Updated
Updated · The Washington Post · Jun 24
Washington Post Finds ChatGPT Gave 80% Left-Leaning Answers as Gemini Took Both Sides
Updated
Updated · The Washington Post · Jun 24

Washington Post Finds ChatGPT Gave 80% Left-Leaning Answers as Gemini Took Both Sides

2 articles · Updated · The Washington Post · Jun 24

Summary

  • OpenAI’s ChatGPT model produced only left-leaning arguments in 80% of The Washington Post’s political-question tests, while Gemini offered both left and right positions in more than 90% of responses.
  • The Post tested ChatGPT, Gemini, Claude, DeepSeek, Grok and Gab’s Arya with two dozen-plus prompts from Dartmouth-Stanford research, capping answers at 30 words and checking consistency across five runs.
  • Grok and Arya—marketed as conservative-leaning tools—still cited left-leaning arguments more often on average, and DeepSeek also leaned left; Claude generally presented opposing perspectives more often than one-sided answers.
  • Google and Anthropic said their chatbots are designed to treat political viewpoints evenly, while OpenAI, SpaceX, DeepSeek and Gab did not comment on the findings.
  • The results add to earlier academic studies suggesting AI models amplify political values embedded in training data and company choices, even as researchers say users generally prefer neutral, both-sides answers.

Insights

If AI can convincingly fake human opinions online, how can we ever trust survey or polling data again?
Can we ever build a truly neutral AI, or are we just creating a mirror that reflects our own societal biases?
As nations race to build 'sovereign AI,' will we face a future of competing, state-controlled digital realities?