AINews wire, 8 h ago
Rude prompts get worse results from AI chatbots, Searchable study finds
Reported by TechRadar

A study by research firm Searchable, reported by TechRadar, finds that a third of US chatbot users swear at or insult their AI tools. The analysis covered roughly eight million chat messages across Claude, Gemini and ChatGPT, collected between September 2025 and September 2026. Anthropic has responded with updated rules for Claude that target cruelty in conversations.
The messages were scored for rude or offensive words per 10,000 words, grouped by age and location. The average across the sample was 19.1. Nevada ranked worst at 22.7, while New Jersey was the most polite state at 11.4. Among cities, Las Vegas scored 22 and New York 12.3, the lowest of the cities named. The report also pointed to differences between men and women, age groups and education levels.
Searchable also followed each exchange to its end. Of complaints that contained swearing, 31% were resolved, against 41% of plain corrections. Another 35% of rude complaints stayed unfixed, versus 24% of plain ones. Where a fix arrived, rude and polite users needed a similar number of replies.
The report suggests that a plain, factual correction works better than an insult. TechRadar reported the findings on 17 November 2026.
For readers, the finding matters if they use chatbots for research, planning or help with smart home devices. A calm, specific follow-up may save time when an answer is wrong. The article does not say what Anthropic's new rules cover in detail, how they will be enforced, or whether other chatbot makers will follow. It also gives no word list or method for the scoring, so the exact figures should be read as the researchers' own measure. Readers who want the wider context can compare it with other recent chatbot coverage on our site.
Enjoying this? Get the AI brief
Every Thursday: AI that matters for your home.
✓ Free · seven editions · one click to leave
Original reporting by the sources above. Summary by smart-home.space.