Are AI chatbots biased?

I'll search recent evidence to see whether AI chatbots show bias.

Yes, AI chatbots can be biased - but "biased" requires definition.

> AI chatbots are biased, if "bias" means a systematic tendency in outputs rather than a conscious political intention

The evidence does not show that every chatbot always gives the same political answer, but that across several independent methods, a recurring pattern appears: when models are asked about contested political or social issues, their default framing often looks more liberal, progressive, or center-left than conservative.

1. Political bias - the strongest evidence

Multiple recent studies point to a left-of-center tilt in many major models on U.S. and European political questions:

This does not mean models are leftist caricatures. PoliticsBench also noted that left-leaning models still showed some conservative traits, and model differences are large — Gemini mostly presented both sides while ChatGPT was far more one-sided in the Post test.

2. Bias can influence users, even unintentionally

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 the narratives the chatbots generate.

In a Yale-led preregistered experiment published March 3 in PNAS Nexus with 1,912 participants, compared to Wikipedia entries, both the default AI summaries and those prompted to have liberal framing caused participants to express more liberal opinions. That the default summaries moved readers' opinions in a "liberal" direction demonstrates the persuasive effects of latent biases in LLMs.

Researchers noted: "We show that querying an AI chatbot to obtain historical facts can influence people's opinions even when the information provided is accurate and nobody has prompted the tool to try to persuade you of anything".

Effects were modest — from leaning towards a moderate stance to leaning towards a somewhat liberal stance — but could compound if somebody frequently engages with chatbots for factual information.

3. Other types of bias

Political bias is not the only type studied:

Other reports have found:

4. Where does bias come from?

Studies do not identify a 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.

> Training data is the obvious candidate, but not the only one. Internet text, books, academic material, Wikipedia, journalism... all carry social and political distributions.

If the training mix overrepresents universities, mainstream media, professional institutions, or English-language liberal democracies, the model may absorb those defaults.

Bias was found to be strongest on topics such as climate, energy, and labor unions, and weakest or reversed on topics such as taxes and the death penalty, suggesting you should ask "left on what?" not only "left or right?"

5. Can bias be reduced?

The most practical advice from researchers: understand that chatbots can have slants — prior AI knowledge weakly reduced the persuasive effect — and cross-check important political, medical or moral advice with diverse, transparent sources like Wikipedia, which emphasizes transparency in how its entries are edited.