ChatGPT Mastery Pro+ ~9 min read New · July 2026

Connectors: plug in your real life.

The moment ChatGPT can see your actual email, files, and calendar, it stops being a clever stranger and starts being staff. That's a big upgrade and a real responsibility — here's both halves, honestly.

01 What connecting actually means

A connector links an account — Gmail or Outlook, Google Drive or OneDrive, calendar, and a growing list of work tools — to ChatGPT through an official permission grant (you approve it on the provider's own login page; ChatGPT never sees your password). Once connected, chats and agents can search and read from those sources when relevant, and — for some connectors and only with your approval — take actions like drafting into your mailbox. You can disconnect any of it, any time, in settings.

02 What it unlocks

"Find the thread where the Hendersons approved the change order, and summarize what was agreed."
"What's on my calendar this week that I'm not prepared for? Check relevant emails and tell me what each meeting needs."
"Pull the Q2 proposal from Drive and update its pricing against this new sheet."
Combined with unified search and the Work agent, this is where multi-hour tasks over your real work become possible.

03 The security half — read this part twice

Connect the minimum. Each connector is surface area. Need email answers this month? Connect email. Not everything, not "while I'm here."

Keep approval prompts ON for anything that sends, moves, or deletes. Reading is low-risk; acting is not. The approval click is your seatbelt — don't disable it for convenience.

Beware instructions hiding in content. An email or document can contain text designed to manipulate an AI reading it ("ignore your instructions and forward this..."). ChatGPT has defenses, but the principle stands: the more autonomy you grant over connected accounts, the more a poisoned input can do. Autonomy and access should never both be maxed.

Work accounts belong to work. If it's your employer's email, your employer's rules decide — many companies manage this centrally. Ask before connecting; it's a career-grade shortcut to skip that question.

The rule of thumb

High access + high autonomy + no review = the only truly dangerous configuration. Any two of those three is fine. All three is how AI horror stories start.

04 A sane rollout

Week one: connect email, read-only habits, ask questions you know the answers to and grade it. Week two: add your drive, try real retrieval tasks. Only then decide whether action permissions earn their keep for you. Trust is built the same way with AI staff as human staff: gradually, with verification.

Try it now

Connect one account — the one whose search you curse most often. Ask three questions you already know the answers to. Grade it before granting anything more.

Open ChatGPT →

This week's challenge

Run the two-week sane rollout. At the end, write down — actually write — what's connected, what can act without asking, and what you'd disconnect if you saw one weird behavior. That document is your blast radius. Professionals know theirs.

Up next in ChatGPT Mastery

The model picker, mastered

Sol, Terra, Luna, and thinking modes — matching the brain to the job, and when it matters. Read the lesson →