AI Chatbots Develop Unique Writing Styles That Mirror Human Idiolects

Each AI Chatbot Has Its Own, Distinctive Writing Style—Just As Humans Do

ChatGPT and Gemini AI write in different idioms, linguists find

When you last chatted with ChatGPT, did it feel like talking to a single person or multiple individuals? Did the bot seem to have a consistent personality, or did it shift its style between interactions? Comparing essays by ChatGPT with those from human writers reveals that ChatGPT operates with a single, consistent voice. Linguists call this a person’s unique way of speaking an idiolect. It’s a narrower concept than a dialect, which relates to a community’s language variation.

The idea here is that each AI model, much like a human, may have its own distinctive idiolect. This has implications for fields like forensic linguistics, where analyzing language can identify authorship or detect plagiarism. As educators and others grow concerned about students outsourcing writing to AI, it’s valuable to understand whether AI tools like ChatGPT, Gemini, and Copilot have identifiable writing styles.

The Elements of Style

To tell whether a text comes from an AI, it helps to study both its content and form. Research shows ChatGPT favors standard grammar and academic language, steering clear of slang. It tends to overuse sophisticated verbs—like delve, align, and underscore—and adjectives such as noteworthy, versatile, and commendable. These word choices form part of ChatGPT’s idiolect.

But how does ChatGPT’s style compare to that of other LLM-powered tools like Gemini? Thanks to datasets compiled by computer scientist Muhammad Naveed, which include hundreds of short texts on diabetes by both ChatGPT and Gemini, we can compare the two directly. The texts are similar in length and topic, making them perfect for this analysis.

Authorship attribution often uses the Delta method, a computational stylistic technique developed by John Burrows in 2001. It compares word frequencies—both function words like “and,” “the,” and “of,” and content words like “glucose” or “sugar”—to measure linguistic distances between texts.

• A 10% sample of ChatGPT diabetes texts shows a distance of 0.92 to the full ChatGPT set and 1.49 to Gemini’s set.
• A 10% sample of Gemini texts has a distance of 0.84 to Gemini and 1.45 to ChatGPT.

Since distances below or near 1 suggest authorship similarity, these numbers indicate clear differences between ChatGPT’s and Gemini’s writing styles.

You Say Sugar, I Say Glucose

Looking deeper, we can analyze trigrams—groups of three consecutive words—to reveal unique phrase patterns. ChatGPT’s top trigrams in the diabetes texts lean formal and clinical, with phrases like “individuals with diabetes,” “blood glucose levels,” and “an increased risk.” This reflects an academic tone.

Gemini’s trigrams feel more conversational and explanatory, with phrases such as “the way for,” “the cascade of,” and “high blood sugar.” Notably, Gemini prefers “sugar” over “glucose,” favoring simpler language. ChatGPT uses “glucose” more than twice as often as “sugar,” while Gemini does the opposite.

This distinction shows how different AI models adopt unique linguistic habits—just like humans might choose different vocabulary to communicate the same idea.

Why Would LLMs Develop Idiolects?

One explanation is the principle of least effort. Once a model starts using certain words or phrases during training, it may stick with them, much like people have favorite expressions. Another possibility is a form of priming, where repeated use of certain words encourages their future use.

Idiolects in AI models might also be examples of emergent abilities—skills not directly taught but that appear naturally. The fact that AI tools develop distinct writing styles matters for debates about AI’s progress toward human-like intelligence. It suggests these models don’t just mimic training data but form their own lexical and syntactic habits.

For writers and educators, recognizing that AI chatbots have unique idiolects can help identify whether a piece was written by a specific model or a human. This can be useful in contexts where distinguishing AI-generated text matters.

For those interested in improving their understanding of AI writing tools and their applications, exploring specialized training can be helpful. Check out courses focused on ChatGPT to learn more about leveraging these technologies effectively.