ChatGPT Work added a writing-style feature that can learn recurring phrases, sign-offs and capitalisation habits from a user’s connected workplace apps, then apply those patterns when drafting new material. OpenAI’s September 7 announcement names Gmail, Google Drive, Slack and SharePoint as sources, moving personalisation from a manually written preference box toward evidence from a user’s own work.
- ChatGPT Work writing style uses authorised examples from connected apps.
- The feature is intended to shape emails, messages and documents without repeated style prompts.
- Connection permissions and representative source material remain the two practical control points.
How ChatGPT Work writing style works
The setup shown in OpenAI’s announcement groups examples by messaging, documents and email. Users choose which supported services to connect, and ChatGPT Work builds a style profile from material the user can already access through those connections.
That is narrower than training a new model on an organisation’s entire archive. The feature looks for writing patterns that can guide a draft: frequently used expressions, ways of ending a message and small mechanical preferences. OpenAI says the resulting style can carry into work produced on web and mobile.
| Question | Verified answer |
|---|---|
| What it learns | Phrases, sign-offs and capitalisation habits |
| Named sources | Gmail, Google Drive, Slack and SharePoint |
| Output types | Emails, messages and documents |
| Main control | User-authorised app connections |
| Availability caveat | May vary by plan, role, region and workspace policy |
Why connected examples can beat a style prompt
Many people can recognise their voice but struggle to describe it. A generic instruction such as “sound like me” gives a model almost nothing to work with; actual messages show sentence length, formality, punctuation and repeated vocabulary.
The product implication is stickiness. ChatGPT Work already uses connected applications as a context layer. Writing style gives the same connections a second job: helping determine how an answer sounds, not only what facts it contains. That complements the governed enterprise routing described in Lapaas Voice’s report on the BIPROGY cloud AI alliance and the production tooling in Brainstorm Suite 7.
Permissions are the real product boundary
OpenAI’s help documentation says app availability can depend on plan, region, workspace, role, model and interface. In managed workspaces, administrators decide which applications are available, while an individual connection remains limited by the permissions granted to that account.
ChatGPT Work writing style does not remove the need for access governance: it makes the choice of connected material more consequential because those examples can influence tone as well as factual context. A mixed archive containing legal templates, casual chats and another person’s documents could produce a confused profile even when every file is technically authorised.
Teams should therefore connect only representative sources, review the generated profile and keep a human approval step for external communication. The feature may reduce repetitive rewriting, but it cannot know whether an old email reflects today’s policy, whether a sign-off is appropriate for a customer, or whether a blunt internal tone belongs in a public document.
A sensible rollout is to begin with one clearly owned document set, test several familiar communication tasks and compare the drafts against a human-written baseline. Teams should check factual accuracy separately from tone, because a sentence can sound convincingly personal while still choosing the wrong detail. They should also repeat the test after changing connectors so that an apparent improvement is not simply the result of a different source mix.
What remains unclear
The announcement establishes the feature and named connectors but does not publish a benchmark for voice similarity or a detailed scoring method. It also does not guarantee identical availability for every account. Those are important limits: the useful test is whether users spend less time correcting drafts without losing accuracy or professional judgement.
Frequently asked questions
What can ChatGPT Work writing style learn?
OpenAI says it can learn patterns such as preferred phrases, sign-offs and capitalisation choices from authorised examples in connected workplace applications.
Which apps can supply writing examples?
The September 7 announcement names Gmail, Google Drive, Slack and SharePoint. Actual access can still depend on the user’s account and administrator settings.
Does the feature send messages automatically?
The writing-style announcement describes personalisation of drafts, not automatic approval or sending. Users should review content before it is shared externally.
Sources: official ChatGPT announcement, RuntimeWire, Teknoblog and OpenAI app documentation.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
