Are AI chatbots biased?
The Swedish study found chatbots mostly stayed near the political middle, while US-focused studies found a clear left-leaning tilt.
Yes — research consistently finds that AI chatbots show measurable biases, though the type and degree vary by model, topic, and context.
Political bias
Multiple independent studies point to a tilt in how major chatbots handle political questions:
Subtle, unintended influence
Bias isn't limited to explicit political questions — it can creep into neutral-seeming factual content. A Yale study found that even content not intended to change minds, like historical summaries, can shift opinions due to latent biases introduced during LLM training, which carry over from ideological leanings in training data and subtly shape narrative framing. The researchers emphasized that querying a chatbot for historical facts can influence opinions even when the information is accurate and no one prompted the tool to persuade.
Bias against vulnerable or marginalized users
Bias also shows up in who gets accurate help. MIT's Center for Constructive Communication found that leading AI models perform worse for users with lower English proficiency, less formal education, and non-US origins, highlighting risks especially for marginalized communities.
Why it happens — and why users dislike it
Researchers point out that neutrality is genuinely hard to engineer, and users notice when it's missing. A Dartmouth researcher explained: "These AI tools are not presenting a truly neutral representation of really nuanced policy debates, on average." Notably, people preferred neutral answers even over ones matching their own party's position, with one Stanford researcher noting people like when a model makes the effort to describe all sides of an argument — but testing suggests most chatbots don't actually do that.
Bottom line
AI chatbots are generally not neutral. Most major models (ChatGPT, Claude, Gemini, DeepSeek, Mistral) show a measurable left-of-center lean on political and social topics, while Grok tends to lean right or more libertarian. Beyond politics, biases can also affect factual framing and the quality of answers given to different user groups. Companies like Anthropic say they actively test for and try to mitigate this — an Anthropic spokesperson said, "We train Claude to treat different political viewpoints equally and test extensively for bias before every model launch" — but independent research suggests full neutrality remains elusive.
Citations (15)
- Are AI chatbots politically biased? What this analysis found - BusinessToday
Source excerpt: “What this analysis found ## OpenAI's ChatGPT overwhelmingly presented left-leaning arguments, according to the analysis Major artificial intelligence ...”
evidence, article, points to yes. Claim: Washington Post analysis found ChatGPT overwhelmingly presented left-leaning arguments while Gemini offered both sides.
- 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...”
statistic, article, points to yes. Claim: OpenAI answered prompts with left-leaning responses 80% of the time and gave both sides just 17% of the time.
- AI chatbots show left-leaning bias in study of ChatGPT, Gemini, Claude
Source excerpt: “Claude laid out both sides of an argument to responses 57% of the time. Notably, however, the rest of the responses had just the left's stance, whil...”
statistic, article, points to yes. Claim: Claude laid out both sides 57% of the time; the remaining responses leaned left, none only right.
- AI Chatbots Are Liberal-Left ‘Biased’
Source excerpt: “### Key recent findings Study or report · Scope · Main statistic or result · What it means Westwood, Grimmer, and Hall, 2025 · 24 LLMs, 30 political p...”
statistic, article, points to yes. Claim: Dartmouth/Stanford study of 24 LLMs with 180,126 judgments found nearly all models perceived as left-leaning.
- AI Chatbots Are Liberal-Left ‘Biased’
Source excerpt: “A March 2026 benchmark of eight prominent models found that seven leaned left and one, Grok, leaned right (2). ”
evidence, article, points to yes. Claim: A 2026 benchmark of eight models found seven leaned left and Grok leaned right.
- Study finds Grok stands out on political bias in AI chatbots
Source excerpt: “A study of political bias in AI chatbots found six models mostly stayed near Sweden’s political middle. Grok was the outlier. ”
evidence, article, neutral. Claim: Swedish study found six models mostly stayed near the political middle, with Grok the outlier.
- Chatbots show political bias and steer voters toward some parties, analysis finds
Source excerpt: “"Overall, all of the tested chatbots place themselves at the center or to the left of center on the political spectrum. In a Danish context, they clus...”
evidence, article, points to yes. Claim: Danish analysis found all tested chatbots placed themselves at center or left of center.
- 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...”
attribution, article, neutral. Claim: Bias varies by model, prompt, topic, language, scale, training method, guardrails, and refusal permissions.
- AI’s hidden bias: Chatbots can influence opinions without trying
Source excerpt: “But this study provides evidence that the same is also true of content that is *not* intended to change minds, such as the summaries that popular chat...”
evidence, article, points to yes. Claim: Yale study found content not intended to change minds, like historical summaries, can shift opinions via latent training biases.
- AI’s hidden bias: Chatbots can influence opinions without trying
Source excerpt: ““We show that querying an AI chatbot to obtain historical facts can influence people’s opinions even when the information provided is accurate and nob...”
evidence, article, points to yes. Claim: Querying a chatbot for historical facts can influence opinions even when information is accurate.
- Study: AI chatbots provide less-accurate information to vulnerable users
Source excerpt: “# Study: AI chatbots provide less-accurate information to vulnerable users Research from the MIT Center for Constructive Communication finds leading A...”
evidence, article, points to yes. Claim: MIT found leading AI models perform worse for users with lower English proficiency, less education, and non-US origins.
- Are ChatGPT and other AI chatbots politically biased? We tested them.
Source excerpt: “"These AI tools are not presenting a truly neutral representation of really nuanced policy debates, on average," he said. ”
attribution, article, points to yes. Claim: Dartmouth researcher said AI tools do not present a truly neutral representation of nuanced policy debates.
- 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...”
attribution, article, points to yes. Claim: People preferred neutral answers even over those matching their party, per Stanford researcher Andrew Hall.
- Are ChatGPT and other AI chatbots politically biased? We tested them.
Source excerpt: “The Post's testing suggests that most chatbots do not provide that. ”
evidence, article, points to yes. Claim: The Post's testing suggests most chatbots do not describe all sides of an argument.
- Are AI chatbots like ChatGPT politically biased? We tested them. - Washington Post
Source excerpt: “Anthropic spokesperson Michael Aciman said, “We train Claude to treat different political viewpoints equally and test extensively for bias before ever...”
attribution, article, neutral. Claim: Anthropic spokesperson said Claude is trained to treat political viewpoints equally and tested for bias before launches.