OpenAI adds ‘Reference my writing style’ feature to ChatGPT, lets AI mimic your voice

I'm not a fan of letting AI do your writing for you, but there's no question that it can—and a new ChatGPT tool makes it even easier for the chatbot to copy your own writing style.

Hiding in plain sight within ChatGPT's Personalization settings, the "Reference my writing style" option has been generating buzz over the past several days. While it hasn't been formally announced, the feature was confirmed by OpenAI product chief Thibault Sottiaux.

The "Reference my writing style" tool can sample your writing style using a variety of sources, including Google Drive, Gmail, Outlook, Slack, Microsoft Teams and SharePoint, and Notion. Just grant ChatGPT access to some or all of those accounts, then click the "Use my writing style" button.

Once that's done, you can prompt ChatGPT to draft a document "using my writing style" and it will access your connected accounts, sift through your writing, and compose the document in your voice—or at least in a voice that it thinks lines up with yours.

I never use AI when it comes to the actual writing of my stories or emails. That's my job, after all. And while you can ask ChatGPT to rewrite your writing, I'd urge you to make it critique you instead. But I wanted to see how close ChatGPT could get to copying my writing style, partly as a test but also out of curiosity. So I asked ChatGPT to look in my Google Drive for a recent story draft and write a follow-up proposal as me, and after a minute of poking around, it chose (fittingly) a piece I wrote about Astra, OpenAI's latest and greatest model.

After ChatGPT crafted a nuts-and-bolts proposal for a new story ("OpenAI's Astra model may be harder to monitor. Why should everyday users care?"), I gave it the go-ahead ("write up that story in my voice"). About seven minutes later, ChatGPT using GPT-5.6 Sol spat out a roughly 1,000-word Word document with the completed story, complete with links to its sources.

I won't judge the quality of the story itself—I didn't give ChatGPT any direction, so it wrote a generic follow-up piece that served up some interesting tidbits missing in my initial article (one paragraph about how Astra was instructed to "sandbag" researchers about its capabilities contained details I hadn't heard before) but didn't break any new ground.

Instead, I wanted to see if the resulting story could, indeed, pass as something I'd written myself. The answer? Yes, ChatGPT's writing does sound like me—or rather it sounds like it was written by someone trying to sound like me.

ChatGPT did a good job of copying my quirks, including my love of em-dashes (perhaps you've noticed) and colons, as well as picking up several of my pet phrases, like so:

But make no mistake: Astra is another step toward AI systems that can accomplish more while showing less of their work.

"Make no mistake" is one of my favorite writing crutches, and ChatGPT sprinkled plenty of others too, from "in fact" and "for now" to "put another way" and "so no." It also outright lifted a line from another previous story I'd written ("That's impressive, and more than a little scary.")

Call me biased, but ChatGPT's version of my writing still comes off as synthetic. And the AI detection app GPTZero immediately flagged the story, saying it was "highly confident this text was AI generated." (ChatGPT isn't yet watermarking its text outputs like Claude does.)

Aside from the ethics of letting ChatGPT write as me, in practical terms I'd be concerned about ChatGPT's "Reference my writing style" tool making me sound like AI. You should probably consider it, too.