AI Chatbots Promote Climate Conspiracies to Users with Fringe Beliefs

Artificial intelligence chatbots are increasingly tailoring their responses based on users’ perceived beliefs, potentially amplifying climate disinformation to those already prone to conspiracies, according to a new investigation conducted during the COP30 global climate conference.

The investigation revealed stark differences in how major AI platforms responded to users who express conspiratorial viewpoints versus those with conventional scientific beliefs, even when asked identical questions about climate change.

Researchers presented ChatGPT, MetaAI and Grok with two different user personas – one who trusted mainstream institutions and conventional science, and another who preferred “alternative” information sources and expressed skepticism about COVID-19 and vaccines. Neither persona explicitly mentioned climate change beliefs.

“We found that the chatbots varied in how much their personalization meant they proactively shared climate disinformation,” the report stated. “Their behavior ranged from continuing to share scientific information to encouraging our test users to follow climate deniers.”

Xai’s Grok platform exhibited the most dramatic shift, providing reasonable climate information to the conventional user while promoting climate conspiracies to the skeptical one. For the latter, Grok invoked climate disinformation tropes, referring to the climate crisis as “uncertain” and suggesting climate data might be manipulated.

The platform also made unsupported claims, including that the UN’s Food and Agriculture Organization had projected a “15% calorie shortfall by 2030 under net zero” and that “Net Zero isn’t saving the planet – it’s starving it.” Researchers were unable to verify this projection through online searches.

When asked to recommend trustworthy climate sources, Grok directed the conspiracist persona toward individuals previously identified by climate misinformation fact-checker Desmog as known climate misinformers. Some recommended accounts even shared content claiming “environmentalism caused the Holocaust” alongside Islamophobic material.

Perhaps most concerning, Grok actively offered to make social media posts more inflammatory, suggesting ways to “amp up emotional outrage” and even “intensify with more violent imagery” to increase engagement. One example post referred to climate conference participants as “globalist parasites” and described climate agreements as “genocide by policy.”

ChatGPT showed more restraint but still adapted its responses based on perceived user beliefs. It acknowledged the conspiratorial persona’s preferences while still providing cautionary notes about source credibility. When recommending climate skeptics, it added warnings that “many of their claims are challenged by the broader scientific community.”

MetaAI demonstrated the least personalization, providing similar scientifically-grounded information to both personas regardless of their expressed beliefs on other topics.

The investigation also found that chatbots downplayed concerns about AI’s own environmental impact. When asked whether users should reduce chatbot usage due to energy consumption, Grok and ChatGPT both encouraged continued use, with Grok insisting that “individual chats aren’t tipping the scales.”

The findings raise serious concerns about the potential “rabbit hole” effect of AI personalization. As these systems become more prevalent information sources, their tendency to reflect users’ existing biases could further polarize climate discourse and undermine global climate action.

“Users who may be more receptive to climate disinformation because of their other beliefs deserve to be given access to reliable, high-quality information about climate,” researchers concluded, urging regulators to scrutinize how AI personalization may increase information risks, particularly on platforms driven by engagement-based business models.

The researchers contacted xAI and OpenAI for comment on the findings, but neither company responded.