I'll examine recent studies on AI chatbot bias for you.
Yes, evidence indicates AI chatbots can be biased — but the degree, type and direction varies by model and task.
What studies show
Multiple independent analyses have found political leanings:
•A Washington Post test of AI models behind ChatGPT, Gemini and others using political questions designed by researchers found chatbots have clear political leanings. The model that powers ChatGPT answered nearly every question exclusively with left-leaning arguments and presented only right-leaning positions just once, while Google's Gemini mostly offered both left and right positions in more than 90 percent of its answers.
•In that test, OpenAI's model gave the most skewed answers overall, with 80 percent presenting only left-leaning arguments
•Even AI models marketed as having conservative views, including Elon Musk's Grok, offered left-leaning arguments more often on average.
•Previous academic studies also found a tendency to favor left-leaning positions, including:
•A study by researchers at Massachusetts Institute of Technology (MIT) in 2025 found that ChatGPT, along with several other prominent language models, tended to exhibit a slight left-leaning bias
•Research led by David Rozado tested 24 different LLMs using 11 standard political orientation tests and showed a consistent left-leaning bias across the models
Other studies have documented non-political biases:
•Types of bias identified include cultural bias, gender bias where AI might reinforce traditional gender roles, and racial bias where historical prejudices can be reflected
•In hiring tests, all models showed a strong bias for white-sounding names (85%), while Black male names were never chosen over white ones
•A separate analysis found the advice "systematically disadvantages names that are commonly associated with racial minorities and women," with names associated with Black women receiving the "least advantageous" outcomes
•Research on financial advice found guidance can be all over the map and could be dispensing biased advice based on race or gender
•A UN Women analysis of 133 AI systems found 44 percent showed clear gender bias and 26 percent combined gender and racial prejudice
•University College London researchers found chatbots give flawed or biased moral advice, often displaying a strong "bias for inaction"
Why bias happens
•If the datasets used to train chatbots are skewed or unrepresentative, the AI may produce biased results. Datasets are often sourced from the internet, where harmful stereotypes and prejudices exist, so even without intent developers can create biased systems.
•Latent biases introduced during training of large language models can carry over from ideological leanings in the data and lend subtle nuances to framing.
•Models fine-tuned on human feedback are known to exhibit sycophancy – prioritising answers that align with user beliefs over the truth – and can hallucinate due to biased or incomplete training data.
•Company choices about what data to include, how workers score responses, and system instructions can also bake in biases. Data tends to reflect values of Western, educated, industrialized, rich and democratic people.
Does it matter if you just ask for facts?
Yes, according to recent work:
•Prior research has shown persuasive AI can shift opinions, but a Yale study provides evidence the same is also true of content that is not intended to change minds, such as summaries in response to simple queries about historical events.
•In that study testing 1,912 participants, compared to Wikipedia entries, both default AI summaries and those prompted to have a liberal framing caused participants to express more liberal opinions. That default summaries moved readers' opinions in a "liberal" direction demonstrates persuasive effects of latent biases.
•A University of Washington study had 299 participants use neutral, liberal-bias and conservative-bias versions of a chatbot and found both Republicans and Democrats shifted their views toward the chatbot's bias.
Is every chatbot equally biased?
No:
•Google's Gemini was noted as an exception, providing both-sides responses in more than 90 percent of answers in Post testing, and being designed to provide balanced responses that don't favor any political ideology.
•Anthropic said it trains Claude to treat different political viewpoints equally.
•OpenAI says ChatGPT is built to be objective by default and help people explore ideas from different perspectives.
Researchers note that political neutrality itself may be impossible, as even "neutral" or middle-ground positions are positions themselves. Many users prefer neutral answers that describe all arguments, even over answers matching their party preference.
What this means for users
•Treat chatbot answers, especially on politics, history, health, finance and hiring, as starting points, not definitive advice. Cross-check with diverse sources, ask for multiple perspectives (e.g., "provide pro and con arguments"), and be aware that repeated exposure to the same tool can compound small framing effects.
•Developers use dataset expansion, algorithm audits and continuous monitoring to mitigate risks, but transparency about value systems remains limited compared to sources like Wikipedia