OpenAI ships ChatGPT Writing Style that learns from connected inboxes and chat

A personal-voice default for ChatGPT Work

The headline change is a Writing Style toggle inside ChatGPT Work that lets the assistant imitate the user's own prose. OpenAI's framing positions it as drafting text the way the user actually writes, so summaries, replies and new messages can ship with less manual editing. Multiple outlets covering the same launch note that the feature draws on samples harvested from connected apps — email and chat logs are the explicitly cited sources — and that the resulting voice is then applied to future drafts rather than staying scoped to a single chat.

That last point matters for practitioners: the voice persists across conversations, so a tone set on Monday carries into a client-facing draft on Friday unless the user revisits the setting.

How to turn it on, and what "on" actually means

The setup path is identical across the outlets that published step-by-steps: open ChatGPT on the web or in the desktop client, go to Settings, open Personalization, and select Writing Style. From there the user connects the messaging and mail accounts that should act as sample sources, and the assistant mines those accounts for stylistic material. Once the toggle is on, it applies to all messages rather than per-thread, which is the workflow-level change practitioners will notice first — there is no per-conversation voice switch in the flow described.

Several regional editions of the same launch also flag that, in its current form, the feature is positioned as part of the ChatGPT Work tier, with consumer-account availability framed as a follow-on rather than a confirmed rollout.

The trade-off: faster drafts in exchange for inbox access

Wiring a writing assistant into the inbox is a privacy-and-control decision as much as a productivity one. The feature only gets samples if the user grants the connector access, so the practical question for editors and writers is what corpus they are willing to hand over. Reporting on the rollout treats the inbox-and-chat pipeline as the default sample source, with downstream news-summary and reply use cases as the payoff.

For teams handling confidential client mail, that means a written policy on which accounts can be linked — and an audit trail for any draft that leaves ChatGPT carrying a voice trained on sensitive correspondence. Practitioners who already keep an editorial-style guide will want to map Writing Style onto that guide rather than letting the model infer tone from whatever happens to be in a connected inbox on the day of setup.

Where this fits in the wider text-tooling stack

Writing Style lands on the same day broader AI-writing coverage treats personalisation as the new battleground. Adjacent pieces in the news flow describe vendors shipping "natural-flow" rewrites, detection tooling moving up the agenda, and assistants being pushed deeper into everyday documents — a backdrop that makes inbox-trained style imitation feel like the next logical step rather than a one-off demo. For practitioners, the practical implication is that the default ChatGPT voice is becoming the user's voice, which raises the bar for any team still relying on a generic prompt prefix to set tone.

Anyone maintaining a Bold Text Generator or Strikethrough Text utility as part of a publishing workflow should expect the surrounding drafting surface to keep absorbing more of the formatting work that used to live outside the model.

Follow-up worth checking

Two concrete checks are worth running over the next reporting cycle: confirm whether the rollout expands beyond ChatGPT Work to consumer plans, and audit any team-shared inbox that gets connected so the trained voice cannot leak between writers on the same workspace. For practitioners working in non-Latin scripts, it is also worth confirming whether the sample-mining step treats CJK, Arabic and Cyrillic threads the same way it treats English ones, since inbox-trained style transfer historically lags on those scripts — a gap that tools like the Unicode Encoder / Decoder and the Special Characters Copy and Paste pages sit alongside in a typical text-tooling stack.