ChatGPT dots prompts: 10 copy-ready tasks to give your dot

A ChatGPT dot is only as useful as the job you give it. Because it can keep working after you close the chat, a vague prompt has more time to go wrong. This guide gives you ten prompts to copy, plus the six habits that make them work.

If you haven't set up a dot yet, start with our guide to what dots are and who can get one. Dots are currently available on Pro and Business Premium plans, and on Enterprise if an admin turns them on.

6 habits for writing dot prompts

These come from OpenAI's dots documentation and a practical guide by The Rundown.

1. Name the job in one sentence. "Keep the offsite on track" beats "help with the offsite."
2. Name the exact sources. Say which file, channel, calendar or inbox label to read. Don't say "check my stuff."
3. Put the limits in the prompt. "Draft replies but don't send them" is the single most useful line. Also set the same limit in Settings, Personalization, Custom rules. OpenAI's guidance is that app permissions control when ChatGPT asks, but they don't give an app access it lacks.
4. Ask for source labels. A brief that says where each fact came from is easy to check. One that doesn't can invent an owner or a decision.
5. Say when to ask you. "Message me if a deadline is at risk or you need a decision" keeps it from guessing.
6. Run it once before you schedule it. Read the result, fix the prompt, then turn it into a schedule. The Rundown's guide recommends this order too.

10 prompts to copy

Fill in the capitalised brackets. Use a time zone name like Asia/Kolkata when you give a time.

1. The morning brief

Prepare one morning brief using only these sources:
Project notes: [EXACT FILE OR LINK]
Calendar: [EXACT ACCOUNT AND CALENDAR]
Time zone: [YOUR TIME ZONE]

Look ahead seven days for deadlines. Return:
1. Today's meetings and what to prepare for each.
2. A table of tasks with Task, Owner, Due date, Status and Source.
3. Up to three suggested priorities, with a reason for each.
4. Anything missing or unclear that needs a person.

Do not invent owners or decisions. If something isn't in the sources, write "not stated". Don't send or change anything.

Once you're happy with the result, follow up with: "Create a weekday morning brief at 8:30 a.m. [YOUR TIME ZONE]. Use the same sources and output rules. Read fresh data every run."

2. Keep an event on track

This example comes from OpenAI's own documentation. Swap in your own event:

Help me keep our offsite on track. Read the plan and venue emails I shared, and make a list of decisions, deadlines, and things we're waiting on. Focus on anything that affects the budget or needs an answer this week. Draft suggested replies for my review, but don't send them.

3. A weekday channel check-in

Also based on OpenAI's documentation:

Check the connected planning channel each weekday at 9 AM [YOUR TIME ZONE] for the next four weeks. Update our offsite list when something changes. Message me in ChatGPT if a deadline is at risk or you need a decision from me.

4. Inbox triage, drafts only

Every weekday at [TIME] [YOUR TIME ZONE], look at unread emails in [LABEL OR FOLDER] from the last 24 hours.
Sort them into: needs a reply today, needs a reply this week, and no reply needed.
For each "needs a reply today", write a short draft in my voice and save it as a draft. Never send an email.
Send me one summary in ChatGPT with each email's sender, subject and your one-line reason for the group.
If an email mentions money, a contract or a legal issue, don't draft a reply. Just flag it.

Our inbox triage prompt is a good version to try in a normal chat first.

5. Weekly project review

Every Friday at [TIME] [YOUR TIME ZONE], review [TASK LIST OR PROJECT NOTES LINK].
Tell me: which deadlines changed, which tasks have no owner, and which tasks say "done" without any evidence in the notes.
Show the source for each item. Don't edit the list. Put suggested edits in a separate section for me to approve.

6. Meeting preparation

Prepare a brief for my meeting with [PERSON OR COMPANY] on [DATE AND TIME].
Use only: the calendar invite, the last three email threads with them, and [NOTES FILE].
Return: what we agreed last time, what's still open, three questions to ask, and anything I should double-check before I go.
Label every fact with its source. If you're not sure the emails are about the same client, say so instead of guessing.

After the meeting, our meeting follow-up email prompt can turn your notes into the message.

7. Research watch

Every Monday at [TIME] [YOUR TIME ZONE], look for news on this question only: [YOUR NARROW QUESTION].
Use only these sources: [LIST 3 TO 5 SITES OR FEEDS].
Report only what is new since your last report. For each item give the date, the source link and one sentence on why it matters.
If nothing has changed, say "no changes". Don't fill space.

Use our research question refiner if your question is too broad.

8. Turn an interview into content

Read the interview transcript I've attached. Prepare three things for my review:
1. A 5-bullet summary in plain words.
2. Five short quotes, copied exactly, with the timestamp or line where each appears.
3. Two draft social posts of under 60 words each.
Don't publish anything. Put the word CHECK next to anything you're unsure of.

9. Track feedback

Every weekday at [TIME], read new messages in [FEEDBACK CHANNEL OR FORM].
Add each new item to [TRACKER LINK] with: date, what happened, how many people mentioned it, and a severity of low, medium or high.
If something looks like a bug, write a short note describing how to reproduce it. Don't change any code and don't reply to customers.
Message me only when an item is high severity.

10. A student's weekly plan

Every Sunday at 6 p.m. [YOUR TIME ZONE], read my exam and assignment dates from [CALENDAR OR SPREADSHEET].
Build a plan for the next seven days: what to study each day, for how long, and which deadline it supports.
Keep study blocks to 45 minutes, with a 10-minute break between them, and leave [FREE DAY] free.
Put the plan in a message to me. Don't add anything to my calendar until I say yes.

Students should check the plan requirement first: dots aren't on Free, Go or Plus. The same prompt also works as a normal chat if you paste the dates in.

Rules worth adding in Custom rules

In Settings, Personalization, Custom rules, add a rule for each action you never want without approval. Sensible starters:

• Sending or replying to any email or chat message.
• Publishing or posting anything.
• Buying anything or entering payment details.
• Deleting or overwriting files.
• Sharing a file or changing who can see it.

Choose "Ask before taking action" for each one. You can loosen a rule later once you trust the dot with that task.

Mistakes to avoid

• Connecting everything on day one. Add one app at a time.
• Skipping the test run. A schedule multiplies any mistake in the prompt.
• Trusting an "all clear." A missing update isn't proof that nothing went wrong. Ask for "no changes" to be stated.
• Carrying old deadlines forward. When a date changes, tell the dot to replace the old one.
• Putting private information in a channel you don't control. Check your organisation's data policy first.

Where to go from here

You can adapt any of these ten as an ordinary ChatGPT prompt. The ChatGPT prompts page has hundreds more, including the delegate a task prompt and the weekly review prompt. The prompt builder can help you write your own, and our GPT-6 Sol and Luna guide covers the models behind ChatGPT.

Quick checklist

• One job, one sentence.
• Exact sources, exact time zone.
• "Draft, don't send" in the prompt and in Custom rules.
• Source labels on every fact.
• Tell it when to ask you.
• Run once, check, then schedule.

Your dot can make mistakes. Read what it produces before you act on it, and never rely on it alone for anything involving money, health or legal matters.

Prompts to try

Browse all 1594 prompts →