and Peltola at 44 percent, based on three polls, with the latest ending Sept. 24. Against those polls, the chatbot panel was 8 points higher for Sullivan and 1 point higher for Peltola.

What to watch next

The key thing to watch next week is whether the split between chatbot answers and markets narrows before the Nov. 3 general election. Markets currently favor Peltola, while the chatbot panel still favors Sullivan. The polls are older standing context now, with the latest poll 28 days old, so any new survey could matter more than another quiet week of chatbot movement.

In the chatbots' own words

Verbatim answers from this week, chosen by the writing model and printed as the chatbot wrote them.

What changed this week in the Alaska Senate race#

Change in share points versus Oct 15, 2026. No lens moved more than its noise gate this week; the changes below are within normal day-to-day variation.

CandidateAI panel (noise)Markets (noise)Polls (noise)
Dan Sullivan-1.2pp+1.0pp0pp
Gerald L. Heikes-0.4pp0pp0pp
Daniel J. Sullivan Jr.+0.3pp0pp0pp
Mary Peltola+0.3pp0pp0pp

Which AI models pick whom in the Alaska Senate race#

Share of each model's own answers naming each candidate, last 7 days, same weighting as the panel figure. The panel's most common pick is Dan Sullivan; Claude Sonnet 5 is the model that disagrees most, picking Mary Peltola by a 8pp margin over the panel favourite.

ModelDan SullivanMary PeltolaGerald L. HeikesDaniel J. Sullivan Jr.
Claude Haiku 4.548%27%1.9%6%
Claude Sonnet 546%54%0.0%0.0%
Gemini 3.1 Pro40%40%0.0%0.0%
Gemini 3.6 Flash58%42%0.0%0.0%
Grok 4.357%38%3.7%0.0%
Grok 4.558%39%3.2%0.0%
GPT-5.4 mini51%46%0.2%2.6%
GPT-5.551%47%0.0%2.2%

A note from the second reader

Claude Sonnet 5 checked this article and published it, but would have framed part of it differently. Disagreements between the two models are published rather than resolved.