One of the features of something written by ChatGPT (and other AI chatbots) is that it reads like it’s written by AI. A new feature in OpenAI’s app seems to be trying to change that. If it harvests different writing styles as training data in the process, all the better.
The feature, Writing Style, was spotted by a user on X. It shows OpenAI’s app looking for additional connections to user messaging, document, and email sources. The function allows ChatGPT to “write in your voice by referencing examples from your connected apps.”
Written by ChatGPT?
Is this new in ChatGPT? pic.twitter.com/DJfIWKjy4a
— Gael Breton (@GaelBreton) September 4, 2026
Not everybody will have access to Writing Styles, at least for the moment. According to Digital Trends, it’s still in testing. Only some will see the option to link their apps to their chatbot. A messaging app, your document storage (Dropbox or Google Drive), and an email account would give ChatGPT a very good overview of how anyone communicates digitally. When it rolls out, though, it’s worth checking whether OpenAI will opt users in to further train the company’s systems. If it rolls out.
Writing Style seems like a decent enough idea… at first. Pulling examples of a user’s writing style should allow responses, emails, and written content to sound like the person using ChatGPT. Whether these replications are pulled through as training data or not, the setup probably won’t last too long. Eventually, the repository being used as an example will populate with GPT-written text. At that point, it’ll stop sounding like a user and start a recursive loop that will degrade the quality of communications overall.
There are ways around this. Watermarking AI-written responses and locking the initial example sample to the date when it was first connected are options. But a document clear-out, a device swap, and other common digital chores could easily wipe these stopgaps out, leaving your AI sounding like… well, like an AI again. Still, that’s what testing is for.
