Does swearing at AI chatbots get you better answers? - The News International
Swearing at AI chatbots is common, and it doesn’t pay off. One in three American chatbot users (33%) has spoken at or insulted one, according to a new analysis from the AI visibility platform Searchable. The chatbot’s next reply fixed under a quarter of complaints containing swearing or insults (24%), compared with 35% of plain-language corrections.
The analysis covers more than 8 million anonymous messages from US users between September 2025 and September 2026. Searchable counted swearing, insults and aggressive language per 10,000 words and ranked all 50 states and 80 metro areas.
Over a third of complaints containing swearing or insults (35%) were still unresolved when the chat ended, compared with 24% of plain corrections.
However, of the sweary complaints, 73% came after the user had already complained or sworn earlier in the chat, compared with 43% of plain complaints. Swearing tends to be a sign of a conversation that has already gone wrong.
The most common trigger is the chatbot not doing what it was asked (19%), followed by a wrong answer (18%) and a misunderstood request (15%).
Searchable CEO and co-founder Chris Donnelly explains: “AI chatbots’ capabilities have come a long way in a short space of time, and their users’ expectations have grown with them. You can ask one about almost any topic, and it will search the web and read hundreds of pages to answer, even when the answer it gives isn’t correct.”
He continued: “It’s easy to assume a quick response is an accurate one, but that isn’t always the case, and a lot of the frustration in our data comes from expectations not being met. Just over half of the mean messages came straight after a chatbot ignored an instruction, got something wrong or misunderstood the request.”
Men use a third more swear words and insults than women (22.0 against 16.5 per 10,000 words). Users aged 35 to 44 use over one and a half times as many as 18- to 24-year-olds (21.7 against 13.3).
Geographically, Nevada is the meanest state, with 22.7 swear words, insults and slurs per 10,000 words, twice the rate of New Jersey (11.4), the politest. Among cities, Las Vegas is the meanest at 22.0, while New York is the most polite at 12.3.
Hannah Fry previously compared AI chatbots to an improvisational actor that can play any character. “If you prompt it in a particular way, it will respond in kind,” she says, so “it’s probably better to be polite.”
Lab research is mixed, but some of it supports that advice. In a 2024 Waseda University and RIKEN study, very rude prompts dragged GPT-3.5’s score on a standard knowledge test down from 60.0 to 51.9. Penn State researchers found in 2026 that plain prompts did best on GPT-5-nano (82% against 71% for threatening messages).
Donnelly concluded his remarks by adding that “Hannah Fry’s improv actor comparison is a useful way to think about it. A chatbot works with what you give it, so telling it exactly what went wrong gives it something to fix, while an insult gives it nothing constructive to work with.”
