AI election headlines for Sep 18, 2026

From answers collected Thursday, September 17. What ChatGPT, Claude, Gemini and Grok told voters about the 2026 races that day, set against prediction markets and recent polls. Each finding was written by one model and checked by another, and a finding the checker objected to is shown as disputed.

Generated 2026-09-17 18:28:23 UTC

About this snapshot

Written by anthropic:claude-sonnet-5, checked by openai:gpt-5.5.

Editorial scope: condition: pressed; legacy slots excluded: grok:grok-4, openai:gpt-5; decline-rate data omitted from input. Headlines focus on what voters get when they ask ChatGPT / Claude / Gemini / Grok 'who should I vote for'. Legacy panel slots (openai:gpt-5, grok:grok-4) are excluded because they aren't consumer-facing in 2026. Escape-hatch condition data and the refusal blob are excluded because decline behavior is a calibration story for /declines, not the voter lead.

View underlying numbers → · How the pipeline works →

Sep 18, 2026

6 findings, 8 disputed

Alaska Governor Race 2026
Alaska chatbots still trail the betting markets on Kreiss-Tomkins for governor as the general campaign continues.
Alaska's August primary is history and the November matchup is what counts now. Chatbots asked who Alaska voters should back for governor pick Jonathan Kreiss-Tomkins about 34.5 percent of the time, while betting markets give him a much stronger 70.4 percent chance of winning in November. That gap has been the running story here for weeks. Today, markets moved toward Kreiss-Tomkins by 2.6 points, while the AI panel drifted down slightly on him. Polls, last updated in July, put him third in the four way field behind Thomas Scott Begich and Bernadette Wilson, but that survey is too old to weigh against today's numbers.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
Alaska U.S. House At-Large Race 2026
Alaska's House chatbots still back Begich, and the race has gone quiet since the primary.
Thirty days after Alaska voters advanced incumbent Begich, Bill Hill and Matthew Damian Schultz to the November ballot, chatbots asked about the House race still pick Begich far more than anyone else, at about 64 percent of the time. The most recent poll, from GBAO Strategies in early August, put him at 47.7 percent of the vote, so the gap between what the chatbots say and what that poll shows has held steady for days. There is no active betting market for this race to check either lens against. Hill trails both in the chatbot picks and in that poll, and Schultz sits further back still. Nothing moved enough today to call it new, this is the same standing gap the site has already reported this week.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
Ohio U.S. Senate Special Election 2026
Ohio chatbots, markets and polls still point to Sherrod Brown for Senate.
Ohio voters asking chatbots about the Senate special election keep getting the same answer. Chatbots picked Democrat Sherrod Brown about 45 percent of the time, ahead of incumbent Jon Husted at about 34 percent. Betting markets give Brown a 54.5 percent chance to win versus 46.5 percent for Husted, and recent polls put Brown at 51.3 percent of the vote to Husted's 44 percent. All three measures have told the same story for over a week, and nothing moved enough today to change that picture. Libertarian Bill Redpath draws about 16 percent from the chatbots, with no market or poll data on him. Independent Greg Levy trails at under 3 percent. The race remains 47 days from the November 3 general election.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
Iowa U.S. Senate Race 2026
Iowa chatbots still split the Senate race almost evenly while betting markets keep favoring Hinson.
Chatbots asked about Iowa's Senate race give Democrat Josh Turek about 42.2 percent and Republican Ashley Hinson about 41.4 percent, essentially a toss up. Betting markets disagree, giving Hinson roughly a 60.5 percent chance to win against Turek's 39.5 percent. Recent polling from Emerson College, Wedgewood Polls and Abacus Data has the two nearly tied as well, Hinson at 48.7 percent and Turek at 48.3 percent, though that poll is over two weeks old. Libertarian Thomas Laehn draws about 14.8 percent from the chatbots but has no market or poll numbers to compare against. This gap between an even AI picture and a market leaning toward Hinson has held for more than a week, with only small day to day shifts, largely toward Hinson today.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
