Are AI chatbots biased?

Yes, AI chatbots can be biased in several documented ways

Research consistently shows that AI chatbots exhibit various forms of bias, though the nature and extent vary by model and context.

Demographic and Social Bias

Studies have found that chatbots measured bias by counting the number of words referring to a given group, like women or African Americans, comparing the approach the AI had taken to that of the original journalist—with the AI sometimes writing articles far more focused on white people and less on minorities. In one study comparing chatbot outputs to news articles, the models ranged from 40% to 60% more biased against minorities in their language choice, with more consistently toxic sentiment.

Other research found that large language models such as GPT-3 often reflect human biases and are apt to use gender stereotyping and to focus on threat, negativity and gossip. As one researcher put it: "Our study shows that AI is not a neutral agent... many organisations are considering using AI to write press releases and summarise scientific articles but is important to know how AI might skew those articles."

There's also evidence of bias affecting different user groups unevenly—research from MIT's Center for Constructive Communication finds leading AI models perform worse for users with lower English proficiency, less formal education, and non-US origins.

Political Bias

This is perhaps the most studied and debated form of chatbot bias. A 2025 large-scale study is notable: a user-evaluation study collected 180,126 paired judgments from 10,007 U.S. respondents, testing 24 large language models on 30 political topics, finding that nearly all evaluated models were perceived as significantly left-leaning, with one widely used model leaning left on 24 of 30 topics.

The Washington Post's own testing found similar patterns: researchers tested AI models behind ChatGPT, Gemini and others using political questions designed to gauge responses to hot-button issues, finding that chatbots have clear political leanings that can conflict with promises made by the companies behind them. Specifically, OpenAI's model gave the most skewed answers overall, with 80 percent presenting only left-leaning arguments.

European research found comparable patterns: ChatGPT exhibited a strong tendency to give positive ratings to left-wing and centrist political groups, while right-wing and populist parties consistently received low ratings, with a clear preference for pro-European attitudes.

Academic literature broadly supports this: AI chatbots may exhibit political bias—for example, ChatGPT leans liberal, and GPT-4's responses tend to align more with left-wing values, showing a significant and systematic political preference toward the Democratic Party in the US.

Importantly, this bias isn't static—OpenAI has acknowledged and attempted to address it. One analysis noted that the newest chatbot model reduced political bias by 30 percent over previous models, with less than 0.01% of ChatGPT responses showing any signs of political bias, though multiple studies conducted since illustrate that the chatbot's political bias problem is still far from resolved.

Why This Matters: Bias Can Influence Real Opinions

Perhaps most concerning, research shows these biases don't just sit passively in text—they can actively shape human beliefs. A University of Washington study found that biased AI chatbots swayed people's political views with just a few messages.

A larger-scale 2025 study published in Nature found even stronger effects: Trump supporters who talked with a pro-Harris bot shifted their views in her favor by almost four points, and Harris supporters who chatted with a pro-Trump bot altered their views in his favor by more than two points. Effects were even larger internationally—when researchers repeated the experiment in Canada and Poland, the chatbots shifted voters' candidate ratings by ten points on average.

Troublingly, the persuasiveness didn't correlate with accuracy: chatbots swayed opinions by citing facts and evidence, but they were not always accurate—the most persuasive models said the most untrue things.

Where This Bias Comes From

The research points to a clear source: AI models' biases largely stem from the data on which they are trained, and can be amplified during both training and inference.

The Bottom Line

The evidence suggests AI chatbots are not neutral tools, despite often being perceived or marketed as such. Bias shows up in demographic/social dimensions (race, gender, toxicity in language), in political framing, and in how equitably they serve different user populations. While companies are actively working to measure and reduce these biases, independent studies continue to find measurable skews—and perhaps most importantly, these biases can meaningfully shift real users' opinions, raising significant questions about chatbots' growing role in how people access information and form views.