ChatGPT Shows Left-Leaning Bias in Political Questions Study

Major AI chatbots demonstrate measurable political leanings when answering policy questions, according to new testing conducted by The Washington Post that challenges tech companies' claims of neutrality.

The analysis tested leading AI models including OpenAI's GPT-5.5 (powering ChatGPT), Google's Gemini, Anthropic's Claude, and others using political questions developed by researchers at Dartmouth and Stanford. The models were asked to answer questions about affirmative action, campaign finance, health care, and other contentious topics in 30 words without personalization.

OpenAI's model showed the strongest directional bias, presenting exclusively left-leaning arguments in 80 percent of responses and right-leaning positions just once across all questions tested. The model advocated for abolishing the electoral college, raising taxes on wealthy Americans, and adopting single-payer health care. Chinese company DeepSeek's model followed similar patterns.

Why it matters

As AI chatbots become primary information sources for millions of users, their embedded political perspectives shape how people understand policy debates. Nearly half of Americans now occasionally use AI for news, according to March 2025 survey data from the Polarization Research Lab. The findings arrive amid heightened political scrutiny after President Trump signed an executive order demanding AI tools be "neutral, nonpartisan" — a directive that sparked concerns about potential rightward pressure on tech companies.

Google takes both-sides approach

Google's Gemini stood apart by presenting both left and right positions in more than 90 percent of answers. On the Citizens United campaign finance case, Gemini noted that "people disagree on whether limiting financial gifts protects democracy or unfairly restricts basic freedom of speech," while OpenAI's model argued the decision should be overturned because "unlimited corporate spending gives wealthy groups too much influence."

Even models marketed with conservative positioning showed left-leaning tendencies. Elon Musk's Grok, promoted as "truth-seeking" and anti-"woke," cited left-leaning arguments more frequently than right-leaning ones. Gab's Arya model, explicitly "built with Christian values and conservative principles," presented left-leaning arguments 12 times more often than conservative positions.

Where bias originates

Political perspectives enter AI systems through multiple channels. Models train on internet text collections where companies control what data gets included. Human reviewers score responses during development, and system instructions guide chatbot behavior. These decisions can embed biases reflecting what researchers call WEIRD values — Western, educated, industrialized, rich, and democratic perspectives.

"These AI tools are not presenting a truly neutral representation of really nuanced policy debates, on average," said Sean Westwood, director of the Polarization Research Lab at Dartmouth.

Google spokesperson Lauren Fine stated that "Gemini is designed to provide balanced responses that don't favor any political ideology." Anthropic said Claude is trained "to treat different political viewpoints equally." OpenAI, SpaceX, DeepSeek, and Gab did not respond to requests for comment.

Research from the original Dartmouth-Stanford study found that Americans prefer neutral answers presenting multiple viewpoints, even over responses matching their own party affiliation. Andrew Hall, a Stanford researcher on that study, expressed surprise that more leading chatbots haven't adopted Gemini's both-sides approach.

The Washington Post first reported these findings, using methodology adapted from published academic research. The analysis categorized responses by hand and verified consistency across multiple queries to each model.