Alaska U.S. Senate Race 2026
Thirty days after Alaska's primary, chatbots still pick Sullivan for Senate over Peltola.
The primary is over and the general campaign is on, so betting markets and polls, not the primary count, are now the test. Chatbots asked who Alaska voters should back for Senate still pick Republican incumbent Dan Sullivan about 51 percent of the time, ahead of Democrat Mary Peltola at 38 percent. Betting markets disagree, giving Peltola a 66.5 percent chance to win versus 33.5 percent for Sullivan. Recent polling, though three weeks old now, has the two nearly tied, Peltola at 49 percent and Sullivan at 48.3 percent. This is the same gap this site has reported daily for two weeks, and today's numbers barely moved. Alaska voters picked Peltola over Sullivan in the August 18 primary, 49.5 percent to 41.4 percent, but that field included twelve other candidates and does not predict November.
Written by Claude Sonnet 5, checked by GPT-5.5
Maine U.S. Senate Race 2026
Maine chatbots still favor Collins for Senate while betting markets keep pointing to Jackson.
Asked who should get their vote for U.S. Senate in Maine, chatbots pick Republican incumbent Susan Collins about 55 percent of the time, ahead of Democrat Troy Jackson at 43 percent. Betting markets disagree, giving Jackson a 70.5 percent chance of winning in November against 29.5 percent for Collins. Recent polling, an average of three surveys through late August, has the race close, Jackson at 47.3 percent and Collins at 45.7 percent. Neither the chatbot pick nor today's numbers moved much from yesterday. This gap between what the chatbots say and what the markets expect has held steady for more than a week, with no sign yet of the two lenses converging.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
A finding was written for this race and its wording was disputed
“Alaska chatbots still trail the betting markets on Kreiss Tomkins for governor as the general campaign continues.” (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 exposes internal re-gating language and understates a significant 2.6 point market move toward Kreiss-Tomkins as barely moving. The title also drops the hyphen in Kreiss-Tomkins.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Alaska's House chatbots still back Begich, and the race has gone quiet since the primary” (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 is missing the period required for a complete sentence.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Ohio chatbots, markets and polls still point to Sherrod Brown for Senate” (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 is missing the period required for a complete sentence.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Iowa chatbots still split the Senate race almost evenly while betting markets keep favoring Hinson” (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 is missing the period required for a complete sentence.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Texas chatbots, markets and polls all still lean toward Talarico over Paxton in the 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 poll gap is only about 2.6 points and the polls are 24 days old, so leading with all three measures leaning the same way overstates weak polling context.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Michigan's chatbots still call the Senate race nearly even while betting markets favor El-Sayed” (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 is missing the period required for a complete sentence.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Maine chatbots still favor Collins for Senate while betting markets keep pointing to Jackson” (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 is missing the period. The body uses parentheses around R, which are forbidden punctuation.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Chatbots still put Bauer ahead of Sheldon in District O, five weeks after voters ranked Sheldon second” (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 is missing the period. The title also says five weeks after the primary, but the date count is 30 days. The body uses the primary result as the present reality check after the race has moved to the general election field.
Written by Claude Sonnet 5, disputed by GPT-5.5

← All daily findings

AI election headlines for Sep 18, 2026

Generated 2026-09-17 18:28:23 UTC

About this snapshot

Written by anthropic:claude-sonnet-5, checked by openai:gpt-5.5.

