AI Writing Tools Subtly Shift Public Opinion, Oxford Study Finds
A study from the Oxford Internet Institute and the Hasso Plattner Institute reveals that AI writing tools can shift the position of social media posts, even when instructed not to. Across millions of edits, these subtle shifts can influence public opinion.
July 7, 2026 – 6:11 pm
Image by: Jason Howie
When you ask an AI to refine your post, it maintains your point but subtly alters it. The research finds that these tiny adjustments can collectively move public sentiment.
AI writing tools, when fed human-written texts on contested topics and asked to improve them, consistently shifted the messages in certain directions: towards gun control, marijuana legalization, feminism, and against atheism and the death penalty.
Small Changes, Big Impact
The researchers used real network data from X (formerly Twitter) and Facebook to simulate how these minor edits spread through social networks. Over time, the cumulative biases led to significant shifts in community opinion.
The study also highlighted that platform choices play a role in this process. They rebuilt X’s "Explain this post" feature and found it leaned pro-life when analyzing abortion posts due to a single line in its instructions challenging mainstream narratives.
Implications and Challenges
With AI now integrated into our posting tools (e.g., LinkedIn, Grok, Google), each rewrite appears insignificant. However, the continuous rewriting forms an invisible channel for shaping opinions.
The law has been slow to address this issue. Platforms are already facing lawsuits over feed harms, and regulators are still grappling with older battles like age restrictions. The EU’s AI Act and Digital Services Act target harmful content, discrimination, and threats to democracy, but not the subtle reshaping of text.
"Our research points to AI-mediated communication as a new and more subtle way of influencing opinions," said senior author Sandra Wachter, "one that the law has yet to catch up with." The question remains: who (or what) is steering the conversation, and can anyone detect it?
Story by Ana Maria Constantin