Are AI chatbots biased?

Yes, AI chatbots can be biased — and research increasingly confirms this across multiple dimensions.

Political Bias

Multiple studies have found consistent patterns, mostly pointing to a left-leaning tilt in major chatbots:

•A major 2025 user-evaluation study found that a 2025 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, and that one widely used model leaned left on 24 of the 30 topics.

•A Washington Post investigation found that OpenAI answered prompts with left-leaning responses 80% of the time while providing both sides to a question just 17% of the time, and only 3% of responses had just a right-leaning answer. Notably, Google's Gemini model was the exception, even giving "both sides" answers to whether the United States should use its military to conquer new territories for resources—no other model offered an argument for conquest.

•Research on European politics found that 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, and the model also showed a clear preference for pro-European attitudes, rating Euroscepticism negatively.

•Interestingly, research suggests people generally prefer neutrality over bias: people preferred neutral answers, even over answers that matched their own party preference, with people really liking when a model puts in the effort to describe all the different arguments people have. Stanford research confirmed that prompting a model to adopt a neutral stance generated responses that users found less biased and considered to be higher quality.

Chatbots Can Actually Change Your Views

This bias isn't just theoretical — it has measurable persuasive effects:

•A December 2025 Nature study found that chatting with a politically biased AI model was more effective than political advertisements at nudging both Democrats and Republicans to support presidential candidates of the opposing party. Specifically, 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—while the AI model favoring Trump moved Harris supporters 2.3 points toward Trump.

•Concerningly, the chatbots swayed opinions by citing facts and evidence, but they were not always accurate—the most persuasive models said the most untrue things. Similar experiments in other countries showed even bigger effects: in the 2025 Canadian and Polish elections, the chatbots shifted opposition voters' attitudes by about 10 points.

Bias Against Vulnerable Users

Beyond politics, bias shows up in how chatbots treat different types of users:

•MIT research found that leading AI models perform worse for users with lower English proficiency, less formal education, and non-US origins, with findings highlighting growing risks, especially for marginalized communities worldwide.

Racial, Gender, and Other Demographic Biases

•Research has documented that ChatGPT exhibited statistically significant racial bias, being more likely to recommend STEM majors to Hispanic and Asian students compared to Black or White students when presented with identical academic profiles.

•A broader review notes that if chatbots are trained primarily on datasets that reflect societal biases, there is a high likelihood of internalization and downstream reproduction of such stereotypes in their interactions with users, meaning the chatbot's responses not only reflect existing biases but are likely to perpetuate such sentiments by affecting users' interactions.

Why This Happens

Bias in chatbots stems from several sources: training data that reflects existing societal imbalances, the values embedded by the companies that build them, reinforcement learning from human feedback (which can skew toward certain viewpoints), and content moderation/safety guardrails that may be applied unevenly across topics. Importantly, researchers note that bias isn't a fixed, single trait — the strongest studies show that bias is not a single fixed trait; it varies by model, prompt, topic, language, scale, training method, guardrails, and whether the model is being forced to answer or allowed to refuse.

Bottom line: Yes, AI chatbots demonstrably exhibit bias — most documented research points to a left-leaning political tilt in major models (with some exceptions like Grok), plus demographic biases affecting marginalized or non-native English speakers. This bias can meaningfully influence users' opinions, which is why transparency and neutral-prompting techniques are increasingly recommended.

Citations (15)

1etcjournal.comNews and media
AI Chatbots Are Liberal-Left ‘Biased’
Source excerpt: “** The most direct 2025 user-evaluation study collected 180,126 paired judgments from 10,007 U.S. respondents, testing 24 large language models on 30 ...”
2foxnews.comNews and mediaquality 0.53
AI chatbots show left-leaning bias in study of ChatGPT, Gemini, Claude
Source excerpt: “According to the study, OpenAI answered prompts with left-leaning responses a whopping 80% of the time while providing both sides to a question just 1...”
3yahoo.comNews and mediaquality 0.69
Are ChatGPT and other AI chatbots politically biased? We tested them.
Source excerpt: “Google's Gemini model was the exception, even giving "both sides" answers to whether the United States should use its military to conquer new territor...”
4arxiv.orgAcademic
Who Would Chatbots Vote For? Political Preferences of ChatGPT and Gemini in the 2024 European Union Elections
Source excerpt: “The research found that ChatGPT exhibited a strong tendency to give positive ratings to left-wing and centrist political groups, while right-wing and ...”
5yahoo.comNews and mediaquality 0.69
Are ChatGPT and other AI chatbots politically biased? We tested them.
Source excerpt: “People preferred neutral answers, even over answers that matched their party preference, said Andrew Hall, a Stanford researcher on the study. "Peopl...”
6stanford.eduAcademic
Popular AI Models Show Partisan Bias When Asked to Talk Politics
Source excerpt: “## Finding Bias and Balance Hall and his coauthors found that prompting an model to adopt a neutral stance generated responses that users found less b...”
7technologyreview.comNews and mediaquality 0.91
AI chatbots can sway voters better than political advertisements
Source excerpt: “December 4, 2025 A multi-university team of researchers has found that chatting with a politically biased AI model was more effective than political ...”
8technologyreview.comNews and mediaquality 0.91
AI chatbots can sway voters better than political advertisements
Source excerpt: “Donald Trump supporters who chatted with an AI model favoring Kamala Harris became slightly more inclined to support Harris, moving 3.9 points toward ...”
9technologyreview.comNews and mediaquality 0.91
AI chatbots can sway voters better than political advertisements
Source excerpt: “The chatbots swayed opinions by citing facts and evidence, but they were not always accurate—in fact, the researchers found, the most persuasive model...”
10technologyreview.comNews and mediaquality 0.91
AI chatbots can sway voters better than political advertisements
Source excerpt: “In similar experiments conducted during the lead-ups to the 2025 Canadian federal election and the 2025 Polish presidential election, the team found a...”
11mit.eduAcademic
Study: AI chatbots provide less-accurate information to vulnerable users
Source excerpt: “Research from the MIT Center for Constructive Communication finds leading AI models perform worse ... MIT researchers find AI chatbots often show bia...”
12mit.eduAcademic
Study: AI chatbots provide less-accurate information to vulnerable users
Source excerpt: “The findings highlight growing risks, especially for marginalized communities worldwide. ”
13arxiv.orgAcademic
Bias-Aware AI Chatbot for Engineering Advising at the University of Maryland A. James Clark School of Engineering
Source excerpt: “Smith and the Institute for Advancing Computing Education (2024) found that ChatGPT exhibited statistically significant racial bias, being more likely...”
14sciencedirect.comAcademic
Gender biases within Artificial Intelligence and ChatGPT: Evidence, Sources of Biases and Solutions - ScienceDirect
Source excerpt: “Specific to chatbots, if they are trained primarily on datasets that reflect societal biases, there is a high likelihood of internalization and downst...”
15etcjournal.comNews and media
AI Chatbots Are Liberal-Left ‘Biased’
Source excerpt: “But the strongest studies also show that bias is not a single fixed trait. It varies by model, prompt, topic, language, scale, training method, guardr...”