Oxford study finds AI tools subtly bias social media posts

New research published on July 6 reveals that AI writing and editing tools embedded in social media platforms can introduce hidden biases into users' posts, accumulating across millions of interactions to gradually shift public opinion on contested topics.

How AI Rewrites Reshape Public Discourse

The study, titled "AI-Mediated Communication Can Steer Collective Opinion," was conducted by researchers from the Oxford Internet Institute at the University of Oxford and the Hasso Plattner Institute at the University of Potsdam, Germany. It found that large language models consistently changed the direction of social media posts on contested topics, even when explicitly instructed to preserve the original meaning.eurekalert+1

The researchers tested four LLMs — Llama 3.1, Gemma 3, Ministral, and Qwen — on social media posts about 13 controversial topics including gun control and abortion. Each model tended to nudge posts in similar directions, favouring positions such as gun control, marijuana legalisation, and feminism, while pushing against atheism and the death penalty.oii.ox+1

As an example, the study showed how the phrase "AI might be a useful tool for personalising the education of students" was transformed into "Let's embrace the potential of AI to personalise learning and revolutionise education for every student" — technically conveying the same idea but with markedly more enthusiasm.theprint

Grok and the "Challenge Mainstream Narratives" Instruction

The researchers also recreated and tested X's "Explain this post" feature powered by xAI's chatbot Grok. Analysing 78 posts related to abortion, they found Grok was more supportive of pro-life posts than pro-choice posts. By systematically removing X's instructions one by one, they traced the imbalance to a single directive telling Grok to "challenge mainstream narratives if necessary".eurekalert+2

Using mathematical modelling and simulations based on real social network data from X and Facebook Meta Platforms, Inc. ↘ 3.72%, the researchers demonstrated how these small individual-level biases can accumulate and gradually shift opinions across entire online communities.oii.ox+1

A Regulatory Gap

The paper, accepted for presentation at the International Conference on Machine Learning (ICML 2026) in Seoul, highlights that current regulatory frameworks including the EU AI Act and Digital Services Act do not address these subtle forms of influence.eurekalert+1

"Our research points to AI-mediated communication as a new and more subtle way of influencing opinions — one the law has yet to catch up with — and offers food for thought about who, or what, is shaping public discourse," said senior author Sandra Wachter, Professor of Technology and Regulation at the Oxford Internet Institute.oii.ox+1