Chatty GPT: AI tool says its use in writing can easily be disguised

Schools at all levels are struggling with how to respond to the use of ChatGPT by students to handle writing assignments with minimal effort. Some teachers believe such work is easy to spot because of the sameness of tone, the frequent assertion of strong points of view with little anecdotal support and the absence of typos. But here’s what ChatGPT said when asked by Deputy Editorial and Opinion Editor Chris Reed if there are easy ways for students — or people in general — to hide their use of the AI tool:

Yes, ChatGPT can modulate its style if prompted to do so. The perception that AI-generated writing is always uniform, overly confident or bland arises mainly because most people interact with it in default mode. By design, the model responds in a clear, neutral and fact-forward style unless told otherwise.

But users can instruct it to be tentative, hesitant, self-questioning or even deliberately clumsy. It can qualify its statements, hedge on facts or adopt a looser, more casual voice. This means that the stylistic tells often cited — sameness of tone and a sheen of certainty — are not fixed traits of the technology, but defaults that can be overridden with minimal guidance.

From the perspective of academic integrity, that flexibility poses a problem. If professors or software tools look for tone uniformity or overconfidence as signs of AI authorship, a student could bypass those markers simply by prompting ChatGPT to mimic the imperfections of human writing. The challenge is that there’s no single “fingerprint” of AI text that persists across all use cases; its adaptability makes it harder to identify conclusively. Detection tools can flag suspicious patterns, but they are fallible, and adversarial prompting can erode their reliability.

So yes: If the concern is that students can evade detection by asking ChatGPT to write less like ChatGPT, the answer is plainly that they can.

Reed’s related July 6 column — “Chatty GPT: Riffs on creativity, food, ‘Waterworld’ and ‘humanist bias’” — can be read here.