AI Biases Sway Public Understanding of History

Study finds AI chatbot summaries of historical events can shape reader opinions.

Published on Mar. 4, 2026

Got story updates? Submit your updates here. ›

A new study has found that even subtle biases in large language models (LLMs) used by AI chatbots can influence how the public perceives and understands historical events. Researchers asked 1,912 participants to read summaries of the 1919 Seattle General Strike and the 1968 Third World Liberation Front protests, with some summaries framed from a liberal or conservative perspective. They found that the AI-generated summaries, even with default framing, led to more liberal opinions compared to Wikipedia summaries, while conservative-framed summaries resulted in slightly more conservative views.

Why it matters

As more people turn to AI chatbots to learn about the world, the potential for these systems' underlying biases to shape public understanding of history and current events is concerning. This study highlights how even small differences in how information is presented can sway people's perspectives, which could have broader societal implications.

The details

The study, conducted by Daniel Karell and colleagues, explored the effects of both unintentional and intentional political biases in LLMs. Participants were asked to read summaries of two 20th century historical events - the 1919 Seattle General Strike and the 1968 Third World Liberation Front student protests - generated by GPT-4o and compared to Wikipedia summaries. Some AI summaries were explicitly framed from a liberal or conservative perspective, while others used the model's default framing. After reading the summaries, participants were asked to share their opinions on issues related to the events, such as the appropriateness of labor strikes and the use of curricula to advance social justice causes.

  • The study was conducted in 2026.

The players

Daniel Karell

The lead researcher who explored the effects of political biases in large language models.

GPT-4o

The large language model used to generate summaries of the historical events for the study.

Got photos? Submit your photos here. ›

What they’re saying

“People seeking unbiased information from LLMs may be subtly influenced by the models' latent biases, and this may have consequences for society writ large.”

— Daniel Karell, Lead Researcher (Mirage News)

The takeaway

This study highlights the potential for AI-generated content to sway public understanding of history and current events, even in subtle ways. As AI chatbots become more prevalent, it will be crucial to address and mitigate biases in the underlying language models to ensure the public has access to accurate and unbiased information.