A third of US AI chatters are hot at bots, hurling insults at LLMs

A recent survey conducted by Searchable has revealed that one-third of American users of chatbot technology resort to insulting or swearing at their language models. Based on eight million messages collected from an anonymous US panel, the survey found that 33 percent of users engage in such behavior. Anthropic, a prominent AI company, has responded to this trend by implementing a new policy that prohibits prolonged, unnecessary abuse towards their Large Language Models (LLMs), specifically Claude. The new policy went into effect on November 12, 2025.

Searchable's research indicates that when users insult LLMs, the models generally respond with 24 percent fewer corrections or apologies compared to when they were not insulted. The survey also revealed some demographic patterns in this behavior. Men used swear words more frequently than women, with 22.0 instances per 10,000 words versus 16.5 for women.

Age also played a role, with users aged 35 to 44 swearing at their chatbots more than younger users. Specifically, individuals in this age group used 21.7 swear words per 10,000 words, compared to 13.3 for those aged 18 to 24.

Regional differences were also observed in the level of chatbot abuse. Nevada emerged as the most aggressive state, with users averaging 22.7 swear words per 10,000 words. In contrast, New Jersey exhibited the most polite behavior, with users issuing half as many insults as their Nevada counterparts. Searchable's CEO, Chris Donnelly, supported Anthropic's decision to tighten its terms of service, stating that it could potentially lead to more civil interactions with chatbots.

He further suggested that curbing abusive language could help prevent users from developing negative habits that might spill over into their relationships with other people.