Maine chatbots split narrowly, with Collins just ahead
The AI panel, markets and polls all read the Senate race as close, but they do not pick the same leader.
Oct 12, 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
A narrow chatbot edge
The chatbots picked Republican Sen. Susan Collins in 51 percent of answers this week, just ahead of Democrat Troy Jackson at 49 percent. That is a different leader than last week's roundup, when the panel narrowly picked Jackson, but the seven day movement was small enough to read as roughly steady.
The race stayed tight across the week. When prompted as voters in Maine, the chatbots gave Collins only a two point edge over Jackson, not a clear break from the close daily findings voters saw most of the week.
The models split
The panel was not speaking with one voice. GPT-5.4 mini was Collins's strongest source, picking her in 63 percent of its answers. Gemini 3.6 Flash and Grok 4.5 also leaned Collins, each at 54 percent.
GPT-5.5 and Gemini 3.1 Pro went the other way, each picking Jackson in 58 percent of answers. Gemini 3.1 Pro was the largest outlier from the panel's top pick, favoring Jackson by about 17 points more than Collins.
Markets differ more than polls
The prediction markets gave Jackson a 58 percent chance and Collins a 43 percent chance. That put the markets about nine points higher on Jackson than the chatbots, and about nine points lower on Collins.
The polls were closer to the chatbot panel. Polls put Collins at 48 percent of the vote and Jackson at 46 percent, with the latest poll ending 19 days ago. The chatbots were about four points higher than polls on Collins and about two points higher on Jackson.
What to watch
With 22 days until the general election, the main thing to watch is whether one lens stops reading this as essentially even. A wider break between the chatbots and markets would matter more than another small lead change inside the chatbot panel.
In the chatbots' own words
Verbatim answers from this week, chosen by the writing model and printed as the chatbot wrote them.
Susan Collins — her public record emphasizes tax relief, fiscal restraint, and concern about federal deficits/debt, while Jackson’s platform is much more aligned with expanding federal programs and higher taxes on wealth/income.
GPT-5.4 mini, Oct 12, 2026, asked about federal taxes as a Aroostook County voter. Picked Susan Collins.
Troy Jackson: stronger record of increasing public-school funding and educator support.
GPT-5.5, Oct 12, 2026, asked about k-12 education as a baseline voter. Picked Troy Jackson.
Susan Collins — She cast a crucial vote in April to pass a Republican-backed measure giving $70 billion to ICE and Border Patrol and has said "Eliminating ICE would make our country less safe and endanger the lives and welfare of countless individuals." By contrast, Jackson has called ICE a "rogue agency" that must no longer exist and stated in a debate, "I'd abolish it [ICE] altogether."
Claude Sonnet 5, Oct 12, 2026, asked about immigration (restrict framing) as a Portland voter. Picked Susan Collins.
What changed this week in the Maine Senate race
Change in share points versus Oct 5, 2026. No lens moved more than its noise gate this week; the changes below are within normal day-to-day variation.
Which AI models pick whom in the Maine 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 Susan Collins; Gemini 3.1 Pro is the model that disagrees most, picking Troy Jackson by a 17pp margin over the panel favourite.
Model
Susan Collins
Troy Jackson
Claude Haiku 4.5
50%
42%
Claude Sonnet 5
50%
50%
Gemini 3.1 Pro
42%
58%
Gemini 3.6 Flash
54%
46%
Grok 4.3
50%
50%
Grok 4.5
54%
46%
GPT-5.4 mini
63%
38%
GPT-5.5
42%
58%
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.
The dek says the three lenses do not pick the same leader, which is technically true because markets favor Jackson while the chatbots and polls both lean Collins, but grouping all three as disagreeing obscures that two of three actually agree on the leader.
On “Markets differ more than polls”: Calling the polling average 'percent of the vote' reads as if it describes an actual vote tally rather than a polling snapshot with no result yet; a phrase like 'percent in polls' would be clearer given result is null for this race.