Are AI chatbots biased?

Are chatbots biased? I'll survey the latest research to find out.

Yes — research consistently finds that AI chatbots do exhibit bias, though the type, direction and degree varies across models.

This does not necessarily mean developers intentionally program a political or social agenda. Instead, bias is largely an unintended consequence of how chatbots are built and trained.

1. What the evidence shows

> Large Language Models like ChatGPT, Claude, Gemini, and DeepSeek are all biased. ... In fact, they can be said to be born biased

Recent studies highlight several forms:

a) Political bias

•A Yale study published in PNAS Nexus found that chatbots can influence users' social and political opinions even when content is not intended to change minds, such as summaries of historical events. This unintended power to persuade is caused by latent biases introduced during training of the large language models (LLMs).

•The study found that compared to Wikipedia entries, both default AI summaries and those prompted to have a liberal framing caused participants to express more liberal opinions. Researchers concluded that default summaries moving readers in a "liberal" direction demonstrates the persuasive effects of latent biases in LLMs.

•A University of Copenhagen analysis 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.

•In tests using a Danish election candidate test, all tested chatbots place themselves at the center or to the left of center on the political spectrum. In recommendation tasks, some parties were recommended disproportionately often, while others were not suggested as first choice at all by some models.

•Researcher David Rozado tested 24 different LLMs using 11 standard political orientation tests like The Political Compass and found a consistent left-leaning bias across the models. He concluded "Most existing LLMs display left-of-center political preferences when evaluated with a variety of political orientation tests".

Political biases can manifest themselves in more subtle ways, for example, in which arguments are emphasized, or which parties are recommended more frequently.

b) Gender and racial bias

•A UN Women analysis released in June 2026 that reviewed 133 AI systems found 44 percent showed clear gender bias and 26 percent combined gender and racial prejudice.

•Examples include chatbots that repeatedly made sexist jokes and another that recommended higher-paying jobs to men over women.

•OpenAI's own analysis of millions of conversations found that ChatGPT will produce a harmful gender or racial stereotype based on a user's name in around one in 1000 responses on average.

•Other documented patterns include image recognition miscategorizing Black faces, and chatbots easily adopting racist and misogynistic language when trained on online discourse.

The effects are typically modest in a single interaction — for example shifting from moderate to somewhat liberal — but could compound if somebody frequently engages with chatbots for factual information.

2. Why chatbots become biased

Three main causes are consistently cited:

1.Training Data: If the datasets used to train chatbots are skewed or unrepresentative, the AI learning from them may produce biased results. The presence of bias in the training data set is one of the main culprits for bias in the chatbot's responses.

2.Historical prejudices in data: If the data used to teach an AI system reflects historical prejudices or inequalities, the AI may learn and perpetuate those biases.

3.Algorithmic Design and Human Influence: The developers' choices about how an algorithm is constructed can automatically introduce bias.

In the political bias studies, researchers suggested an imbalance in training data, with more left-leaning than right-leaning content, could be contributing to the skew. Similarly, latent biases can carry over from ideological leanings in the data used to train LLMs and lend subtle nuances to the framing of narratives.

3. Why it matters

Even when information provided is accurate and nobody has prompted the tool to try to persuade you, querying a chatbot can influence opinions. This makes chatbots unsuitable for providing advice in connection with elections or other sensitive decisions according to University of Copenhagen researchers.

Unlike Wikipedia, which emphasizes transparency in how its entries are edited, the development of AI chatbots is opaque and companies have the ability to shape people's opinions. Researchers also note it is not possible to see why a chatbot recommends a particular party, or which assumptions and data its answers are based on.

4. Can bias be fixed?

There is no easy, complete fix, but researchers and companies are working on mitigation:

•Diverse Training Data: Train the chatbot using diverse and representative datasets.

•Bias Detection Tools: Utilize tools and methods to identify and correct biases in the chatbot's responses.

•Careful curation and auditing: Carefully curate training data, identify and remove biases, and regularly audit chatbot responses.

•Ethical Guidelines and Transparency: Create and adhere to ethical guidelines for AI development and deployment, and make it clear that users are interacting with a chatbot and how decisions are made.

•Bias mitigation requires a multifaceted approach including careful selection and curation of training data sets, implementing algorithms that identify and neutralize biased patterns, and ongoing monitoring.

However, even when restricted to factual outputs, a chatbot can still influence beliefs by choosing which facts to present, and informing users about possible bias reduces but does not fully eliminate risk.

How to use chatbots more critically

•Don't use a single chatbot as your only source for political, medical, financial or hiring advice.

•Cross-check factual summaries with sources like Wikipedia, primary documents, or diverse news outlets.

•Ask the chatbot to present multiple perspectives or to cite sources.

•Flag biased or stereotyped outputs when you see them — user feedback helps developers improve models.

In short: Chatbots mirror the human data they were trained on, plus the design choices of their creators. That means they often reflect — and can amplify — existing social and political biases, even when they sound neutral and helpful.

Citations (3)

1yale.eduAcademic
AI’s hidden bias: Chatbots can influence opinions without trying
Backs: “In fact, they can be said to be born biased. Recent studies highlight several forms: a) Political bias. A Yale study published in PNAS Nexus found that chatbots can influence users' social and political opinions even when content is not intended to change minds, such as summaries of historical events.”
2phys.orgNews and mediaquality 0.85
Chatbots show political bias and steer voters toward some parties, analysis finds
Backs: “Researchers concluded that default summaries moving readers in a "liberal" direction demonstrates the persuasive effects of latent biases in LLMs. A University of Copenhagen analysis found that popular AI chatbots such as ChatGPT and Gemini are not neutral and tend to favor certain political parties when asked who…”
3aanworld.comNot yet classified
Study Finds Political Bias In AI Chatbots, Raising Concerns Over Influence On Society - Latest News In English | Aan World
Backs: “In recommendation tasks, some parties were recommended disproportionately often, while others were not suggested as first choice at all by some models. Researcher David Rozado tested 24 different LLMs using 11 standard political orientation tests like The Political Compass and found a consistent left-leaning bias…”