Are AI chatbots biased?

YesSources conflict

Most sources report a left-leaning bias in ChatGPT, while the Euronews study reports ChatGPT shifting rightward over time.

AI chatbots can indeed exhibit various forms of bias, and this is a well-documented issue in AI research. Let me search for some current information to give you a comprehensive answer.## Yes, AI chatbots can be biased in multiple documented ways

Research consistently shows that AI chatbots exhibit various forms of bias, stemming primarily from their training data and design choices.

Political Bias

This is one of the most studied areas. Multiple academic studies have found consistent patterns:

•A University of East Anglia study published in Public Choice found that the artificial intelligence platform ChatGPT shows a significant and systemic left-wing bias, with responses favoring the Democrats in the US, the Labour Party in the UK, and President Lula da Silva's Workers' Party in Brazil.

•Similarly, research comparing models found that Gemini exhibited a more pronounced liberal and left-wing tendency compared to ChatGPT, and these political biases varied depending on the language used for inquiry.

•A large-scale 2025 study involving 180,126 paired judgments from 10,007 U.S. respondents, testing 24 large language models on 30 political topics, found that nearly all evaluated models were perceived as significantly left-leaning, and that one widely used model leaned left on 24 of the 30 topics.

•Interestingly, this isn't static—one 2025 study found ChatGPT is seeing a rightward shift in political values over time, which researchers attributed to a change in the datasets used to train their models, the number of interactions with users, or changes and updates to the chatbot.

•Some research also found that ChatGPT not only produces more left-leaning text and images but also often refuses to generate content that presents conservative perspectives.

•Even OpenAI has acknowledged this issue, stating their models stay near-objective on neutral or slightly slanted prompts, and exhibit moderate bias in response to challenging, emotionally charged prompts. They reported that newer models show improved bias levels and greater robustness to charged prompts, reducing bias by 30% compared to our prior models.

Gender Bias

Chatbots also reflect gender biases present in their training data:

•ChatGPT's training data includes a disproportionate amount of text reflecting male-centric perspectives, so when asked about leadership, it often associates traditionally masculine traits with effective leadership while neglecting nurturing traits.

•In medical contexts, studies found GPT-4 was significantly more likely to recommend treatments and further clinical testing for males than females.

•A UC Santa Cruz study on empathy found GPT-4o empathized more when told the person it was responding to was female, with this over-empathizing present when told the person it was chatting with was female, and more similar to a typical human response when told the person was male.

Why This Happens

The root causes are structural: when training data reflects societal stereotypes or underrepresents certain demographics, chatbots will learn to replicate as well.

Why This Matters

Bias in chatbots isn't just academic—it has real behavioral effects:

•A University of Washington study found that AI chatbots with political biases can sway the opinions and decisions of users who interact with them, and this effect occurred regardless of a person's own political party and after only a few conversational exchanges.

•Researchers warn that unchecked bias could lead chatbots to deliver "skewed information," which could further polarise society or create "echo chambers" that reinforce a user's particular beliefs.

The Bottom Line

Bias in AI chatbots is a well-documented, active area of research—spanning political leanings, gender stereotypes, and emotional responses. It stems largely from the data these models are trained on (which reflects existing societal biases) and the decisions made during fine-tuning. Companies like OpenAI are actively working to measure and reduce this bias, but it remains an evolving challenge rather than a solved problem.

Citations (19)

