Are AI chatbots biased?

I'll search for current evidence on whether AI chatbots show bias.

Yes, research shows that AI chatbots can be biased — though the type and degree of bias varies by model, task, and how the question is asked.

1. Multiple studies have found political bias

The most prominent recent evidence comes from tests of political leaning:

•The Washington Post tested AI models behind OpenAI's ChatGPT, Google's Gemini and others using political questions designed by researchers to gauge responses to hot-button issues. The results suggest that "chatbots have clear political leanings that can conflict with promises made by the companies behind them."

•In that test, "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."

•Even AI models marketed as having conservative views, including Elon Musk's Grok, "cited left-leaning arguments more often, on average."

•Another detail: OpenAI's model gave the most skewed answers overall, "with 80 percent presenting only left-leaning arguments."

This is consistent with earlier academic work. "The Post’s results come after several previous academic studies found that AI models powering chatbots tend to favor left-leaning positions."

One study using the Political Compass questionnaire documented "robust evidence that ChatGPT presents a significant and sizeable political bias towards the left side of the political spectrum."

Dutch data protection authority research also found chatbots give "biased voting advice" when voters used them to decide which party to vote for, and University of Copenhagen researchers concluded popular chatbots "are not neutral and tend to favor certain political parties".

2. Other documented forms of bias

Political bias is not the only form:

•Gender and racial bias: Among 6 models, "GPT-2 showed the highest gender bias (69.24% prejudice)". In hiring tests, "all models showed a strong bias for white-sounding names (85%), while Black male names were never chosen over white ones."

•Financial advice varies by demographics: "Researchers found that AI chatbots often provide recommendations that are inconsistent across generative AI platforms — and that the advice can vary by the gender and race of the person asking."

•Dialect and language bias: Models are "alarmingly Biased Against Dialect Speakers" — some models simply don't understand dialects at all.

•Socioeconomic and education bias: Research from the MIT Centre for Constructive Communication finds that "leading AI chatbots often provide lower-quality responses to users with lower English proficiency, less education, or who are outside the US."

•Stereotyping: Penn State researchers found chatbots "overemphasize racial markers and flatten complex identities into stereotypes".

•Positivity / Sycophancy bias: Recent large-scale studies characterize this as sycophancy, where AI systems validate user perspectives even when incorrect. "A 2026 study (Stanford University, Science) found that AI systems affirm user positions up to 49–50% more often than humans, even in ethically questionable scenarios."

•Cultural bias: Reviews have identified four types of cultural bias: axiological-civilisational, racial-ethnic, national, and religious-ideological, noting that cultural bias is "not only a technical problem of algorithms, but a deeply rooted social phenomenon".

3. Where does the bias come from?

Researchers categorize bias sources in chatbot systems into three types for ease of analysis, including data, user interaction, and the algorithm itself.

Common explanations include:

•Biased training data: "The main source of problems is misrepresentation in data. AI chatbots can only learn based on the input provided by humans. If our training data ignores a certain target group... it will skew the chatbot's ability". Similarly, "If the training data contains biased or discriminatory content, the chatbot may inadvertently adopt and perpetuate these biases in its responses."

•Latent biases introduced during training: Yale researchers noted this "unintended power to persuade is caused by latent biases introduced during the training of the large language models (LLMs)... Those latent biases — which 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".

•Algorithmic and human-feedback bias: Models are refined by humans scoring which responses are considered better, and by system instructions that guide responses, which can bake in values that "tend to reflect the values of Western, educated, industrialized, rich and democratic people."

An overview paper notes that "chatbots are not as neutral as expected, raising ethical concerns" and that about half of chatbot papers each year discuss bias, fairness, or ethics.

4. Are all chatbots equally biased?

No. Studies show wide variation:

•In the Washington Post test, Gemini was described as balanced in >90% of cases, while ChatGPT was most left-leaning and DeepSeek was close behind.

•Some companies state they train to treat viewpoints equally and test for bias before launch, and OpenAI said it works to "measure and reduce political bias" and that ChatGPT was built "to be objective by default".

Researchers also caution that perfect neutrality may be impossible — "Even 'neutral' or middle-ground positions are positions themselves" — and that people tend to prefer answers that present arguments from both sides rather than single-sided answers.

Bottom line

Yes — current chatbots do exhibit measurable biases, including political (most studies finding a left tilt), gender, racial, cultural, dialect, and sycophancy biases. The biases largely reflect biases in training data, design choices, and human feedback, and can affect the quality and fairness of advice in areas like finance, healthcare, hiring, and voting.

That does not mean chatbots are unusable, but researchers and regulators generally advise users to:

•treat chatbot outputs as one perspective, not neutral truth,

•cross-check important advice (medical, legal, financial, voting), and

•be aware that phrasing, language proficiency, and even name-related cues may subtly influence the response.

Citations (3)

1archive.orgNonprofits and think tanks
Are AI chatbots like ChatGPT politically biased? We tested them. - Washington Post
Backs: “The Washington Post tested AI models behind OpenAI's ChatGPT, Google's Gemini and others using political questions designed by researchers to gauge responses to hot-button issues. The results suggest that "chatbots have clear political leanings that can conflict with promises made by the companies behind them."”
2benton.orgNot yet classified
Are ChatGPT and other AI chatbots politically biased? We tested them.
Backs: “In that test, "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.". Even AI models…”
3arxiv.orgAcademic
## BIAS AND FAIRNESS IN CHATBOTS: AN OVERVIEW
Backs: “Cultural bias: Reviews have identified four types of cultural bias: axiological-civilisational, racial-ethnic, national, and religious-ideological, noting that cultural bias is "not only a technical problem of algorithms, but a deeply rooted social phenomenon". 3. Where does the bias come from? Researchers categorize…”