A Washington Post test prompted major chatbots with political issues like taxes and affirmative action, then tallied whether answers pushed only one side. Results highlight how different training approaches lead to varying degrees of balance across the board.
WHOA! Whoever thought OpenAI--with all the donations from their top brass to right wing, has actually a substantially left leaning model than Anthropic that the administration has been crossing swords with.. 🤔
May be this also holds for WaPo vs. NYT too..
https://www.washingtonpost.com/technology/interactive/2026/06/24/are-ai-chatbots-like-chatgpt-politically-biased-we-tested-them/?itid=hp-top-table-main_p001_f008 via @washingtonpost
Gemini delivered balanced responses far more often than competitors, suggesting its design prioritizes neutrality over decisive takes on contested topics.
With AI chatbots shaping how people learn about policy debates, the study leaves open how much these patterns affect daily interactions or long-term opinions.
Many users praised Google as the fairest model in the Washington Post test on political bias in AI chatbots, while negative users accused Google and OpenAI of lying or forcing Marxism.
No Digg Deeper questions have been answered for this story yet.
@ianbremmer LOL. If it weren’t for xAI, we would be force fed Marxism
Definitely. I’m really glad you posted this. It’s great that Google has very little “only” one side argument. I’m also really excited about what @NewsGuardRating may do to create a pipeline of high-quality journalism void of the extremes as well. I really wanted to do that @IQSS so I’m glad to see it getting done. We need more AI/newsstand feeds that offer balance.
@SkiMoDee1 google much more reliable according to this study.
are ai chatbots politically biased? what the washington post found from testing:
@ianbremmer You mentioned the high reliability of google while ignoring their well-known history of absurd inaccuracies. So…right back at ya, chief.
@SkiMoDee1 the x mechahitler bit was also problematic. that’s why you need not to cherry pick examples.
@SkiMoDee1 i said more reliable than the others in the context of this study. which is pretty self evident when you read it.
@ianbremmer Anyone remember when Elon Musk's Grok went full MechaHitler?
@ianbremmer True. I found black Vikings to be highly reliable.
https://www.telegraph.co.uk/business/2024/02/21/google-chatbot-ethnically-diverse-images-vikings-knights/
@ianbremmer Who owns Wapo again? The source of this investigation may have a political interest...
@ianbremmer Let’s get real. The political right has a higher saturation of falsehoods and conspiracy information.
@ianbremmer Well done to Google and xAI for both trying to balance.
@ianbremmer It’s hard to be nuanced and rational and be right wing. That’s why no matter how hard Twitter tried to game their ai to be right wing, it’s still more left leaning than right.
Grok is highly dichotomic and will adopt a position forcefully without looking at alternative arguments, whether it falls on the Right or Left on an issue When you try to force it to examine a counterposition, it typically repeats itself or shortens its responses to avoid complexity or adding in additional information.
The world is complex and it's rarely "this side is right, this side is wrong" without nuance, but the owner is a sociopath who can't see grey area when it comes to his own self-interest, and his bot has been tuned for dogmatism.
@ianbremmer Less written content from right vs left. LLMs train on available text, so it's representative sampling, not bias. In fact٫ Google Gemini artificially shifted parameters for "unbiased" perception, distorting history-e.g. US Founding Fathers as Black/Asian/Native groups in images.
@Not_Tom_Jones @ianbremmer You mean when someone prompted it for it?
@ianbremmer @SkiMoDee1 xAI is truth-seeking. If an answer is left or right coded, but the objective evidence lay in its favor, it doesn't care.
@ianbremmer Chatbots aren't biased. Their models are human-designed and the content they are fed is also (still largely) human-made ... so the question is, are the models showing biased or is the content they are consuming? Or ... did the WaPo prompts effect the data?
@ianbremmer "Both sides" like "believe scientists" vs "climate change is a hoax"?
@heidilegg @SkiMoDee1 very little only liberal, very little only conservative. so head and shoulders best according to this study.
