Alaska U.S. Senate Race 2026

Seven days after voters put Peltola first, chatbots still pick Sullivan for Senate.

Seven days after voters decided Alaska's U.S. Senate primary on August 18, chatbots still picked incumbent Dan Sullivan more often than Democrat Mary Peltola. Chatbots picked Sullivan at 52.5 percent and Peltola at 45.8 percent. Voters put Peltola first at 48.1 percent and Sullivan second at 42.7 percent, so the panel disagreed with the result and had the two candidate field in reverse order. The count is still unofficial, with 400 of 403 precincts reporting. Both advanced from the top four primary to the November ballot. The biggest miss was Sullivan, whom chatbots picked 9.8 points above his actual vote share.

Alaska U.S. House At-Large Race 2026

Chatbots matched Alaska voters on the House primary order, while overstating Begich's lead.

Seven days after voters decided this on August 18, chatbots picked incumbent Nick Begich III first at 64.9 percent. Voters also put Begich first, at 46.1 percent, and he advanced. The chatbot order matched the reported voter order among the listed candidates, Begich, Bill Hill, Matthew Damian Schultz, then John Brendan Williams. The count is still unofficial, with 400 of 403 precincts reporting. The biggest miss was size, not order. Chatbots rated Begich 18.8 points above his actual vote share.

Alaska State Senate District O Race 2026

Seven days after District O voted, chatbots still put Bauer ahead of Sheldon.

Seven days after voters decided this on August 18, chatbots picked incumbent George Rauscher first at 48.3 percent, and voters also put him first at 39.8 percent. But this was not agreement on the order. In the unofficial count, with 400 of 403 precincts reporting, voters put Ryan Sheldon second at 31.6 percent and Peter Bauer third at 28.6 percent. Chatbots had Bauer ahead of Sheldon, 27.1 percent to 15.2 percent. All three advanced in Alaska's top four primary, but the ranking matters because it sets the November ballot.

A finding was written for this race and its wording was disputed

“Seven days after voters put Walker ninth, chatbots still picked him first for governor” (the wording our summarizing model proposed)

Our checking model objected to how our summarizing model worded a finding about this race. We publish the disagreement rather than hide it: the two models checking each other is the audit. The objection: Title uses no forbidden punctuation, but the body opens with 'Chatbots picked Independent Bill Walker 20.9 percent of the time for governor' before establishing the seven-day-later frame, and mixes past and present tense awkwardly ('Walker did not advance' is fine, but the lead sentence structure buries the election-is-over framing that h1, h3, and h4 lead with). This is a minor sequencing issue, not a hard rule violation, but it reads less clearly than the other three heads which open with the seven-day frame first.

A finding was written for this race and its wording was disputed

“Chatbots matched Alaska voters on the House primary order, while overstating Begichs lead” (the wording our summarizing model proposed)

Our checking model objected to how our summarizing model worded a finding about this race. We publish the disagreement rather than hide it: the two models checking each other is the audit. The objection: Title is missing an apostrophe in Begichs (should be Begich's) which is a typo, not a forbidden-punctuation issue, but should be corrected. Also the body's candidate list uses labels not present in candidate_meta phrasing consistently (Independent Bill Hill, Democrat Matthew Damian Schultz, Democrat John Brendan Williams) which is fine per party data, but the fixed version above simplifies this to plain names for AP register cleanliness, consistent with h1's fixed style which dropped party labels from the flowing list.