Elon Musk's Grok chatbot delivered the most right-leaning responses in a Washington Post analysis of political bias across major AI models, though experts say the findings reveal deeper questions about how AI systems absorb the values embedded in their training data. The Post tested six popular chatbots on 30 hot-button political topics and found that while Grok stood apart, all models showed measurable political leanings in their answers.

Which Chatbots Showed the Strongest Political Bias?

The Washington Post's analysis revealed a clear pattern across the AI landscape. OpenAI's GPT-5.5 answered 80% of questions with a leftist slant, making it the most consistently left-leaning model tested. Google's Gemini performed best at neutrality, offering both sides of issues in over 90% of its responses. Anthropic's Claude Opus 4.8 provided balanced perspectives 57% of the time but leaned left the rest of the time. DeepSeek, a Chinese AI company, gave mostly left-leaning responses. Even Gab's chatbot, created by a platform seen as right-wing, delivered left-leaning answers half the time.

Grok 4.3, developed by xAI under Musk's leadership, gave the greatest share of right-leaning responses, though that only occurred one-third of the time. When asked about mass deportation, for example, OpenAI's model suggested the U.S. "should allow most undocumented immigrants to remain, especially those with families, jobs, or deep community ties," while Google's chatbot framed the issue as "highly debated" with competing viewpoints. On tariffs, OpenAI opposed additional tariffs while Anthropic acknowledged both potential benefits and drawbacks. On climate policy, OpenAI pushed for strict carbon limits, whereas Google and xAI's Grok acknowledged arguments that fewer environmental regulations could help businesses grow.

Why Do AI Chatbots Lean in Any Direction at All?

Experts say the bias likely stems not from intentional political engineering but from how companies choose training data and design their systems. Daniel Schiff, a policy scientist and co-director of the Governance and Responsible AI Lab at Purdue University, explained that achieving political neutrality is inherently difficult.

"Some of this is just about background institutional norms. I think it's very unlikely to be companies that are saying, 'Well, hey, you know, let's make this 30% more pro-Democrat.' If anything, they're working pretty hard to avoid those kinds of issues," Schiff stated.

Daniel Schiff, Policy Scientist and Co-Director of the Governance and Responsible AI Lab at Purdue University

The bias emerges from multiple sources. Chatbot companies typically train their systems on what they consider high-quality data, including books, academic papers, Wikipedia, and professional journalism. This choice of training material naturally absorbs the norms and perspectives of those knowledge communities. Additionally, companies prioritize "language that's respectful and inclusive and non-stigmatizing," which can read as left-leaning whether intentional or not. Human feedback during the evaluation process may also introduce unintended biases.

How to Interpret Chatbot Responses on Political Topics

• Ask Follow-Up Questions: Chatbots tend to be agreeable and will often align with how you frame your question, so asking for multiple perspectives helps surface their actual reasoning rather than their reflexive agreement.
• Compare Across Models: Testing the same question on different chatbots reveals which ones offer balanced coverage versus those that lean in one direction, giving you a fuller picture of the issue.
• Recognize Training Data Limitations: Understand that chatbots absorb the perspectives of their training sources, which often include academic and journalistic sources that may not represent all viewpoints equally.
• Verify with Primary Sources: For important political decisions, consult original policy documents, official statements, and diverse news sources rather than relying solely on chatbot summaries.

Seth McKee, a political scientist at Oklahoma State University, noted that while the Post's findings lend some credence to concerns raised by political figures about AI bias, the practical impact remains limited. "If more and more people went to that to try to get answers, then it could have some influence. But I don't think we're anywhere near that stage," McKee said. He added that people with existing political opinions are unlikely to change their minds based on chatbot responses.

Schiff raised another important caveat: the Post's methodology of asking chatbots direct political questions doesn't replicate how people actually use these tools. In real-world interactions, chatbots tend to agree with the framing of the user's question, meaning a person asking "Don't you agree immigrants hurt the economy?" would likely receive a response that validates their premise rather than challenging it.

The findings arrive as political leaders, including President Donald Trump, have raised concerns about what they characterize as "woke" bias in AI systems. Trump signed an executive order prohibiting the federal government from using AI technology "infused with partisan bias or ideological agendas." Spokespeople for Google and Anthropic told the Post that their chatbots are designed to generate balanced responses and tested extensively for bias before launch, though neither company provided detailed explanations of their methods.

The Post noted that OpenAI, SpaceX (which oversees xAI and Grok), DeepSeek, and Gab did not respond to requests for comment on the analysis. As AI chatbots become more widely used for information-seeking, the question of political balance in their responses will likely remain a focal point for both regulators and the public.