1uea.ac.ukAcademic
Fresh evidence of ChatGPTs political bias revealed by comprehensive new study
Source excerpt: “Published today in the journal Public Choice, the findings show that ChatGPT’s responses favour the Democrats in the US, the Labour Party ... # Fresh...”
evidence, article, points to yes. Claim: UEA study in Public Choice found ChatGPT shows systemic left-wing bias favoring Democrats, Labour, and Lula's Workers' Party.
2uea.ac.ukAcademic
Fresh evidence of ChatGPTs political bias revealed by comprehensive new study
Source excerpt: “Published today in the journal Public Choice, the findings show that ChatGPT’s responses favour the Democrats in the US, the Labour Party in the UK, a...”
evidence, article, points to yes. Claim: UEA study found ChatGPT responses favor Democrats in the US, Labour in the UK, and Lula's party in Brazil.
3arxiv.orgAcademic
Language-Dependent Political Bias in AI: A Study of ChatGPT and Gemini
Source excerpt: “A comparative analysis revealed that Gemini exhibited a more pronounced liberal and left-wing tendency compared to ChatGPT. ”
evidence, official document, points to yes. Claim: Gemini exhibited a more pronounced liberal and left-wing tendency than ChatGPT.
4arxiv.orgAcademic
Language-Dependent Political Bias in AI: A Study of ChatGPT and Gemini
Source excerpt: “The study also found that these political biases varied depending on the language used for inquiry. ”
evidence, official document, points to yes. Claim: Political biases varied depending on the language used for inquiry.
5etcjournal.comNews and media
AI Chatbots Are Liberal-Left ‘Biased’
Source excerpt: “** The most direct 2025 user-evaluation study collected 180,126 paired judgments from 10,007 U.S. respondents, testing 24 large language models on 30 ...”
statistic, article, points to yes. Claim: 2025 study with 180,126 paired judgments from 10,007 U.S. respondents tested 24 LLMs on 30 political topics.
6etcjournal.comNews and media
AI Chatbots Are Liberal-Left ‘Biased’
Source excerpt: “It found that nearly all evaluated models were perceived as significantly left-leaning, and that one widely used model leaned left on 24 of the 30 top...”
statistic, article, points to yes. Claim: Nearly all models were perceived as left-leaning; one widely used model leaned left on 24 of 30 topics.
7euronews.comNews and mediaquality 0.82
ChatGPT may be shifting ‘rightward’ in political bias, study finds
Source excerpt: “Chinese researchers have found that ChatGPT, OpenAI’s popular artificial intelligence (AI) chatbot, is seeing a rightward shift in political values. ”
evidence, article, points to yes. Claim: A 2025 study found ChatGPT is seeing a rightward shift in political values over time.
8euronews.comNews and mediaquality 0.82
ChatGPT may be shifting ‘rightward’ in political bias, study finds
Source excerpt: “The researchers give three theories for this rightward shift: a change in the datasets used to train their models, the number of interactions with use...”
attribution, article, points to yes. Claim: Researchers attributed the shift to training data changes, user interactions, or chatbot updates.
9psypost.orgNews and mediaquality 0.84
Scientists reveal ChatGPT's left-wing bias — and how to "jailbreak" it
Source excerpt: “The research shows that the system not only produces more left-leaning text and images but also often refuses to generate content that presents conser...”
evidence, article, points to yes. Claim: ChatGPT produces more left-leaning text and images and often refuses to generate conservative-perspective content.
10openai.comCompanies
Defining and evaluating political bias in LLMs
Source excerpt: “# Defining and evaluating political bias in LLMs ## Overview and summary of findings Based on this evaluation, we find that our models stay near-objec...”
attribution, official document, points to yes. Claim: OpenAI acknowledged its models stay near-objective on neutral prompts but show moderate bias on charged prompts.
11openai.comCompanies
Defining and evaluating political bias in LLMs
Source excerpt: “GPT‑5 instant and GPT‑5 thinking show improved bias levels and greater robustness to charged prompts, reducing bias by 30% compared to our prior model...”
statistic, official document, points to yes. Claim: Newer models reduce bias by 30% compared to prior models.
12sciencedirect.comAcademic
Gender biases within Artificial Intelligence and ChatGPT: Evidence, Sources of Biases and Solutions - ScienceDirect
Source excerpt: “Representation bias In the case of ChatGPT, the training data includes a disproportionate amount of text reflecting male-centric perspectives. As a r...”
evidence, official document, points to yes. Claim: ChatGPT's training data includes disproportionate male-centric text, associating masculine traits with leadership.
13sciencedirect.comAcademic
Gender biases within Artificial Intelligence and ChatGPT: Evidence, Sources of Biases and Solutions - ScienceDirect
Source excerpt: “Such biases have already been proven to occur in treatment plans recommended by GPT-4, where studies done to measure the sex difference in likelihood ...”
evidence, official document, points to yes. Claim: GPT-4 was more likely to recommend treatments and clinical testing for males than females.
14ucsc.eduAcademic
AI chatbots perpetuate biases when performing empathy, study finds - News
Source excerpt: “They also found the bot empathized more when told the person it was responding to was female. ”
evidence, article, points to yes. Claim: UC Santa Cruz study found GPT-4o empathized more when told the person was female.
15ucsc.eduAcademic
AI chatbots perpetuate biases when performing empathy, study finds - News
Source excerpt: “The researchers noticed that this over-empathizing was present when told that the person it was chatting to was a female, and was more similar to a ty...”
evidence, article, points to yes. Claim: Over-empathizing occurred when told the person was female, and responses were more human-like when told male.
16sciencedirect.comAcademic
Gender biases within Artificial Intelligence and ChatGPT: Evidence, Sources of Biases and Solutions - ScienceDirect
Source excerpt: “When training data reflects societal stereotypes or underrepresents certain demographics, chatbots will learn to replicate as well.”
evidence, official document, points to yes. Claim: Training data reflecting stereotypes or underrepresenting demographics causes chatbots to replicate bias.
17psypost.orgNews and mediaquality 0.84
Just a few chats with a biased AI can alter your political opinions
Source excerpt: “A new study shows that artificial intelligence chatbots with political biases can sway the opinions and decisions of users who interact with them. ”
evidence, article, points to yes. Claim: University of Washington study found politically biased chatbots can sway users' opinions and decisions.
18psypost.orgNews and mediaquality 0.84
Just a few chats with a biased AI can alter your political opinions
Source excerpt: “This effect occurred regardless of a person’s own political party and after only a few conversational exchanges. ”
evidence, article, points to yes. Claim: The persuasion effect occurred regardless of party and after only a few conversational exchanges.
19euronews.comNews and mediaquality 0.82
ChatGPT may be shifting ‘rightward’ in political bias, study finds
Source excerpt: “If left unchecked, researchers warned AI chatbots could start to deliver “skewed information,” which could further polarise society or create “echo ch...”
attribution, article, points to yes. Claim: Researchers warn unchecked bias could deliver skewed information, polarise society, or create echo chambers.