Most people use ChatGPT like a search engine. They type a question, get an answer, close the tab. That’s leaving most of its value on the table.
The real leverage comes from setting it up once, deeply, so it knows your work, your voice, and your context — and can act on all three.
Here’s how to build that kind of assistant.
Start With a Dedicated Project, Not a Random Chat
Don’t just open a new conversation every time. Create a persistent project or workspace in ChatGPT and treat it as your assistant’s permanent home. Every conversation inside that project inherits the context you’ve built into it — your connected tools, your instructions, your history.
This matters because a one-off chat has no memory of you. A well-configured project does.
Connect It to the Tools You Actually Use
The assistant is only as useful as what it can see. Connect it to:
- Email — so it knows what’s pending, who’s waiting on you, and what tone you use
- Calendar — so it understands your real constraints and upcoming commitments
- Task manager — so it can pull your to-do list without you re-explaining it every time
- Cloud storage — so it can reference documents, notes, and past work you care about
- Slack or team chat — so it understands the ongoing conversations in your work
Set these integrations to read-only at first. You want the assistant to be deeply informed before you give it any write access. Trust builds incrementally.
Train It on Your Voice
Before you ask it to draft anything on your behalf, give it this prompt:
Analyze the emails and messages I’ve written over the past 12 months. Exclude anything I forwarded or copied. Build a voice profile that captures my tone, sentence length, how I open and close messages, and any phrases I use often.
Once it has that profile, every draft it produces will actually sound like you — not like a generic AI press release. You’ll still edit, but you’re starting from 80% done instead of zero.
Give It a Standing Context Prompt
At the start of your assistant project, write a setup prompt that tells it what it has access to and how to behave. Something like:
Every time I ask you something in this project, you have read-only access to my Gmail, Google Calendar, Notion workspace, and task list. Use that context proactively. Don’t ask me to re-explain things you can look up. Default to being brief unless I ask for detail.
This single prompt changes the quality of every interaction that follows. You stop being the one who has to remember and re-explain everything.
Practical Things It Can Do Right Now
Once the foundation is in place, the assistant can handle tasks that used to eat real time:
Weekly planning. Ask it: What’s on my calendar and in my inbox that I need to handle this week? What can you take off my plate? It’ll surface deadlines, flag things you’ve been ignoring, and often point out that it already drafted something you asked for earlier.
Meeting prep. Before a client call or a presentation, ask it to pull together a prep doc using everything it knows — past emails with that person, relevant files, notes from previous meetings. A 10-second prompt produces a document that would have taken you 30 minutes to build manually.
Drafting outreach. If you need to reach out to 15 potential partners, give it a clear brief and ask it to write personalized drafts for each one, saved to your drafts folder. You review and send. The research and the writing happen without you.
Summarizing scattered context. Working on a project that’s spread across a dozen threads and documents? Point the assistant at those sources and ask it to synthesize a one-pager. It doesn’t need you to paste anything in — it can read the sources directly.
The Prompt That Sounds Simple but Isn’t
Once your assistant knows your work well, one of the most useful prompts you can give it is also the shortest:
What do I need to hear right now?
Because it has your emails, your schedule, your task list, and your recent conversations, it has enough signal to tell you what you’re avoiding, what’s overdue, or where you’re overcomplicating something. It’s not therapy — but it’s surprisingly good at cutting through noise when you’re overwhelmed.
One Thing to Get Right Before You Start
Don’t connect everything and immediately give the assistant permission to send emails or post on your behalf. The read-only phase is where you build trust. You watch it draft things. You check whether its judgment matches yours. You correct the voice profile when it slips into corporate-speak.
After a week or two of that, you’ll know exactly how much autonomy to give it — and where you still want to stay in the loop.
The setup takes maybe an hour total. After that, the assistant earns that hour back within a few days. That’s the trade worth making.
