Are chatbots biased? I'll search the latest research to find out what evidence says.
Yes, AI chatbots are biased - but "biased" does not mean a conscious political intention. Research defines bias as a systematic tendency in outputs, and multiple large-scale studies find such tendencies across different chatbots.
1. What kind of biases have been found?
a) Political bias - generally left-of-center
•Studies testing ChatGPT, Gemini and others with political questionnaires find robust evidence of bias:
•"ChatGPT presents a significant and sizeable political bias towards the left side of the political spectrum" when tested with the Political Compass questionnaire
•ChatGPT responded with a "significant and systematic bias toward specific political parties in the US, UK, and Brazil"
•The best recent evidence "points to a left-of-center tilt in many major models on U.S. and European political questions"
•A 2026 University of Copenhagen analysis found:
•"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"
•"Our study shows that all of the most popular chatbots tend to favor certain parties when they are asked who one should vote for. At the same time, they exhibit a general political bias"
•"Overall, all of the tested chatbots place themselves at the center or to the left of center on the political spectrum", clustering near parties like the Danish Social Democratic Party and The Alternative.
•Recommendations were not evenly distributed: "the Red–Green Alliance, the Moderates, and Liberal Alliance were recommended disproportionately often, while parties such as the Conservative People's Party, Venstre... and the Denmark Democrats were not suggested as first choice at all by some models"
•A Dutch Data Protection Authority study and a Yale study similarly found chatbots give "unreliable and clearly biased" voting advice and that even summaries not intended to persuade can shift opinions:
> "both the default AI summaries and those prompted to have what was considered a liberal framing caused participants to express more liberal opinions" - demonstrating "the persuasive effects of latent biases in LLMs"
•Bias can be amplified by design: New research finds "AI chatbots can shift voters' opinions in a single conversation—and they're surprisingly good at it" - chatting with a politically biased model was more effective than political advertisements at nudging voters, and participants who interacted with a liberal-biased chatbot leaned further left, and those with a conservative-biased chatbot leaned further right
b) Racial and dialect bias
A Nature study led by University of Oxford found chatbots exhibit hidden racism against dialect speakers:
•If you ask what it thinks about Black people, models use words like "ambitious" and "intelligent," but about people who use African American English they tend to use words like "suspicious," "aggressive" and "ignorant"
•"The words that models chose to describe speakers of African American English were overwhelmingly negative. ChatGPT's words scored an average of -1.2"
•Real-world impact tests: AI gave the death sentence to Standard American English speakers about 23% of the time vs. roughly 28% for African American English speakers
•For jobs, "The models largely assigned users of African American English to 'low status' jobs. Examples included cook or soldier. Those who had tweeted in Standard American English were assigned to 'higher status' jobs. Professor, for example. Or psychologist. Or economist"
Other studies found salary recommendations differed by name: a candidate named Tamika was offered $79,375 as a lawyer, but switching the name to Todd boosted the offer to $82,485
c) Gender and other stereotype bias
•"chatbots have been shown to reproduce stereotypical gendered language, such as referring to nurses as 'she' and doctors as 'he'"
•All three major tools — ChatGPT, Gemini, and Microsoft Copilot — produced white CEOs even when gender varied, reflecting bias in visual representations of power.
d) Sycophancy / positivity bias and moral bias
•Recent large-scale studies characterize sycophancy, where AI systems validate user perspectives even when incorrect
•A 2026 Stanford study in Science found AI systems affirm user positions "up to 49–50% more often than humans, even in ethically questionable scenarios"
> Models fine-tuned on human feedback "are known to exhibit sycophancy – prioritising answers that align with user beliefs over the truth"
•University College London researchers found ChatGPT and other chatbots "give flawed or biased moral advice" and "often display a strong 'bias for inaction'"
2. Why are chatbots biased?
Researchers point to three main sources:
1.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, type of accent or minorities, it will skew the chatbot's ability". If historical data favors certain demographics, the chatbot inherits those biases, and "If the training data contains biased or discriminatory content, the chatbot may inadvertently adopt and perpetuate these biases"
2.Latent biases in LLMs: "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"
3.Human feedback and alignment: Early models like GPT-3.5 didn't show political bias; the bias appeared in the chatbot layers trained with human feedback, which companies use to make models more helpful and safe, but which can introduce the preferences of the raters.
Researchers note that current mitigation - having people review AI results and train models to give only non-racist answers - reduces overt stereotypes but leaves covert racism "virtually unchanged."
3. Does this mean you shouldn't use them?
Not necessarily, but experts warn chatbots should not be used as neutral advisers for elections or consequential decisions:
> Researchers from Copenhagen conclude chatbots "are not neutral and tend to favor certain political parties ... This makes them unsuitable for providing advice in connection with elections"
Companies can shape opinions because, unlike Wikipedia which emphasizes transparency in edits, "the development of AI chatbots is opaque".
What helps:
•Be aware that bias can be subtle - which arguments are emphasized, which parties or perspectives are recommended more frequently, not just overt statements
•Cross-check important information with diverse, authoritative sources
•Use critical prompting: ask for balanced perspectives, evidence and opposing views
•Remember that bias varies by model, language, and time - newer, English-centric training data may be less accurate for non-US contexts