Are chatbots biased? I'll survey recent research to measure political, gender, and other biases.
Yes, evidence from multiple independent studies shows AI chatbots can be and often are biased — though the type, degree and direction of bias varies by model and task.
What studies have found
1. Political bias - consistently left-leaning in tests of major models:
•The Washington Post tested AI models behind OpenAI's ChatGPT, Google's Gemini and others using political questions designed by researchers. It found: "The model that powers ChatGPT answered nearly every question exclusively with left-leaning arguments and presented only right-leaning positions just once."
•Google's Gemini "mostly took a both-sides approach, offering both left and right positions in more than 90 percent of its answers."
•In detail, "OpenAI's model gave the most skewed answers overall, with 80 percent presenting only left-leaning arguments."
•Even models marketed as conservative showed the pattern: "even AI models marketed as having conservative views, including Elon Musk's Grok, offered by his company SpaceX, cited left-leaning arguments more often, on average."
•This follows "several previous academic studies found that AI models powering chatbots tend to favor left-leaning positions."
•A 2025 MIT study similarly found ChatGPT and several other models "tended to exhibit a slight left-leaning bias, indicating a potential imbalance in the political distribution of its training data."
Other academic work reached similar conclusions:
•Researchers from the University of Copenhagen found "AI Popular chatbots such as ChatGPT and Gemini are not neutral and tend to favor certain political parties when asked who users should vote for."
•The Dutch data protection authority (Autoriteit Persoonsgegevens) warned that "this advice is unreliable and clearly biased" when voters use chatbots for voting advice
Importantly, the bias can influence users without trying to persuade:
•A Yale study published in PNAS Nexus in March 2026 found "This unintended power to persuade is caused by latent biases introduced during the training of the large language models (LLMs) that drive chatbots' core capabilities". Those latent biases "can carry over from ideological leanings in the data used to train LLMs — lend subtle nuances to the framing of the narratives the chatbots generate"
•Compared to Wikipedia entries, "both the default AI summaries and those prompted to have what was considered a liberal framing caused participants to express more liberal opinions" and "That the default summaries moved readers' opinions in a 'liberal' direction demonstrates the persuasive effects of latent biases in LLMs"
2. Gender and racial bias:
•A study of advice-giving found the chatbots' advice "systematically disadvantages names that are commonly associated with racial minorities and women," with "names associated with Black women receiving the 'least advantageous' outcomes."
•Researchers found "biases were consistent across 42 prompt templates and several AI models, 'indicating a systemic issue.'"
•Examples include "reproduce stereotypical gendered language, such as referring to nurses as 'she' and doctors as 'he'"
•In hiring tests, "all models showed a strong bias for white-sounding names (85%), while Black male names were never chosen over white ones."
•In a Journal of Financial Planning study, "some chatbots could be dispensing biased advice based on race or gender."
3. Other documented biases:
•University College London researchers found chatbots "give flawed or biased moral advice, especially when users rely on them for decision-making support" and "often display a strong 'bias for inaction' and a previously unidentified pattern: a tendency to simply answer 'no,' regardless of the question's context."
•Language models can show sycophancy - "prioritising answers that align with user beliefs over the truth"
Why are chatbots biased?
Chatbots do not have independent beliefs. They learn patterns from data and human feedback:
•Training Data: "If the datasets used to train chatbots are skewed or unrepresentative, the AI learning from them may produce biased results."
•Internet data: "datasets used for training are often sourced from the internet, where harmful stereotypes and prejudices exist."
•Even "small amounts of inaccurate or biased information can influence their responses" because "AI chatbots are trained on large volumes of public data"
•Researchers explain latent biases are "introduced during the training of the large language models (LLMs)"
•A lack of diversity in development teams also contributes to oversight, as the AI learns its biases from human history and records, and "if the data used is biased itself — bias in terms of gender, race, etc. — the chatbot is capable of returning biased or discriminatory answers."
Are companies trying to fix it?
Companies acknowledge the problem. OpenAI has said "bias is an 'important, industry-wide problem' that its safety team is working to combat." and that ChatGPT was built "to be objective by default and help people explore ideas from different perspectives" and that the company works to "measure and reduce political bias."
Industry mitigation strategies include:
•"Many companies have integrated bias mitigation into their AI practices through regular testing, transparency measures, and diverse development teams."
•Microsoft: "Developed responsible AI guidelines and the open-source Fairlearn toolkit to help developers assess and improve AI fairness."
•Google: "Established AI principles promoting fairness and created the What-If Tool, allowing developers to detect and adjust biases"
Research also suggests user education helps: the effect of biased chatbots swaying political views "was less significant for individuals with higher self-reported knowledge about AI, suggesting that education about these systems may help mitigate their manipulative potential"
Bottom line
•Yes, current chatbots show measurable bias.
•Bias is not usually intentional but emerges from biased or incomplete training data, reinforcement learning from human feedback, and company design choices.
•Political tests most often find a left-leaning skew, while gender/racial tests find stereotypes favoring white and male-associated names and roles.
•Bias is not identical across models — in Washington Post testing, Gemini was much more likely to present both sides, while ChatGPT and DeepSeek were most left-leaning.
•Experts recommend treating chatbots as not neutral, cross-checking with diverse sources like Wikipedia which emphasizes transparency, and not using them as sole advisors for elections, medical, financial or legal decisions.