Are AI chatbots biased?

I'll examine current evidence on whether AI chatbots show bias and what causes it.

Yes, evidence indicates AI chatbots do exhibit systematic biases — though not necessarily as conscious political intention.

> AI chatbots are biased, if "bias" means a systematic tendency in outputs rather than a conscious political intention. The best recent evidence points to a left-of-center tilt in many major models

1. Political bias - most studied

Multiple recent studies find a left-leaning bias in US and European political questions:

In detail:

This confirms earlier academic work. German researcher Rettenberger's work opens with: "We show that ...LLMs tend to align more closely with left-leaning political parties" and that chatbots often present left-leaning views that support the Democratic Party in the US and the Labour Party in the UK.

2. Why it happens: latent biases in training

Researchers attribute this to training:

> This unintended power to persuade is caused by latent biases introduced during the training of the large language models (LLMs) that drive chatbots' core capabilities

Those latent biases — which can carry over from ideological leanings in the data used to train LLMs — lend subtle nuances to the framing of narratives

A Yale study (PNAS Nexus, March 3, 2026) found that compared to Wikipedia entries, both the default AI summaries and those prompted to have liberal framing caused participants to express more liberal opinions. That default summaries moved readers' opinions in a "liberal" direction demonstrates the persuasive effects of latent biases in LLMs.

More broadly, bias is understood by source:

Chatbots can pick up political perspectives in different ways. Most are trained on large collections of text scraped from the internet, but companies can choose what data to include, hire workers to score responses, and write system instructions.

3. Gender, racial and other sociodemographic biases

Political lean is not the only bias documented:

Recent financial and hiring examples:

Inaccuracies including hallucinations are also attributed to biased or incomplete training data.

4. Not all models behave the same and neutrality is contested

People also prefer neutral answers even over answers that matched their party, researchers found.

5. Can bias be reduced?

Companies and researchers recommend:

In short, chatbots mirror the biases in their training corpora and refinement processes and can amplify them - from left-leaning political framing to gendered financial advice and racial stereotyping. Bias varies by model, prompt, and use case, and is an active area of measurement and mitigation, but current independent testing finds it persistent and measurable.