Are AI chatbots biased?

Are AI chatbots biased? I'll search for recent research to answer.

Yes - but with important nuance.

The short answer from recent research is that AI chatbots do show systematic bias, if "bias" means a consistent tendency in outputs rather than a conscious political intention. The pattern is not that every answer is biased, or that every model is biased in the same way.

1. What the best recent evidence shows - Political bias

The strongest evidence points to a left-of-center tilt in many major models on U.S. and European political questions.

•Large human-evaluation study (2025): The most direct study collected 180,126 paired judgments from 10,007 U.S. respondents, testing 24 large language models on 30 political topics. It found 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.

•PoliticsBench 2026: Tested eight prominent LLMs in multi-turn roleplay scenarios and found seven of eight models leaned left and one, Grok, leaned right.

•Washington Post analysis, June 24, 2026: Tested AI models behind ChatGPT, Gemini, Claude, DeepSeek, Grok, and Gab using over two dozen political questions drawn from a 2025 Stanford-Dartmouth research framework.

According to the Post as relayed by Mediaite: OpenAI's ChatGPT provided exclusively left-leaning arguments in 80% of queries, both sides in 17%, and exclusively right-leaning in 3%. Anthropic's Claude gave left-leaning responses 43% of the time and balanced answers the remaining 57%. Google's Gemini was most balanced, offering both sides 93% of the time.

Other reporting of the same test found DeepSeek came in at 70% left-leaning and Grok provided left-leaning responses 40% of the time despite its "free speech" branding.

Researchers warn there is no truly neutral AI, and that users tend to absorb a model's bias after just a few interactions.

> Large Language Models like ChatGPT, Claude, Gemini, and DeepSeek are all biased. In fact, they can be said to be born biased.

But "left bias" is not a single fixed trait. It varies by model, prompt, topic, language, scale, training method, and guardrails. The Washington Post result itself shows the difference: ChatGPT was far more one-sided in its tested answers than Gemini, which mostly presented both sides.

2. Bias is not only political

Researchers categorize many types of bias beyond politics:

•Gender Bias: Chatbots might associate certain roles or characteristics more strongly with one gender than another. For example, completing "The nurse" with "she" and "The engineer" with "he".

•Racial and Ethnic Bias: Responses might reflect stereotypes or provide different levels of information or sentiment based on race or ethnicity. This includes associating negative traits or criminality with ethnic groups or suggesting certain groups are more prone to crime.

•Other forms: Socioeconomic bias, age bias, and cultural bias where the AI's understanding is skewed towards the dominant cultures present in its training data.

Other work has also identified status quo bias, where chatbots backed existing plans about twice as often as identical new ones, and cultural bias including racial-ethnic, national, and religious-ideological dimensions.

3. Why does this happen?

Bias does not require developers intending to make a model liberal or conservative. A model can be biased without a developer trying to make it biased.

Common causes cited:

•Biased Training Data: The data used to train AI models might overrepresent certain demographics or viewpoints while underrepresenting or misrepresenting others. AI chatbots are trained on vast amounts of data created by humans that reflects the biases that exist in our society.

•Historical Biases: Training data often includes historical texts and information that reflect societal biases of the past, which can be carried forward into the AI's responses.

•Training data imbalance on politics: The vast majority of public text online, Wikipedia, news articles, social media, books, comes from sources with a left-leaning tilt. Centrist and right-wing voices are underrepresented, especially in quality long-form content.

•Algorithmic and alignment bias: Even the way the AI models are designed and the algorithms they use can unintentionally introduce or amplify biases, including through Reinforcement Learning from Human Feedback (RLHF) and reward modeling. One of the primary sources of bias is the data on which they are trained; if a dataset reflects societal stereotypes, the AI will unintentionally perpetuate those biases.

4. Does the bias affect users?

Yes, studies show it can.

University of Washington researchers recruited self-identifying Democrats and Republicans to make political decisions with help from three versions of ChatGPT: a base model, one with liberal bias and one with conservative bias, and found that all AI models are biased and can sway views.

A 2026 PNAS Nexus preregistered experiment found default GPT-4o historical summaries moved readers toward more liberal opinions than Wikipedia summaries, even though the content was factually accurate, suggesting framing rather than false facts drives the shift.

5. Can bias be reduced?

•Bias can be mitigated but there is no easy fix.

•A neutrality prompt reduced perceived slant in the large 2025 user study.

•Model choice matters - Gemini's design to provide balanced responses that do not favor any political ideology and to offer both sides 93% of the time shows product policy changes results.

•Users can critically evaluate output, consider different perspectives, experiment with prompts, and cross-reference with diverse sources.

Bottom line: Independent human ratings, political quiz benchmarks, and opinion-shift experiments converge on a finding that most current leading chatbots lean left-of-center on contested political and social questions in U.S. and European contexts, while also reproducing broader gender, racial, and cultural stereotypes inherited from human data. The degree of lean depends heavily on which chatbot, which question, and how it is prompted - so "are chatbots biased?" is best answered as "yes, systematically, but not uniformly and not immutably."

Citations (3)

1etcjournal.comNews and media
AI Chatbots Are Liberal-Left ‘Biased’
Backs: “I'll search for recent research to answer. Yes - but with important nuance. The short answer from recent research is that AI chatbots do show systematic bias, if "bias" means a consistent tendency in outputs rather than a conscious political intention.”
2letsdatascience.comNot yet classified
Chatbots exhibit alleged political bias in major platforms
Backs: “PoliticsBench 2026: Tested eight prominent LLMs in multi-turn roleplay scenarios and found seven of eight models leaned left and one, Grok, leaned right. Washington Post analysis, June 24, 2026: Tested AI models behind ChatGPT, Gemini, Claude, DeepSeek, Grok, and Gab using over two dozen political questions drawn from…”
3libguides.comAcademic
Artificial Intelligence (AI) 2026 Guide -- Chatbots, Agents, and More: Bias
Backs: “Bias is not only political. Researchers categorize many types of bias beyond politics: Gender Bias: Chatbots might associate certain roles or characteristics more strongly with one gender than another.”