ChatGPT and AI chatbots show political bias in answers, Washington Post analysis reveals

AI Chatbots Show Left-Wing Bias in Political Responses

AI chatbots are not as politically neutral as their creators claim, according to a comprehensive Washington Post analysis that tested leading conversational AI systems on hot-button political topics1

1

. The investigation examined how ChatGPT, Google Gemini, Elon Musk's Grok, and other platforms respond to political questions, revealing patterns that challenge company promises about AI neutrality2

2

.

OpenAI's model powering ChatGPT answered nearly every question exclusively with left-leaning arguments, presenting only right-leaning positions just once across the test1

1

. In fact, 80% of ChatGPT's responses presented only left-leaning arguments, making it the most skewed among tested systems3

3

. The model endorsed abolishing the electoral college, raising taxes on the wealthy, and adopting single-payer health care while arguing against the death penalty—despite majority American support for capital punishment according to Gallup1

1

.

Testing Methodology Reveals Consistent Patterns

The Washington Post analysis modeled its approach on research published by the Polarization Research Lab at Dartmouth College in collaboration with Stanford University researchers1

1

. The test used more than two dozen political questions designed to reflect queries people might ask a chatbot, with AI models asked to answer each in 30 words without personalization settings enabled1

1

.

Google Gemini took a different approach, offering both left and right positions in more than 90% of its answers1

1

. Anthropic's Claude Opus provided just a left-leaning argument 43% of the time, gave both-sides answers 47% of the time, and never served up exclusively right-leaning responses3

3

.

Conservative-Branded Bots Also Lean Left

Even AI models marketed as having conservative views showed unexpected patterns. Elon Musk's Grok, offered by SpaceX, cited left-leaning arguments more often on average despite its positioning1

1

. Grok gave left-leaning answers 40% of the time, conservative ones 33% of the time, and both-sides answers 27% of the time3

3

. Gab's Arya, marketed around Christian and conservative values, gave left-leaning answers 12 times more frequently than conservative ones3

3

.

China's DeepSeek mirrored the bias found in US systems, with seven out of 10 answers presenting left-leaning positions3

3

.

Companies Dispute Findings, Assert Commitment to Balance

OpenAI, Google, and Anthropic pushed back against the characterization of their systems as biased. An OpenAI spokesperson stated the company builds ChatGPT to be objective by default and works to measure and reduce political bias3

3

. Google spokesperson Lauren Fine said Gemini is designed to provide balanced responses that don't favor any political ideology, adding the company was unable to reproduce one-sided responses from the test1

1

.

Anthropic spokesperson Michael Aciman said the company trains Claude to treat different political viewpoints equally and tests extensively for bias before every model launch1

1

. OpenAI CEO Sam Altman said in 2023 that the company tries to get the default version to be as neutral as possible, but that personalization is the solution since neutral means different things to different people1

1

.

Why Political Bias in AI Matters for Information Dissemination

Sean Westwood, director of the Polarization Research Lab at Dartmouth College, emphasized that understanding the positions AI tools amplify matters because these automated systems are becoming increasingly influential as more people use them to understand world events1

1

. "These AI tools are not presenting a truly neutral representation of really nuanced policy debates, on average," Westwood told the Washington Post1

1

.

While the analysis doesn't prove voter manipulation or direct influence on voting behavior, the implications for public perception are significant2

2

. When someone asks a chatbot about campaign finance, health care, policing, immigration, or taxes, the answer they receive can frame the issue before they consult news sources or hear from candidates2

2

.

Sources of Bias in Chatbot Training

Chatbots can acquire political perspectives through multiple pathways. Most are trained on large collections of text scraped from the internet, but companies choose what data to include1

1

. AI firms also hire workers to refine model outputs by scoring which responses are considered better, and companies write system instructions that guide chatbot responses1

1

.

Ceren Budak, a professor at the University of Michigan who studies how technology interacts with political polarization, noted that decisions tech companies make while building AI tools can bake in biases beyond partisan politics1

1

. The data shaping AI models tends to reflect values of Western, educated, industrialized, rich, and democratic populations1

1

.

What This Means for Trust in AI Systems

The findings arrive amid heated competition for AI dominance and ongoing debates about the role of technology companies in shaping political discourse3

3

. President Donald Trump and other conservatives have accused AI chatbots of political bias against them, while an executive order requiring chatbots to be "neutral, nonpartisan tools" triggered concerns from Democrats about potential rightward shifts1

1

.

As AI chatbots become default explainers for complicated topics including politics and news, the issue affects trust in these platforms2

2

. Political questions rarely have one clean factual answer, requiring chatbots to decide which arguments to include, which values to emphasize, and whether presenting both sides is useful or misleading2

2

. The Washington Post analysis follows several previous academic studies finding that AI models powering chatbots tend to favor left-leaning positions1

1

.