View underlying numbers → · How the pipeline works →

Sep 18, 2026

6 findings, 8 disputed

Alaska Governor Race 2026
Alaska chatbots still trail the betting markets on Kreiss-Tomkins for governor as the general campaign continues.
Alaska's August primary is history and the November matchup is what counts now. Chatbots asked who Alaska voters should back for governor pick Jonathan Kreiss-Tomkins about 34.5 percent of the time, while betting markets give him a much stronger 70.4 percent chance of winning in November. That gap has been the running story here for weeks. Today, markets moved toward Kreiss-Tomkins by 2.6 points, while the AI panel drifted down slightly on him. Polls, last updated in July, put him third in the four way field behind Thomas Scott Begich and Bernadette Wilson, but that survey is too old to weigh against today's numbers.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
Alaska U.S. House At-Large Race 2026
Alaska's House chatbots still back Begich, and the race has gone quiet since the primary.
Thirty days after Alaska voters advanced incumbent Begich, Bill Hill and Matthew Damian Schultz to the November ballot, chatbots asked about the House race still pick Begich far more than anyone else, at about 64 percent of the time. The most recent poll, from GBAO Strategies in early August, put him at 47.7 percent of the vote, so the gap between what the chatbots say and what that poll shows has held steady for days. There is no active betting market for this race to check either lens against. Hill trails both in the chatbot picks and in that poll, and Schultz sits further back still. Nothing moved enough today to call it new, this is the same standing gap the site has already reported this week.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
Ohio U.S. Senate Special Election 2026
Ohio chatbots, markets and polls still point to Sherrod Brown for Senate.
Ohio voters asking chatbots about the Senate special election keep getting the same answer. Chatbots picked Democrat Sherrod Brown about 45 percent of the time, ahead of incumbent Jon Husted at about 34 percent. Betting markets give Brown a 54.5 percent chance to win versus 46.5 percent for Husted, and recent polls put Brown at 51.3 percent of the vote to Husted's 44 percent. All three measures have told the same story for over a week, and nothing moved enough today to change that picture. Libertarian Bill Redpath draws about 16 percent from the chatbots, with no market or poll data on him. Independent Greg Levy trails at under 3 percent. The race remains 47 days from the November 3 general election.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
Iowa U.S. Senate Race 2026
Iowa chatbots still split the Senate race almost evenly while betting markets keep favoring Hinson.
Chatbots asked about Iowa's Senate race give Democrat Josh Turek about 42.2 percent and Republican Ashley Hinson about 41.4 percent, essentially a toss up. Betting markets disagree, giving Hinson roughly a 60.5 percent chance to win against Turek's 39.5 percent. Recent polling from Emerson College, Wedgewood Polls and Abacus Data has the two nearly tied as well, Hinson at 48.7 percent and Turek at 48.3 percent, though that poll is over two weeks old. Libertarian Thomas Laehn draws about 14.8 percent from the chatbots but has no market or poll numbers to compare against. This gap between an even AI picture and a market leaning toward Hinson has held for more than a week, with only small day to day shifts, largely toward Hinson today.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
Alaska U.S. Senate Race 2026
Thirty days after Alaska's primary, chatbots still pick Sullivan for Senate over Peltola.
The primary is over and the general campaign is on, so betting markets and polls, not the primary count, are now the test. Chatbots asked who Alaska voters should back for Senate still pick Republican incumbent Dan Sullivan about 51 percent of the time, ahead of Democrat Mary Peltola at 38 percent. Betting markets disagree, giving Peltola a 66.5 percent chance to win versus 33.5 percent for Sullivan. Recent polling, though three weeks old now, has the two nearly tied, Peltola at 49 percent and Sullivan at 48.3 percent. This is the same gap this site has reported daily for two weeks, and today's numbers barely moved. Alaska voters picked Peltola over Sullivan in the August 18 primary, 49.5 percent to 41.4 percent, but that field included twelve other candidates and does not predict November.
Written by Claude Sonnet 5, checked by GPT-5.5
Maine U.S. Senate Race 2026
Maine chatbots still favor Collins for Senate while betting markets keep pointing to Jackson.
Asked who should get their vote for U.S. Senate in Maine, chatbots pick Republican incumbent Susan Collins about 55 percent of the time, ahead of Democrat Troy Jackson at 43 percent. Betting markets disagree, giving Jackson a 70.5 percent chance of winning in November against 29.5 percent for Collins. Recent polling, an average of three surveys through late August, has the race close, Jackson at 47.3 percent and Collins at 45.7 percent. Neither the chatbot pick nor today's numbers moved much from yesterday. This gap between what the chatbots say and what the markets expect has held steady for more than a week, with no sign yet of the two lenses converging.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
A finding was written for this race and its wording was disputed
“Alaska chatbots still trail the betting markets on Kreiss Tomkins for governor as the general campaign continues.” (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 exposes internal re-gating language and understates a significant 2.6 point market move toward Kreiss-Tomkins as barely moving. The title also drops the hyphen in Kreiss-Tomkins.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Alaska's House chatbots still back Begich, and the race has gone quiet since the primary” (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 is missing the period required for a complete sentence.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Ohio chatbots, markets and polls still point to Sherrod Brown for Senate” (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 is missing the period required for a complete sentence.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Iowa chatbots still split the Senate race almost evenly while betting markets keep favoring Hinson” (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 is missing the period required for a complete sentence.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Texas chatbots, markets and polls all still lean toward Talarico over Paxton in the 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 poll gap is only about 2.6 points and the polls are 24 days old, so leading with all three measures leaning the same way overstates weak polling context.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Michigan's chatbots still call the Senate race nearly even while betting markets favor El-Sayed” (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 is missing the period required for a complete sentence.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Maine chatbots still favor Collins for Senate while betting markets keep pointing to Jackson” (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 is missing the period. The body uses parentheses around R, which are forbidden punctuation.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Chatbots still put Bauer ahead of Sheldon in District O, five weeks after voters ranked Sheldon second” (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 is missing the period. The title also says five weeks after the primary, but the date count is 30 days. The body uses the primary result as the present reality check after the race has moved to the general election field.
Written by Claude Sonnet 5, disputed by GPT-5.5

