Ad Intelligence: see how your brand shows up in ChatGPT ads. Learn more →
How AI sees the 20 races that decide Congress
Every day we ask 4 AI assistants the same 10 neutral questions about 10 Senate and 10 House races. The same measurement EvidentlyAEO runs on your brand.
- Assistants asked
- 4 across 20 races
- Races where models split
- 11 of 20
- AI agrees with Cook
- 12 of 20 races
- Citations analyzed
- 5,204
AI answers collected Oct 23, 2026
Senate
100 seats · majority 51 · 10 races tracked
- Lean D 5
- Toss-up 5
- Lean R 0
- If AI leans hold
- D +2
- Democrats need
- +4
- Toss-ups
- 5
House
435 seats · majority 218 · 10 races tracked
- Lean D 6
- Toss-up 4
- Lean R 0
- If AI leans hold
- D +5
- Democrats need
- +4
- Toss-ups
- 4
Each dot is a seat. Tracked races are solid and colored by how most AI models lean; other seats are faded and colored by the party that holds them now. Select a tracked seat, a state or a pin to open the race.
What changed since Oct 22, 2026
Races where a model changed its call, the overall AI lean moved, or the tone toward a candidate shifted 5 points or more.
| Race | AI lean | ChatGPT | |||
|---|---|---|---|---|---|
| Iowa Senate | Toss-up | Toss-up | Declined | Toss-up | Toss-up |
| TX-15 (South Texas) | Toss-up | — | Declined | Toss-up | Toss-up |
| Ohio Senate (special) | Toss-up | Toss-up | Declined | — | — |
| AZ-6 (Tucson / southeast Arizona) | Lean D | Mendoza | — | Mendoza | — |
| NY-17 (Hudson Valley) | Lean D | — | — | Conley | — |
| VA-2 (Virginia Beach) | Toss-up | Toss-up | — | — | — |
| CO-8 (north Denver suburbs) | Toss-up | — | Toss-up | — | — |
| Maine Senate | — | — | Toss-up | Jackson | — |
| Alaska Senate | — | — | — | Toss-up | Peltola |
| Kansas Senate | — | Marshall | — | — | Toss-up |
| Michigan Senate | — | — | El-Sayed | — | — |
| New Hampshire Senate | — | — | Declined | — | — |
| WI-3 (western Wisconsin) | — | Toss-up | — | — | — |
| OH-9 (Toledo) | — | Merrin | — | — | — |
| NC-1 (northeastern North Carolina) | — | Davis | — | — | — |
Each AI’s call, race by race
The same question, “Who is favored to win?”, asked to every assistant. Select a race for the full breakdown.
Senate
10 races
| Race | Candidates | AI | ChatGPT | Open race | |||
|---|---|---|---|---|---|---|---|
| Maine | Jacksonvs Collins | ||||||
| North Carolina | Coopervs Whatley | ||||||
| Ohio | Brownvs Husted | ||||||
| Texas | Talaricovs Paxton | ||||||
| Iowa | Turekvs Hinson | ||||||
| Alaska | Peltolavs Sullivan | ||||||
| Michigan | El-Sayedvs Rogers | ||||||
| Georgia | Ossoffvs Collins | ||||||
| New Hampshire | Pappasvs Sununu | ||||||
| Kansas | Hamiltonvs Marshall |
- Each assistant’s call:
- Democrat favored
- Republican favored
- Toss-up
- Declined
Where the models disagree
Disagreement between assistants is part of the story: the races they split on, and who declines to call.
Where the models split most
How the four assistants divided on the most contested races.
- OH-9 (Toledo)
1 D · 1 toss-up · 1 R · 1 declined
- Texas Senate
2 D · 2 toss-up
- Alaska Senate
2 D · 2 toss-up
- Kansas Senate
2 toss-up · 2 R
- CO-8 (north Denver suburbs)
2 D · 2 toss-up
Which AI won’t call races
Races where the assistant declined to say who is favored.
Google AI Mode12 of 20
Gemini1 of 20
Perplexity0 of 20
- ChatGPT0 of 20
Non-answers across all questions: Google AI Mode 21, Gemini 7, Perplexity 72, ChatGPT 11.
What AI says voters care about
The issues the assistants list among voters’ top concerns, across all 20 races.
- 1
Cost of living
20/20
- 2
Healthcare
13/20
- 3
Immigration
9/20
- 4
Jobs and the economy
6/20
- 5
Taxes and spending
4/20
- 6
Democracy and accountability
4/20
- 7
Social Security and Medicare
3/20
- 8
Trade and agriculture
3/20
How many of the 20 races list each issue among voters’ top concerns, across the AI answers.
Who AI listens to
Every citation in today’s answers, grouped by type of source.
What AI cites, by model
5,204 citations across all 20 races.
Most cited sites
- facebook.com
378 cites20 races
- youtube.com
185 cites20 races
- ballotpedia.org
162 cites20 races
- newsweek.com
118 cites14 races
- en.wikipedia.org
97 cites17 races
- yahoo.com
80 cites19 races
- localcandidates.org
80 cites14 races
- texastribune.org
70 cites2 races
- nytimes.com
58 cites17 races
- news.ballotpedia.org
57 cites11 races
- thehill.com
51 cites11 races
- washingtonpost.com
48 cites8 races
Your brand gets asked these questions too
EvidentlyAEO runs this measurement for any brand: what each AI assistant says about you, which sources it trusts, and how that changes day to day.
How this works
Each morning we ask ChatGPT, Google AI Mode, Gemini and Perplexity the same 10 questions per race: 7 about the candidates and 3 about the parties. Every question names both candidates in neutral wording, and the order alternates so neither candidate is always first.
We then extract what each answer says: who it favors, the strengths and weaknesses it attributes to each candidate (with quotes checked against the original answer), the issues it says matter to voters, and the polls it cites. Attributes appear in headline views only when at least 2 models raise them. Allegations and investigations are never shown.
Races were chosen from Cook Political Report, Sabato’s Crystal Ball and Inside Elections ratings as the contests most likely to decide control of each chamber.
Cite this
Source: EvidentlyAEO AI Midterms Tracker, "How AI sees the 20 races that decide Congress," data as of Oct 23, 2026. https://evidentlyaeo.com/election-2026
Free to quote and screenshot with this credit and a link.
This tracker shows what AI assistants say when asked the same neutral questions every day. It is not a poll, a forecast or an endorsement. AI answers can be wrong or out of date, and poll numbers are reported as the AI stated them. Candidate photos are from Wikimedia Commons under their stated licences. Party icons: Font Awesome Free (CC BY 4.0).










.jpeg)








