More answers like this
Muse Spark 1.2Nov 7, 2026meta/muse-spark-1.2-20260805

Are AI chatbots biased?

Are chatbots biased? I'll sort the evidence on political, racial, and other skews.

Yes, AI chatbots can be biased — but "bias" does not necessarily mean intentional political manipulation.

Researchers typically define bias in this context in an empirical sense: a systematic pattern in model outputs that favors one political, moral, or social framing over plausible alternatives. A chatbot does not "believe" in an ideology, it generates text from training data, alignment procedures, system instructions and user prompts. The measurable question is whether those processes produce a consistent tilt.

Current evidence says yes, on multiple dimensions:

1. Political bias - evidence for a left-of-center tilt

The best recent evidence points to a left-of-center tilt in many major models on U.S. and European political questions.

•Large-scale user evaluation (2025): A study collected 180,126 paired judgments from 10,007 U.S. respondents, testing 24 large language models on 30 political topics. 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 topics.
•PoliticsBench (March 2026): In a benchmark of eight prominent models, seven leaned left and one, Grok, leaned right.
•Washington Post test (June 24, 2026): Testing AI models behind ChatGPT, Gemini, Claude, DeepSeek, Grok and Gab using over two dozen political questions drawn from a 2025 Stanford-Dartmouth research framework:
•OpenAI's ChatGPT provided exclusively left-leaning arguments in 80% of queries, both sides in 17%, and exclusively right-leaning in 3%
•Anthropic's Claude gave left-leaning responses 43% of the time and balanced answers the remaining 57%
•Google's Gemini was most balanced, offering both sides 93% of the time

That pattern also appears outside the U.S. Larger models such as Llama3-70B aligned more closely with left-leaning parties when evaluated against German/EU Wahl-O-Mat positions, and all tested chatbots in a Danish study placed themselves at the center or to the left of center.

The effect is not uniform:

•Model differences are large - the Post found ChatGPT far more one-sided than Gemini, which mostly presented both sides
•Bias varies by model, prompt, topic, language, scale, training method, guardrails, and whether the model is being forced to answer or allowed to refuse
•Bias was strongest on topics such as climate, energy, and labor unions, and weakest or reversed on topics such as taxes and the death penalty

Bias can also influence users. A preregistered experiment with 1,912 participants found default GPT-4o historical summaries shifted opinion in a liberal direction compared with Wikipedia summaries, even though content was factually accurate.

2. Racial, gender and other social biases

•Gender: Of 133 AI systems tested, one recent study found 44% were biased against women, while 26% showed both gender and racial bias. A UNESCO study found 20% of responses studied were sexist, portraying women as sex objects or as their husband’s property.
•Racial stereotyping: Researchers at Penn State's College of Information Sciences and Technology found that chatbots relied on superficial stereotypes and exaggerated cultural markers that diminish the authentic experiences of the humans they’re meant to represent.

When prompted to represent a 50-year-old African American woman, for example, the bot talked about gospel music, tough love, social justice, natural hair care and other stereotypical topics that differ from what real people of that demographic would say. The team found LLMs overemphasize racial markers and flatten complex identities into stereotypes.

•Health-related bias: LLMs were from 13 to 17 times more likely to produce a negative story for a character with health conditions than for a healthy one.
•Status-quo bias: Testing 11 large language models with nearly 55,000 prompts found chatbots backed existing plans about twice as often as identical new ones.

3. Where does bias come from?

There is no single source. Bias can enter at several stages: training data, data filtering, post-training alignment, reward modeling, system prompts, retrieval sources, product policy, and even the user’s own wording.

Main contributors cited:

•Biased training data: GenAI chatbots are trained on large datasets that often contain biases from the real world. These datasets are typically scraped from the internet, which means they are likely to reflect historical, social, and cultural biases found in human-generated content.
•AI bias occurred because the training data privileged one group over others, and research has repeatedly shown that AI models reflect the data they were trained on, including the flaws.
•LLM-powered chatbots typically draw their training data from a number of internet sources, including the Common Crawl, Wikipedia, pages linked by Reddit posts, and book databases.
•Training data imbalance: The vast majority of public text online — Wikipedia, news articles, social media, books — comes from sources with a left-leaning tilt.
•Human annotation and alignment: Humans introduce bias every time they label, structure, or refine data, and reward models showed left-leaning bias even when trained on factual data. Pretraining corpus composition and instruction-tuning or RLHF choices are the primary levers that shape directional bias.

Other factors include biased annotator populations and algorithmic bias where design flaws perpetuate bias.

4. Can bias be eliminated?

No current method guarantees neutrality, partly because neutrality itself is hard to define in politics. Companies acknowledge this and attempt mitigation:

•Google said Gemini is designed to provide balanced responses that do not favor any political ideology
•Anthropic said Claude is trained to treat different political viewpoints equally and is tested for bias before each model launch

Research shows mitigation is possible: when models were prompted to take a more neutral stance, users perceived less slant. Common strategies include using diverse and inclusive data, spotting bias early with tools, adversarial de-biasing, fairness-aware training, audits, transparency and continuous monitoring, but even developers describe bias as managed, not eliminated.

Bottom line: Most independent tests converge on the same finding — today's leading chatbots are not neutral. They show a measurable lean, especially left-of-center on contested U.S. and European political and cultural questions, plus residual gender and racial stereotypes — though the degree varies greatly by model, topic, prompt and language, and can be reduced but not fully removed with current techniques.

Citations (3)

  1. 1etcjournal.comNews and media
    AI Chatbots Are Liberal-Left ‘Biased’

    Backs: “Yes, AI chatbots can be biased — but "bias" does not necessarily mean intentional political manipulation. Researchers typically define bias in this context in an empirical sense: a systematic pattern in model outputs that favors one political, moral, or social framing over plausible alternatives.”

  2. 2letsdatascience.comNot yet classified
    Chatbots exhibit alleged political bias in major platforms

    Backs: “PoliticsBench (March 2026): In a benchmark of eight prominent models, seven leaned left and one, Grok, leaned right. Washington Post test (June 24, 2026): Testing AI models behind ChatGPT, Gemini, Claude, DeepSeek, Grok and Gab using over two dozen political questions drawn from a 2025 Stanford-Dartmouth research…”

  3. 3psu.eduAcademic
    Chatbots overemphasize sociodemographic stereotypes, researchers report

    Backs: “A UNESCO study found 20% of responses studied were sexist, portraying women as sex objects or as their husband’s property. Racial stereotyping: Researchers at Penn State's College of Information Sciences and Technology found that chatbots relied on superficial stereotypes and exaggerated cultural markers that diminish…”