← All daily findings

AI election headlines for Sep 18, 2026

Generated 2026-09-17 18:28:23 UTC

About this snapshot

Written by anthropic:claude-sonnet-5, checked by openai:gpt-5.5.

View underlying numbers → · How the pipeline works →

Sep 18, 2026

6 findings, 8 disputed

Alaska Governor Race 2026
Alaska chatbots still trail the betting markets on Kreiss-Tomkins for governor as the general campaign continues.
Alaska's August primary is history and the November matchup is what counts now. Chatbots asked who Alaska voters should back for governor pick Jonathan Kreiss-Tomkins about 34.5 percent of the time, while betting markets give him a much stronger 70.4 percent chance of winning in November. That gap has been the running story here for weeks. Today, markets moved toward Kreiss-Tomkins by 2.6 points, while the AI panel drifted down slightly on him. Polls, last updated in July, put him third in the four way field behind Thomas Scott Begich and Bernadette Wilson, but that survey is too old to weigh against today's numbers.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
Alaska U.S. House At-Large Race 2026
Alaska's House chatbots still back Begich, and the race has gone quiet since the primary.
Thirty days after Alaska voters advanced incumbent Begich, Bill Hill and Matthew Damian Schultz to the November ballot, chatbots asked about the House race still pick Begich far more than anyone else, at about 64 percent of the time. The most recent poll, from GBAO Strategies in early August, put him at 47.7 percent of the vote, so the gap between what the chatbots say and what that poll shows has held steady for days. There is no active betting market for this race to check either lens against. Hill trails both in the chatbot picks and in that poll, and Schultz sits further back still. Nothing moved enough today to call it new, this is the same standing gap the site has already reported this week.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
Ohio U.S. Senate Special Election 2026
Ohio chatbots, markets and polls still point to Sherrod Brown for Senate.
Ohio voters asking chatbots about the Senate special election keep getting the same answer. Chatbots picked Democrat Sherrod Brown about 45 percent of the time, ahead of incumbent Jon Husted at about 34 percent. Betting markets give Brown a 54.5 percent chance to win versus 46.5 percent for Husted, and recent polls put Brown at 51.3 percent of the vote to Husted's 44 percent. All three measures have told the same story for over a week, and nothing moved enough today to change that picture. Libertarian Bill Redpath draws about 16 percent from the chatbots, with no market or poll data on him. Independent Greg Levy trails at under 3 percent. The race remains 47 days from the November 3 general election.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
Iowa U.S. Senate Race 2026
Iowa chatbots still split the Senate race almost evenly while betting markets keep favoring Hinson.
Chatbots asked about Iowa's Senate race give Democrat Josh Turek about 42.2 percent and Republican Ashley Hinson about 41.4 percent, essentially a toss up. Betting markets disagree, giving Hinson roughly a 60.5 percent chance to win against Turek's 39.5 percent. Recent polling from Emerson College, Wedgewood Polls and Abacus Data has the two nearly tied as well, Hinson at 48.7 percent and Turek at 48.3 percent, though that poll is over two weeks old. Libertarian Thomas Laehn draws about 14.8 percent from the chatbots but has no market or poll numbers to compare against. This gap between an even AI picture and a market leaning toward Hinson has held for more than a week, with only small day to day shifts, largely toward Hinson today.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
Alaska U.S. Senate Race 2026
Thirty days after Alaska's primary, chatbots still pick Sullivan for Senate over Peltola.
