Are AI chatbots biased?

Yes — multiple strands of recent research show that AI chatbots exhibit measurable biases, though the type and degree vary by model and dimension.

Political bias
A Washington Post test of major chatbots found that the results suggest that chatbots have clear political leanings that can conflict with promises made by the companies behind them, with the model powering ChatGPT answering nearly every question exclusively with left-leaning arguments and presenting 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. Even models marketed as conservative-leaning weren't immune — AI models marketed as having conservative views, including Grok, cited left-leaning arguments more often, on average. A separate academic comparison analyzing dozens of models across countries similarly found that most models leaned center-left or left, though model scale and openness were not strong predictors, pointing instead toward alignment strategy and institutional context.

A University of Copenhagen study on election-related queries found that popular AI chatbots such as ChatGPT and Gemini are not neutral and tend to favor certain political parties when asked who users should vote for, making them unsuitable for providing election advice. That study also noted all of the tested chatbots place themselves at the center or to the left of center on the political spectrum, clustering close to center-left parties in the Danish context.

Importantly, this political tilt isn't just theoretical — it can shift real opinions. A Yale study found AI chatbots can subtly influence users' social and political opinions through unintended latent biases that can carry over from ideological leanings in the training data. Specifically, compared to Wikipedia entries, both the default AI summaries and those prompted with liberal framing caused participants to express more liberal opinions about historical events.

Bias against certain user groups
An MIT Media Lab study found something arguably more concerning: state-of-the-art chatbots like GPT-4, Claude 3 Opus, and Llama 3 sometimes provide less-accurate and less-truthful responses to users with lower English proficiency, less formal education, or those from outside the US, and refuse to answer questions at higher rates for these users, sometimes responding with condescending language. The effects compounded for people at intersections of disadvantage — users with less formal education who were also non-native English speakers saw the largest declines in response quality. Country of origin mattered too: Claude 3 Opus in particular performed significantly worse for users from Iran compared to the US and China.

Age and other demographic bias
A UOC-led study of five popular chatbots found that while chatbots have taken steps to avoid sexism and gender bias, they still continue to discriminate on grounds of age, with significant age-related biases affecting users even as models are more cautious about sexist stereotypes.

Bias in everyday outputs (not just politics)
A UC San Diego study found LLM-generated content skews positive, which has real behavioral consequences: customers are 32% more likely to buy a product after reading a chatbot-generated review summary than the original human review, because LLMs introduce positive framing bias that affects user decision-making. The same study found LLM-generated summaries changed the sentiment of reviews in 26.5% of cases.

Why this happens
Researchers attribute these patterns to training data patterns, human feedback during reinforcement learning, and safety guidelines that inadvertently encode specific value systems.

Bottom line
As one 2026 analysis summarized: a careful answer to "Are AI chatbots biased?" is yes — though the nature and strength of bias is not stable; models change often, system prompts change, and a test from one month may not describe a model a few months later, which is why audits should be repeated and model-specific. Some companies are actively working to reduce this: in October 2025, OpenAI announced that GPT-5 Instant and GPT-5 Thinking models show a 30% reduction in political bias compared to GPT-4o.

In short: yes, chatbots demonstrably carry biases — political, demographic, and framing-related — rooted in their training data and design choices, and these biases can meaningfully shape user perceptions, though the degree varies significantly across models and is an active target of ongoing mitigation efforts.