OpenAI Lets ChatGPT Work Build Writing Profiles From Connected Apps

OpenAI has added a ChatGPT Work feature that studies writing in connected workplace accounts and uses it to shape generated messages. The change turns Gmail, Google Drive, Slack, and SharePoint from places ChatGPT can consult for information into inputs for a personal writing profile.

The profile is designed to pick up details that often make an AI draft feel unlike its sender: preferred phrases, capitalization, formatting, and sign-offs. Once set up, it applies to messages generated in ChatGPT Work on web and mobile. Setup is available on the web for customers on all paid ChatGPT plans, through the setup page or Settings, then Personalization and Writing style.

A profile, not a reusable prompt

The product change draws a useful line between two roles for workplace connections. A connected app can bring project material into a conversation so ChatGPT can find a file, summarize a document, or look up a message. Writing-style personalization gives those connections a second role: shaping the voice of text ChatGPT Work generates.

Access limits remain the product boundary

The profile does not create a new route into company material. Content available through a Gmail, Drive, Slack, or SharePoint connection remains limited by the linked account’s permissions and any workplace restrictions. In managed workspaces, administrators can control app availability and may disable apps for users altogether.

What determines whether it can work

• The relevant app must be available for the user’s plan, region, role, model, workspace, and ChatGPT interface.
• The connection can access only material the linked account is authorized to reach; app permissions do not grant new access.
• Workspace controls can restrict access even when an individual user has an account connection.

Personalization is distinct from model training

OpenAI distinguishes adapting ChatGPT Work to a person’s style from training a new general model on that person’s inbox or workplace documents. Its Google app data controls say information accessed directly from connected Google apps is not used to train generalized models, except in specified cases such as user feedback or content manually copied into a conversation.

That distinction does not make the feature merely cosmetic. A work assistant that has the right facts but an unfamiliar voice still creates editing work before a message can be sent. OpenAI’s bet is that a profile built from the writing people already produce will make drafts more usable without asking them to explain their style in every prompt.