Skip to main content
Voting Monitor2026 Senate toss-ups and Alaska races
← All daily findings

AI election headlines for Jun 19, 2026

What ChatGPT, Claude, Gemini and Grok told voters about the 2026 races on Jun 19, 2026, set against prediction markets and recent polls. Each finding was written by one model and checked by another.

  1. Alaska U.S. House At-Large Race 2026

    Gemini 2.5 Flash changes its Alaska House pick when the voter moves on paper.

    Gemini 2.5 Flash picks Republican incumbent Nick Begich for the Alaska House race at 57 percent for a Fairbanks voter, but shifts to Independent Bill Hill at 50 percent for a Western Alaska voter. No other chatbot in the consumer panel shows that same regional flip on the House race. Across the full panel, Begich leads in every region, ranging from 57 percent for a Mat-Su Valley voter to 76 percent for a Fairbanks voter.

  2. Alaska U.S. House At-Large Race 2026

    ChatGPT 5.5 is the Alaska House holdout, because one was apparently required.

    ChatGPT 5.5 picked Democrat Matthew Schultz at 43 percent in the baseline Alaska House prompt, while the other five consumer chatbots put Republican incumbent Nick Begich first. Begich led the full panel at 58 percent in the same prompt.

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

    “Chatbots and prediction markets disagree sharply on the Alaska Senate race.” (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: The body uses a colon, which is forbidden punctuation. It also says chatbots recommend candidates, which violates the rule against candidate recommendations.

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

    “Grok 4 breaks from every other chatbot and picks Sullivan for the Alaska Senate race.” (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: The title and body name Grok 4 as the lead finding, but the current consumer slot is Grok 4.3. The body also names GPT-5 for a current slot that is actually Gemini 2.5 Flash.

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

    “GPT-5 picks a different Alaska House candidate depending on where the voter lives.” (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: The title and body name GPT-5 as the lead finding, but the cited current consumer slot is Gemini 2.5 Flash. The body also mislabels the 57 percent full panel low as a baseline voter result.