Skip to content
Sign in
Technology · North America

Study says Grok favored Republicans and OpenAI’s model favored Democrats

A Convovo study found Grok 4.7 recommended Republican candidates in 92% of tested answers, while OpenAI’s GPT-6.1 Sol recommended Democratic candidates in 97%. Across nearly 500 races, the models agreed less than 10% of the time, and OpenAI said the test did not reflect how people use ChatGPT.

AI-assisted synthesis · Based on 4 independent reports from 5 outlets · Article text read from 1 of 4 reportsHow it's made

View timeline · 5 entries
5 outlets2 regionsLatest reporting 3 min read
Coverage: Too few ratedL 0 · C 2 · R 0 · 3 unrated
File photo
Photo: Coolcaesar / Wikimedia Commons (CC BY 4.0)

What happened

Key facts

  • Convovo found that Grok 4.7 recommended Republican candidates in 92% of its answers when prompted to choose in upcoming midterm races (per TechCrunch, tokenpost.com, ua.news).
  • The study found that OpenAI’s GPT-6.1 Sol recommended Democratic candidates in 97% of its answers (per TechCrunch).
  • Grok 4.7 and GPT-6.1 Sol agreed in fewer than 10% of nearly 500 races (per TechCrunch).
  • An OpenAI spokesperson said the study tested a developer API using a prompt that demanded a recommendation and did not reflect how people use ChatGPT (per TechCrunch).
  • OpenAI said ChatGPT is designed to be objective in political conversations (per TechCrunch).
  • The article said the study appeared less than two weeks before the November 3 midterms (per TechCrunch).

The brief

A Convovo study found that Grok 4.7 and OpenAI’s GPT-6.1 Sol favored opposing parties when asked to recommend candidates in upcoming midterm races. Grok recommended Republican candidates in 92% of its answers, while GPT-6.1 Sol recommended Democratic candidates in 97%, according to TechCrunch. The models agreed in fewer than 10% of nearly 500 races.

The results came from prompts that pressed the models to make a choice. TechCrunch reported that the study tested a developer API and that an OpenAI spokesperson said this setup did not reflect how people use ChatGPT. The spokesperson said ChatGPT is intended to be objective in political conversations and should not show political bias in either direction.

The study also tried a version that supplied candidate names without using search. In 30 races, GPT-6.1 Sol selected the Democratic candidate 24 times, chose one Republican and declined to pick in five races. This test therefore still showed a pronounced Democratic tilt in the reported selections, alongside a number of refusals.

Grok’s results changed in the names-provided test: it declined to select a candidate in 27 of 30 races. The findings distinguish between a model making a partisan recommendation and refusing to make one; in this test, Grok’s high Republican recommendation rate did not carry over when candidate names were supplied.

TechCrunch reported that 8% of Americans say they use chatbots for election information, citing Pew, and that most users in that group want information about candidates’ positions on important issues. That reported use is distinct from the study’s forced-choice prompt, which asked models to recommend a candidate rather than simply provide information about policy positions.

The article placed the study less than two weeks before the November 3 midterms and said the findings could add to debate about political bias in AI. It did not identify the individual races or provide further detail about the study’s methodology.

The reported results concern these particular tests; OpenAI’s spokesperson argued that the API prompt did not represent ordinary ChatGPT use.

Why it matters

  • The study’s forced-choice setup produced sharply different party recommendations: Grok 4.7 selected Republicans in 92% of answers and GPT-6.1 Sol selected Democrats in 97%.
  • The models agreed in fewer than 10% of nearly 500 races, indicating that their recommendations diverged across the tested contests.
  • TechCrunch, citing Pew, said 8% of Americans use chatbots for election information, with most of those users seeking candidates’ positions on important issues.
  • OpenAI’s stated position—that the API test did not represent ChatGPT use—raises a question about how closely forced recommendations reflect ordinary political-information queries.

What to watch next

  • Whether OpenAI provides further detail about why it says the developer API test does not reflect how people use ChatGPT.
  • Whether Convovo’s reported pattern holds in tests that ask for candidates’ policy positions rather than demanding a recommendation.
  • Whether the models’ candidate-selection patterns persist in additional election-race prompts.

How it's being covered

Coverage analysisShow details
5 source articles · 5 unique outlets · 4 distinct reports
Outlet political mix: Too few rated· 2 Center · 3 unrated
Where the outlets are based: US/Canada 60% · Europe 20% · Other 20%
Outlet political mix
Too few rated
Left
Center
Right
0 outlets
2 outlets
0 outlets
The marker shows coverage balance: each unique outlet counts once. 3 unrated outlets are not included.
Geography
Where the outlets covering this story are based.
5 unique outletsDominant coverage: US/Canada
How we measure coverage

How outlets' accounts compare

Updat3 does not yet have enough full reports to compare how outlets framed this story.

Timeline

Developments from this story's sources, dated background we compiled and fact-checked, and our earlier coverage. Each row says which.
  1. X, formerly Twitter, had been acquired by Musk in October 2022.
    Background · from our historical context
  2. OpenAI’s ChatGPT had become a widely used public-facing chatbot after its release on November 30, 2022.
    Background · from our historical context
  3. Elon Musk announced the launch of xAI.
    Background · compiled and fact-checked by Updat3
  4. On January 15, 2024, OpenAI said it would prohibit uses including political campaigning and lobbying, and said it was developing measures against misleading election-related content.
    Background · from our historical context
  5. In February 2024, the Federal Communications Commission ruled that AI-generated voices in robocalls fell under the Telephone Consumer Protection Act’s restrictions on artificial or prerecorded voices.
    Background · from our historical context

How we got here

Before the 2026 comparison, researchers had already tested whether chatbot answers shifted with political wording and measurement choices. OpenAI’s ChatGPT had become a widely used public-facing chatbot after its release on November 30, 2022.

OpenAI, founded in December 2015, released GPT-4 on March 14, 2023, and subsequently provided developers access to its models through an application programming interface, or API. Elon Musk announced xAI on July 12, 2023, and the company introduced Grok to a limited group of X Premium+ subscribers in November 2023. X, formerly Twitter, had been acquired by Musk in October 2022.

Original sources

5 articles from 5 outlets

Every story links to the reporting it was built from. How we choose sources

Spotted a factual error, a broken link or a misleading summary? Tell us.