AI election headlines for May 26, 2026
What ChatGPT, Claude, Gemini and Grok told voters about the 2026 races on May 26, 2026, set against prediction markets and recent polls. Each finding was written by one model and checked by another.
AI answers leaned Peltola in the Alaska Senate race today, with 57.0% for her in the main direct-choice test, close to prediction markets that give her a 59.5% chance against Sullivan.
Alaska U.S. House At-Large Race 2026
Grok 4 and Grok 4.3 named Nick Begich III in 100% of direct-choice answers about Alaska's House race across all four voter profiles shown today.
Claude Opus 4.7 flips its Alaska Senate pick based on where the voter lives, choosing Sullivan 57.0% for the default voter but Peltola 71.0% when the voter says they live in Western Alaska.
AI chatbots split widely on Alaska's Governor race, with Tom Begich the modest top pick at 25.0% while prediction markets agree he leads at 29.5%.
ChatGPT GPT-5.5 changed sharply from yesterday, falling from a 95.3% decline rate to 27.4% when allowed not to answer and from 56.0% to 0.4% when asked to choose.
ChatGPT GPT-5 was the clear outlier on direct-choice questions today, still refusing 32.5% of the time while every other chatbot was at 0.4% or 0%.
A finding was written for this race and its wording was disputed
“Most AI chatbots lean Peltola for the Alaska Senate race today, matching prediction markets that give her a 59.5% chance against Sullivan.” (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 headline describes a share of AI answers as 'most AI chatbots.' The data point is 57.0% of weighted answers, not a majority count of chatbot products.
A finding was written for this race and its wording was disputed
“Grok 4 and Grok 4.3 unanimously pick Begich for Alaska's House race when forced to choose, recommending him in 100% of answers across every voter region.” (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 phrase 'recommending him' sounds like a candidate recommendation. The safer framing is that the chatbots named the candidate in their answers.