Using ChatGPT as a personal assistant for client work
Start with the job you want off your desk: a useful brief before the client call, with the latest decision and the question still waiting for an answer. Give that job a repeatable shape before adding more tools.
A good first setup: choose one client, supply a small set of dated material, describe the brief you want, and check its answer against those sources. Keep the first job as preparation you review yourself. Once it works, decide whether it belongs in a recurring task.
Give each part of the setup a purpose
Scroll sideways to compare all columns.
| Part | Use it for | Your decision |
|---|---|---|
| Chat | One request and its follow-up questions | What should be ready when this conversation ends? |
| Project | Related conversations, files and instructions | Which engagement belongs together? |
| Memory | Information reused across conversations, under your settings | Which preferences should persist? |
| Connected apps | Access to supported external sources | Which access does this particular job need? |
| Scheduled task | Work you want to run again | When should it run, and what can that task actually read? |
These are separate choices. For current behavior, see OpenAI's Projects guide and Memory FAQ, checked 21 October 2026. Account and workspace settings affect availability.
A client brief you can try with fictional material
Imagine you advise Harbor Studio on its service launch. Your next conversation needs a decision about the pilot, not another broad summary of the engagement. Use these invented inputs to try the method without uploading client information:
- Monday call note: the operations lead asked for two pilot options. You promised a comparison by Thursday. No pilot date was agreed.
- Tuesday email: the client wants the smaller option first and asks whether it changes the training plan.
- Wednesday working note: you have drafted the comparison. The training question still needs an answer from the delivery lead.
Keep the labels and dates when supplying the text. Then ask:
Prepare my Harbor Studio call brief from these three sources. Give me the latest client request, my outstanding commitment, the decision needed on the call and the questions still open. Attach a source label to each factual statement. Separate a proposal from an agreement. If a date or owner is missing, say so.
The answer to look for: the smaller pilot is the client's latest preference; the comparison is due Thursday; the training impact is unresolved; the pilot date is unconfirmed. That is an expected reading of the fictional inputs, not a transcript of a product run.
An answer that says “the pilot launches Thursday” has merged two unrelated facts. An answer that says “training is unaffected” has filled a gap. Those are useful failure checks because a fluent paragraph can hide both mistakes.
Make the setup reusable in a Project
In ChatGPT, select New project in the sidebar and name the engagement. Add your selected reference material. Open the project's more-options menu and choose Project settings for its instructions. OpenAI documents this path in the Projects guide linked above.
Useful standing instructions are short: distinguish requests from agreements; use the dates in the material; identify conflicting statements; ask before assuming an owner; put unresolved questions at the end. Keep today's meeting-specific request in the chat. Otherwise yesterday's instruction to prepare a launch brief can quietly become the instruction for every future question.
Add recurring work only after the output is useful
“Help me every morning” is too open-ended to judge. “At 8:15 on weekdays, remind me to review today's client commitments” has an observable outcome. A briefing needs an additional decision: where will its fresh information come from?
The scheduled-tasks guide covers that setup and the current project-file restriction. Do not assume that a successful interactive Project conversation proves a scheduled run has identical access. Check one completed run before making it part of your morning.
When a personal assistant fed selected client material fits better
If you enjoy maintaining client Projects and reviewing their drafts, the setup above may be enough. Equerry is another choice when you want to send selected correspondence and notes to a personal assistant, then ask about them from your iPhone or iPad before the next conversation.
| What you send | Forward the Harbor Studio email and save your dated call note in Equerry. Later replies arrive if the Equerry address stays on the thread. |
|---|---|
| What you ask | “What did I promise Harbor Studio, and what should I clarify about the pilot?” |
| What you get back | An answer from the material you shared. When it draws on a saved note, its source chip lets you inspect the passage and open that note. |
You can also save a Meeting Prep Routine and allow calendar access for a brief before upcoming meetings. Equerry does not send the client reply or change your calendar. The useful first test is whether the answer recovers the agreement you would otherwise have to search for.
Frequently asked questions
Can I use ChatGPT as a personal assistant for work?
Yes. Start with one repeatable job, such as preparing a client update from material you supply. Decide what a useful answer must contain, which sources it may use and what you will check before acting on it.
Should I use a Project or a normal chat?
A Project keeps related chats, files and instructions together. A normal chat is a simpler starting point for a single request. If the same client comes up week after week, an ongoing place for the work is usually worth it; for a one-off question, a normal chat is enough.
Does ChatGPT remember information between conversations?
ChatGPT has memory controls, and Projects can reuse their context. Check the settings available to your account. A remembered preference and a dated client decision serve different purposes: keep the decision's source available for checking.
Do I have to connect my whole mailbox?
You can start with selected text or files. Connections are another way to supply information, not a prerequisite for the example here. Review the scope requested by each connection before approving it.
Last updated 2026-10-21.
All product names, logos, and brands are property of their respective owners. Equerry is not affiliated with, endorsed by, or sponsored by any other product mentioned; comparisons are provided for informational purposes.