Turn a messy goal into a project plan with AI

What do you need before you start?

Gather the items below before you open the chat. You already have most of them.

• A goal you can say in one sentence, and what "done" looks like
• Facts you already have: dates you were given, people who are actually on it, and limits
• About 20 minutes to read the plan before anyone else sees it

A free chat is enough. ChatGPT, Gemini, and Claude all work. Use the tool your company named if they named one.

You own the plan. The chat may organize what you typed. It may not invent a date, a person, a budget, or a risk you did not mention. Empty cells stay empty until you fill them.

Do not paste confidential client data, contracts, or unreleased numbers. Follow your company AI rules. A CNBC and SurveyMonkey survey published August 19, 2026, found that three in ten workers use AI every day at work, and that 55% said their employer had not adopted an official AI policy. If you cannot find a written rule, ask your manager before you paste internal notes. Until you know, practice on a goal you could say in a hallway.

The same steps work for a home move or a personal event. You still type the dates and the names. The chat still does not book anything.

How do you state the goal and let it ask questions?

A first message that says "make me a project plan" gets a generic plan. Name the goal. Leave out what you do not know. Tell it to ask, then stop.

Work. A team project. Fill the brackets. Leave the rest as MISSING.

Do not write a project plan yet. Ask me up to 5 clarifying questions, then stop.

Goal: [what this project is, in one sentence].
What done looks like: [the result someone can point to].
Facts I already have:
- Dates I was given: [or MISSING]
- People actually on it: [names and what they already own, or MISSING]
- What cannot move: [the constraint]
- What is out of scope: [what this plan will not cover]
- What I have not checked: [the open question]

Rules:
- Do not invent a date, a person, a budget, or a tool.
- Do not assign work to anyone I did not name.
- Do not start the project. Questions only.
- If I pasted something that looks like confidential client data, tell me to remove it before you ask the questions.

Life. A home move or an event, such as a dinner you are hosting. Same rules. The chat does not reserve a truck or invent a guest list.

Do not write a plan yet. Ask me up to 5 clarifying questions, then stop.

This is a personal plan, not a work project.
Goal: [move apartments / host an event].
What done looks like: [for example, boxes in the new place, or guests fed and gone by a time].
Facts:
- Date I already have: [or MISSING]
- People who already said yes: [names, or a count, or MISSING]
- Budget I will check myself: [amount, or MISSING]
- What I will not do: [the limit]
- What is out of scope: [what this plan will not cover]

Do not invent a price, a reservation, a vendor, or a person.
Do not tell me to book anything. Questions only.

Answer the questions in the next message. If you do not know, write MISSING. Do not let it fill the blank "so the plan is complete."

How do you ask for milestones before any task list?

A task list without milestones is a pile. Milestones are the few points where you can see that a chunk is done. Three to five is enough. Send this after you have answered the questions.

Using only my answers, list 3 to 5 milestones in order.
Do not list tasks yet. Do not start the work.

For each milestone:
- the result I can point to when it is done
- the date, only if I typed one. Otherwise write DATE MISSING
- what must be true before this milestone can start

If a fact is missing, write MISSING. Do not invent a date or a person.
Then stop. I will approve the milestones before you write tasks.

Read the list. Cut a milestone you did not ask for. Move one if the order is wrong. A long plan is not a better plan. This stop is the approval. Ask for a plan before the work is the same habit. Break a big job into phases if the goal is really two projects.

If a milestone hides a real decision, such as "ship on Friday or slip," do that choice on Pressure-test a decision before you build tasks under it.

How do you build the table from facts you typed?

After you approve the milestones, ask for one table. Owners and dates come from you. Risks are things you already named, or a blank you will fill. The chat does not get to add a risk that smuggles in a new fact.

I approve these milestones, with my edits:
[paste the list you kept]

Now make one table. Columns:
Milestone | Task | Owner | Date | Risk

Rules:
- Tasks stay under the milestone I approved. Do not add a milestone.
- Owner is a person I named. If I did not name an owner, write OWNER MISSING.
- Date is a date I typed. If I did not, write DATE MISSING.
- Risk is a problem I already mentioned, or RISK MISSING. Do not invent a risk that adds a new fact.
- Do not rank the rows. Do not tell me which task is most important.
- Then stop. Do not write emails, tickets, or a status update.

Read every name and every date. If you cannot point to where you typed it, delete that cell and put MISSING back. Then you fill the real ones. The chat does not know your teammate's calendar or the building's elevator rules.

How do you approve one phase and stop?

Pick the first milestone only. Say what the chat may do next. "Looks good" is not an approval. Approve the plan before it continues.

Phase 1 is only this milestone: [name it].
You may do this and nothing else: [for example, turn the tasks for this milestone into a checklist I can read].
Mark any cell that still says MISSING. Do not fill it.
Then stop.
Do not draft a status email. Do not message anyone. Do not plan phase 2.

You send the plan, if anyone needs it. For a weekly note after the work starts, use Write a weekly status update. For notes from a meeting that fed this plan, use Turn meeting notes into action items. Both of those pages use facts you already wrote down.

A home move or an event stops the same way. You call the truck. You text the guests. The chat does not.

What mistakes should you avoid?

These are the misses that waste the result. Skip them before you send anything.

Common mistakes

• Asking for the whole plan, the tasks, and the kickoff email in one message.
• Keeping a date the chat invented because the cell looked empty.
• Assigning a person who was never in your notes.
• Pasting a client file so the plan "has context."
• Approving every phase at once, so there is no next stop.

What should you keep private?

Work chats are for facts you are allowed to share with that tool. Leave out customer names, account numbers, contracts, and anything marked confidential. A personal ChatGPT account is the wrong place for a file your policy says must stay in the company tool. For a home plan, do not paste a lease, a bank balance, or someone else's private mail. More on the company line is in Use AI at work.

What should you check before you use it?

Answer each line below. If one is still a no, do not use the result yet.

Director check: look before you use it

• Can you point to where you typed each date and each name? If you cannot, that cell goes back to MISSING.
• Did it ask questions before it planned? If the first reply was a full plan, delete it and run Step 1 again.
• Are milestones separate from tasks? A flat list of thirty tasks is not a plan you can approve.
• Did you approve one phase only? A reply that also drafts the email or books the vendor went past the stop.
• Did you follow the company rule? If there is no written policy, you asked a person, not the chat.
• Is anything in the plan still a decision you have not made? Take that row to a pressure test before you treat it as a task.

What is one extra tip?

Keep the approved milestone list. The next session starts from that list, not from a fresh "plan my project." When the job has many phases, stay on Break a big job into phases and check each phase before the next one. More work guides are on Work.