“It makes a difference if chatbots’ models don’t just average or mirror their training data but develop distinctive lexical, grammatical or syntactic habits in the process, much like humans are shaped by our experiences.”
– Karolina Rudnicka, edited by Madhusree Mukerjee
Scientific American describes how Google’s Gemini model writes more informally than OpenAI’s ChatGPT. They study “idiolects,” more granular than dialects, that focus on differences in the idioms generated in comparable contexts.
One example in the story points to ChatGPT using “blood glucose levels” compared to Gemini’s “high blood sugar” or “blood sugar levels.” They say Gemini also used “blood glucose levels” but only once, suggesting the model was familiar with the phrase and made a deliberate choice to favour the simpler versions.
Each AI Chatbot Has Its Own Distinctive Writing Style—Just as Humans Do | SCIENTIFIC AMERICAN | July 9, 2025 | by Karolina Rudnicka, edited by Madhusree Mukerjee