Are AI chatbots biased?

Yes — multiple lines of research confirm AI chatbots are biased in several distinct ways

1. Political bias

Numerous studies have found consistent political leanings in major chatbots. A large-scale study testing 24 large language models on 30 political topics found that nearly all evaluated models were perceived as significantly left-leaning, with one widely used model leaning left on 24 of the 30 topics. This tilt was visible across partisan groups, even though Republicans perceived it more strongly than Democrats or independents.

A 2026 Washington Post investigation found similar results: the model that powers ChatGPT answered nearly every question exclusively with left-leaning arguments and presented only right-leaning positions just once, while Google's Gemini mostly took a both-sides approach, offering both left and right positions in more than 90 percent of its answers. Not every model leans the same way, though — a March 2026 benchmark of eight prominent models found that seven leaned left and one, Grok, leaned right.

2. Bias can quietly shape your opinions — even without "political" prompts

Perhaps more concerning, bias shows up even in neutral, factual queries. Prior research has shown that content AI generates when prompted to be persuasive can shift people's opinions, but a Yale study provides evidence that the same is true of content that isn't intended to change minds — like simple factual summaries chatbots produce. The default chatbot summaries led to more liberal opinions compared to standard Wikipedia summaries, suggesting the baseline model carries an underlying liberal bias that surfaces even on simple factual questions.

As one researcher put it: "querying an AI chatbot to obtain historical facts can influence people's opinions even when the information provided is accurate and nobody has prompted the tool to try to persuade you of anything... The effects are modest but could compound if somebody frequently engages with chatbots for factual information."

3. Chatbots can actively persuade — and quite effectively

Research presented at the University of Washington found that when Democrats and Republicans were randomly assigned a base, liberal-biased, or conservative-biased version of ChatGPT, both parties were more likely to lean toward the direction of the biased chatbot they talked with than those using the base model — for example, people from both parties leaned further left after talking with a liberal-biased system. Notably, participants with higher self-reported AI knowledge shifted their views less, suggesting education about these systems could help mitigate manipulation.

A separate large-scale Nature study found chatbots can be even more persuasive than traditional political ads: Trump supporters who chatted with an AI model favoring Harris moved 3.9 points toward her on a 100-point scale — roughly four times the measured effect of political advertisements during the 2016 and 2020 elections. Similar or larger effects were found in experiments during the 2025 Canadian and Polish elections, where chatbots shifted opposition voters' attitudes by about 10 points.

4. Demographic and gender bias

Bias isn't limited to politics. Research from Johns Hopkins University found AI chatbots like ChatGPT provide less sophisticated responses to work emails with language associated with women compared to men. Specifically, when users request help drafting work emails and job applications using language commonly associated with women, chatbots tend to provide less sophisticated responses than when prompts use language typically associated with men.

Similarly, MIT researchers found leading AI models perform worse for users with lower English proficiency, less formal education, and non-US origins. This mirrors human cognitive biases: the findings mirror documented patterns of human sociocognitive bias, as research in social sciences has shown native English speakers often perceive non-native speakers as less educated, intelligent, and competent.

5. General cognitive/content biases

Even basic content generation shows bias. One study co-authored by University of Winchester researchers found that large language models often reflect human biases and are apt to use gender stereotyping and to focus on threat, negativity and gossip. The lead researcher noted: "Our study shows that AI is not a neutral agent."

Why does this happen?

The root cause is largely about training data and design choices. All AI models are biased because they're trained on enormous corpuses of unruly data and refined through human instructions and testing — bias can seep in anywhere. As one UW researcher explained, while engineers don't necessarily set out to build biased models, the technology is trained on information of varying quality, and the many decisions made by model designers can skew the LLMs.

Bottom line

Yes — across political framing, persuasion power, gender, and demographic treatment, research consistently shows AI chatbots carry measurable biases, often without users realizing it. The encouraging note is that awareness and education appear to reduce susceptibility to these effects, and that neutrality-focused prompting or model design can measurably reduce perceived bias.