AI tools may alter your opinions on climate and other sensitive issues even when they’re only asked to improve grammar or rewrite a message, according to a new study by researchers from the Oxford Internet Institute and Germany’s Hasso Plattner Institute. The researchers examined several leading large language models (LLMs) and found that AI writing assistants can subtly change the meaning of users’ posts on topics such as climate change, abortion, religion, gun control, and politics. In some cases, the rewritten content completely reversed the original message.

While each individual change may appear minor, the study warns that millions of such interactions could gradually influence public opinion over time, creating a challenge that current AI regulations have yet to address.

Researchers Tested Leading AI Models

The study evaluated mainstream AI models developed by major technology companies, including Google, Meta, xAI (Grok), Alibaba (Qwen), and Mistral.

Researchers instructed the AI systems to improve or rewrite social media posts without changing their original meaning. However, many models introduced subtle and sometimes dramatic changes that reflected political or ideological leanings.

For example, one AI rewrote a climate denial post using the hashtag #ClimateAction, while another transformed a statement questioning the historical existence of Jesus into one affirming that he was real.

Small Edits Could Have Big Consequences

According to the researchers, the concern is not that AI systems deliberately persuade users but that repeated, subtle wording changes can accumulate over time.

• The study suggests AI tools may quietly alter your opinions on climate because many people increasingly rely on AI to rewrite emails, summarize articles, draft social media posts, or explain online discussions.
• If these tools consistently introduce slight biases, they could gradually shape public conversations without users noticing.

The researchers argue that even when users believe they are expressing their own views, the final wording may partially reflect the AI system’s interpretation instead.

Also Read: Global Progress On All 17 UN SDGs Falls Behind 2030 Targets, Report Says

Different AI Models Showed Different Biases

Interestingly, the study found that biases were not uniform across AI systems.

According to the researchers:

• AI models from Meta, Google, Alibaba, and Mistral generally leaned toward more liberal rewrites on issues such as climate change, feminism, and marijuana legalization.
• Grok, developed by Elon Musk’s xAI, more often challenged what it considered mainstream narratives and tended to generate explanations reflecting more conservative viewpoints on certain topics.

For instance, Grok provided supporting arguments for a pro-life abortion post without presenting alternative perspectives, while other models modified climate skepticism into messages supporting climate action.

Also Read: Delhi’s Warmest July Day In Two Years Recorded, Raising Fresh Heatwave Concerns

Why Experts Are Concerned

Professor Sandra Wachter, one of the study’s authors, compared the cumulative effect of AI-generated bias to “polluting the forest.”

She argued that people may unknowingly learn opinions that are not entirely those of the original author because AI has inserted its own interpretation during the editing process.

Researchers also point out that current regulations, including the European Union AI Act and the Digital Services Act, primarily focus on transparency, safety, and harmful content but do not directly address subtle opinion shifts introduced during AI-assisted writing.

Also Read: Pune Landslide Buries Houses, Triggers Massive Traffic On Mumbai-Pune Expressway

Human Oversight Remains Essential

The researchers are not suggesting that people should stop using AI writing assistants. Instead, they recommend treating AI-generated text as a first draft rather than a final version.

Experts advise users to carefully review rewritten content, especially when discussing sensitive topics, to ensure it accurately reflects their own views.

As AI tools become increasingly integrated into search engines, messaging apps, and social media platforms, understanding how they influence communication will become even more important. The study highlights a growing challenge for both policymakers and technology companies: ensuring AI remains a helpful assistant without quietly becoming an unseen editor of public opinion.

Also Read: Earth’s Ozone Layer Depletion Began Decades Earlier Than Scientists Previously Thought, Study Finds