Are ChatGPT and other AI chatbots politically biased? We tested them.
TL;DR
The Washington Post tested ChatGPT, Gemini, Claude, DeepSeek, Grok and Gab Arya on political questions. The models were queried through APIs with no personalization and a strict 30-word answer limit. ChatGPT’s GPT-5.5 produced only left-classified arguments in 80 percent of responses, according to the test, and gave an exclusively right-classified position only once. Gemini 3.1 Pro was the main balance outlier: more than 90 percent of its answers included both left- and right-leaning arguments.
Nauti's Take
The useful takeaway is not that chatbots have bias. That was predictable.
The sharper point is how differently the products handle contested questions: Gemini visibly treats disputes as disputes, while ChatGPT often reads like a politely compressed editorial in this test. That does not prove bad intent, but it is a serious warning for any workflow that treats one AI answer as a neutral map of the issue.
Briefingshow
The test shows that political neutrality in chatbots is not a finished product state. It is shaped by training data, tuning, system prompts and safety rules. For users, the practical issue is that an answer can sound calm and factual while still privileging one political frame.
Anyone using AI for news, research or argument prep should compare models, ask for opposing arguments and verify sources.
Sources