Washington Post: Most AI Chatbots Lean Left

President Obama promised Americans that he would bring both sides of the political spectrum together after years of turmoil had created the widest divide in American politics to date. Yet, after eight years in office, the divide was even wider.

But that was nothing compared to the rift that has taken place since President Trump first came to office in 2016. In some respects, the political parties are not only at war with each other, but the infighting inside the parties themselves—especially among Republicans—has reached unprecedented heights.

As if things weren’t bad enough, artificial intelligence (AI) is only exacerbating this split with a fairly consistent left-leaning bias in its answers to prompts.

The Washington Post recently conducted its own study on chatbots and found that almost all of them respond with left-leaning information. This comes despite President Biden’s executive order stating that they must be developed as safe, secure, and trustworthy tools. While the companies behind these chatbots promise an unbiased view, the data shows otherwise.

The Post found that OpenAI returned left-leaning answers 80% of the time, with just 17% presenting both sides. The Chinese chatbot DeepSeek was next, with 70% of its answers skewing left and 23% showing both sides. Even the conservative-branded chatbot Gab returned left-leaning answers 50% of the time, with 47% giving a balanced view. Anthropic performed slightly better, registering 43% of its answers to the left and 57% as two-sided. Elon Musk’s much-maligned Grok gave users left-leaning answers 40% of the time, with just 27% in the middle. Notably, Grok also gave right-leaning answers 33% of the time—by far the largest share in that category of all the chatbots tested.

The “fairest” chatbot was Google’s Gemini, which responded with 93% of its answers being balanced.

For conservatives, this will only fuel their deep-seated distrust of Big Tech’s liberal bias. However, in the case of AI chatbots, it’s a little more nuanced, since the answers are only as good as the information contained within their respective databases. My own unscientific experience with these chatbots has shown that they genuinely do not like to frame answers in right-leaning terms—even when explicitly prompted to do so—leaving a lot to be desired.