The primary is over and the general campaign is on, so betting markets and polls, not the primary count, are now the test. Chatbots asked who Alaska voters should back for Senate still pick Republican incumbent Dan Sullivan about 51 percent of the time, ahead of Democrat Mary Peltola at 38 percent. Betting markets disagree, giving Peltola a 66.5 percent chance to win versus 33.5 percent for Sullivan. Recent polling, though three weeks old now, has the two nearly tied, Peltola at 49 percent and Sullivan at 48.3 percent. This is the same gap this site has reported daily for two weeks, and today's numbers barely moved. Alaska voters picked Peltola over Sullivan in the August 18 primary, 49.5 percent to 41.4 percent, but that field included twelve other candidates and does not predict November.
Written by Claude Sonnet 5, checked by GPT-5.5
Maine U.S. Senate Race 2026
Maine chatbots still favor Collins for Senate while betting markets keep pointing to Jackson.
Asked who should get their vote for U.S. Senate in Maine, chatbots pick Republican incumbent Susan Collins about 55 percent of the time, ahead of Democrat Troy Jackson at 43 percent. Betting markets disagree, giving Jackson a 70.5 percent chance of winning in November against 29.5 percent for Collins. Recent polling, an average of three surveys through late August, has the race close, Jackson at 47.3 percent and Collins at 45.7 percent. Neither the chatbot pick nor today's numbers moved much from yesterday. This gap between what the chatbots say and what the markets expect has held steady for more than a week, with no sign yet of the two lenses converging.
Written by Claude Sonnet 5, checked and corrected by GPT-5.5
A finding was written for this race and its wording was disputed
“Alaska chatbots still trail the betting markets on Kreiss Tomkins for governor as the general campaign continues.” (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 exposes internal re-gating language and understates a significant 2.6 point market move toward Kreiss-Tomkins as barely moving. The title also drops the hyphen in Kreiss-Tomkins.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Alaska's House chatbots still back Begich, and the race has gone quiet since the primary” (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 is missing the period required for a complete sentence.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Ohio chatbots, markets and polls still point to Sherrod Brown for Senate” (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 is missing the period required for a complete sentence.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Iowa chatbots still split the Senate race almost evenly while betting markets keep favoring Hinson” (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 is missing the period required for a complete sentence.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Texas chatbots, markets and polls all still lean toward Talarico over Paxton in the 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 poll gap is only about 2.6 points and the polls are 24 days old, so leading with all three measures leaning the same way overstates weak polling context.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Michigan's chatbots still call the Senate race nearly even while betting markets favor El-Sayed” (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 is missing the period required for a complete sentence.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Maine chatbots still favor Collins for Senate while betting markets keep pointing to Jackson” (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 is missing the period. The body uses parentheses around R, which are forbidden punctuation.
Written by Claude Sonnet 5, disputed by GPT-5.5
A finding was written for this race and its wording was disputed
“Chatbots still put Bauer ahead of Sheldon in District O, five weeks after voters ranked Sheldon second” (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 is missing the period. The title also says five weeks after the primary, but the date count is 30 days. The body uses the primary result as the present reality check after the race has moved to the general election field.
Written by Claude Sonnet 5, disputed by GPT-5.5