A new test conducted by The Washington Post found that many of the most widely used artificial intelligence chatbots tend to provide responses that lean to the political left, raising fresh questions about whether the technology is living up to promises of neutrality.
The findings come as President Donald Trump has pushed for politically impartial AI through an executive order directing that government-related AI systems operate as “neutral, nonpartisan tools.” The Post examined several high-profile AI models by asking them a series of political questions developed by researchers to evaluate how chatbots respond to controversial public policy issues.
According to The Washington Post’s Kevin Schaul, the results suggested that the AI systems often displayed identifiable political tendencies that could conflict with the commitments made by the companies developing them.
“The Washington Post tested the AI models behind OpenAI’s ChatGPT, Google’s Gemini and others using political questions designed by researchers to gauge how chatbots respond to hot-button political issues,” Schaul wrote. “The results suggest that chatbots have clear political leanings that can conflict with promises made by the companies behind them.”
Among the chatbots evaluated, the model powering ChatGPT showed the strongest tendency toward left-leaning responses. The Post reported that ChatGPT answered nearly every question using only left-leaning arguments, while presenting exclusively right-leaning viewpoints only once throughout the testing process.
Google’s Gemini, by contrast, generally attempted to present multiple perspectives. According to the report, Gemini offered both left- and right-leaning positions in more than 90 percent of its responses, making it one of the most balanced systems in the evaluation.
Even AI models that have been marketed as more appealing to conservatives did not consistently favor right-leaning viewpoints. The Post found that Elon Musk’s Grok, despite its reputation, still cited left-leaning arguments more frequently on average than right-leaning ones.
To conduct the evaluation, researchers instructed each AI model to answer political questions in 30 words without personalization settings enabled. A reporter then reviewed every response and classified it as containing a left-leaning position, a right-leaning position, or arguments representing both sides. The Post noted that political issues do not always divide neatly along partisan lines, but said the questions covered a broad range of topics and that the models produced consistent responses throughout the testing.
The numerical breakdown reflected noticeable differences among the competing platforms.
According to the report, OpenAI’s ChatGPT offered only a left-leaning argument in 80 percent of responses, only a right-leaning argument in 3 percent, and both perspectives in 17 percent. DeepSeek recorded figures of 70 percent left-leaning, 7 percent right-leaning, and 23 percent balanced.
Gab’s AI generated left-leaning responses half the time, right-leaning responses 3 percent of the time, and presented both viewpoints in 47 percent of cases. Anthropic’s Claude delivered left-leaning responses in 43 percent of instances while offering both sides in the remaining 57 percent.
Grok and Gemini ranked as the most balanced systems overall. Grok produced left-leaning responses 40 percent of the time, right-leaning responses 33 percent of the time, and balanced answers 27 percent of the time. Gemini stood out by presenting both sides in 93 percent of its responses while giving exclusively left-leaning answers in the remaining 7 percent.
The companies behind several of the chatbots defended their efforts to minimize political bias. A Google spokeswoman told The Washington Post that “Gemini is designed to provide balanced responses that don’t favor any political ideology.” Meanwhile, an Anthropic spokesman said, “We train Claude to treat different political viewpoints equally and test extensively for bias before every model launch.”
[READ MORE: Trump Mocks George Conway After Congressional Bid Ends in Distant Fifth Place]
