33% of US user sample used profanity toward chatbots… Anthropic policy implemented

A Searchable analysis of more than 8 million messages sent by consenting U.S. panel users to ChatGPT, Gemini, and Claude between September 2025 and September 2026 found that 33% of the sample directed profanity or insulting language at one of the three chatbots at least once. The study counted roughly 19.1 swear, insult, or derogatory terms per 10,000 words entered across the messages, with the 33% figure covering all three chatbots rather than Claude users alone. The finding matters because it quantifies how routinely consumers vent at AI assistants, a signal for providers weighing tone, safety filters, and retention. Anthropic's policy rollout tied to the release is the next item to watch, along with whether other model providers publish comparable usage data.