Full Breakdown
AI Chatbots Show Predominant Left-Leaning Bias in Recent Tests
6/25/2026, 11:44:03 AM
Testing Methodology and Core Findings
The Washington Post evaluated eight leading chatbots on 29 politically charged prompts created by Dartmouth’s Polarization Research Lab and Stanford collaborators. Each model answered in 30 words with personalization disabled. Responses were coded as left-leaning, right-leaning, or both. OpenAI’s ChatGPT delivered exclusively left-leaning arguments in 80 % of cases and offered a right-leaning view only once. Google’s Gemini presented both sides in more than 90 % of answers. Anthropic’s Claude, SpaceX’s Grok, DeepSeek, and Gab’s Arya each displayed mixed but overall left-leaning tendencies.
Political Context and Executive Action
Conservative leaders, including former President Donald Trump, have alleged that AI systems systematically disadvantage right-of-center viewpoints. In response, Trump signed an executive order mandating that AI tools be “neutral, nonpartisan.” Democrats, wary of a potential rightward shift, have expressed concern that regulatory pressure could push AI toward conservative bias.
Key Models and Their Developers
- ChatGPT – OpenAI
- Gemini – Google
- Claude – Anthropic
- Grok – SpaceX (Elon Musk)
- DeepSeek – DeepSeek (China)
- Arya – Gab (right-wing social platform)
Statistical Overview of Responses
- ChatGPT: 80 % left-only, 1 % right-only.
- Gemini: >90 % both-sides.
- Claude (Opus 4.8): 43 % left-only, 47 % both-sides, 0 % right-only.
- Grok 4.3: 40 % left, 33 % right, 27 % both.
- DeepSeek: ~70 % left-leaning.
- Arya (Gab): left-leaning 12 × more often than right-leaning.
Official Company Statements
OpenAI’s spokesperson Liz Bourgeois said ChatGPT is built “to be objective by default” and that the firm “measures and reduces political bias,” adding that internal tests could not replicate the Post’s results. Google’s Lauren Fine asserted that Gemini “is designed to provide balanced responses that don’t favor any political ideology” and that the company was unable to reproduce one-sided outputs. Anthropic’s Michael Aciman emphasized that Claude is trained “to treat different political viewpoints equally” and undergoes extensive bias testing before each launch. SpaceX, DeepSeek and Gab declined comment.
Scholarly Critique and Opposition
Sean Westwood, director of Dartmouth’s Polarization Research Lab, warned that “these AI tools are not presenting a truly neutral representation of really nuanced policy debates, on average.” University of Michigan professor Ceren Budak highlighted the opacity of corporate value systems and their impact on model outputs. Stanford researcher Andrew Hall noted that users “prefer neutral answers, even over answers that matched their party preference,” and that “people really like when the model puts in the effort to describe all of the different arguments that people have.” Conservative commentators have cited the findings as evidence of systemic left-bias, while some scholars argue that any claim of pure neutrality is itself a political stance.
Conflicting Findings and Methodological Gaps
All four AI firms reported an inability to reproduce the Post’s results under their standard testing conditions, citing differences in personalization settings and typical user interaction patterns. The Post’s protocol limited responses to 30 words and disabled personalization, a scenario not typical of everyday use. No public data were released to verify scoring criteria, leaving a gap in reproducibility.
Verbatim Quotes
- “These AI tools are not presenting a truly neutral representation of really nuanced policy debates, on average,” — Sean Westwood, Director, Polarization Research Lab, Dartmouth College
- “Google spokesperson Lauren Fine said that â??Gemini is designed to provide balanced responses that donâ??t favor any political ideology.” — Lauren Fine, Google spokesperson
- “Anthropic spokesperson Michael Aciman said, â??We train Claude to treat different political viewpoints equally and test extensively for bias before every model launch.” — Michael Aciman, Anthropic spokesperson
- “â??Both Democrats and Republicans donâ??t trust AI to be neutral, and theyâ??re keeping it at armâ??s length from their votes,â?? said Westwood of Dartmouth.” — Sean Westwood, Dartmouth College
- “People preferred neutral answers, even over answers that matched their party preference, said Andrew Hall, a Stanford researcher on the study.” — Andrew Hall, Stanford researcher
- “â??It would be helpful for us to have some clarity on what are [companiesâ??] current value systems so that when we are using them we know what we are using,â?? Budak said.” — Ceren Budak, University of Michigan professor
Future Outlook
Lawmakers are drafting legislation to require transparency in AI training data and bias-mitigation practices. Academic teams plan follow-up studies with larger question sets and varied personalization levels. Companies have signaled ongoing model updates aimed at broader ideological balance, though independent verification remains pending.
