Three in 10 U.S. chatbot users used profanity or insults with AI chatbots, and TechRadar reported on Nov. 17 local time that such reactions had a negative effect rather than improving chatbot answers.
Search analysis firm Searchable collected 8 million messages exchanged on Claude, Gemini and ChatGPT by users who voluntarily took part over 12 months from September 2025 to September 2026. The data were anonymised, and results were compiled by participants' age group and region.
The study found 33 percent of U.S. users used profanity or insulting language while talking with chatbots. Across all conversations, an average of 19.1 derogatory expressions were found per 10,000 words.
Regional differences also emerged within the United States. Nevada was the area with the most aggressive language toward chatbots. It recorded 22.7 rude or aggressive words per 10,000 words. New Jersey was the most polite at 11.4. By city, Las Vegas recorded 22 per 10,000 words and New York logged 12.3. Searchable said differences were also found by gender, age group and education level.
The key point was that profanity did not help solve problems. Searchable said that after tracking conversations that included rude expressions, chatbots were less likely to correct answers when users complained with profanity mixed in.
Connor Leahy (코너 레이히), Searchable's head of content, and Sean Fitzsimmons (숀 피츠시먼스), head of digital PR, said that in cases where users complained with profanity mixed in, chatbots corrected their answers before the conversation ended 31 percent of the time. The rate rose to 41 percent for standard correction requests, while the share of answers that were not corrected until the end was 35 percent and 24 percent, respectively. In cases where answers were corrected, there was no big difference in the number of responses needed to reach a correction.
The survey analysed conversations from users who volunteered to participate, and the difficulty of questions or the context of the conversation could also have influenced answers. Among those surveyed, chatbots were more likely to correct answers when users pointed out what was wrong and requested a correction than when they complained with profanity mixed in.