These 14 project management prompts help you plan, run and close a project with ChatGPT or Claude: a charter and scope, a work breakdown with estimates, a RACI, a risk register and pre-mortem, weekly status reports, change requests and a retrospective. Each prompt asks for the facts only you have, and returns a format you can paste into a doc or a tracker.
How to use these prompts
Every prompt works in ChatGPT, Claude or Gemini. Replace the blanks in double braces, or keep the default after the colon: {{deadline:end of Q1}} means "end of Q1" unless you change it.
Give the model real inputs: kickoff notes, the brief, last week's updates, the task list. When you paste long material, put it near the top and the instructions after it. Anthropic's prompting best practices recommend placing "long documents and inputs near the top of your prompt, above your query, instructions, and examples." The prompts below mark where pasted material goes, so move it up when it is long.
Ask for tables. Project documents are easier to check and import when every row has the same fields.
For meetings, the meeting notes prompts cover agendas, notes and follow-ups. For setting the goals a project serves, see the OKR prompts.
Which prompts help you start a project?
1. Project charter from kickoff notes
A one-page charter keeps everyone agreeing on the same project.
Turn these kickoff notes into a one-page project charter for
{{project:migrating our billing system}}.
Sections: purpose (why now), objectives with measures, scope (in and
out), key deliverables, milestones, stakeholders and their role, budget
or effort limits, assumptions, and open questions. Use only what's in the
notes. Put anything missing under "open questions" instead of guessing.
Notes:
{{notes}}
2. Scope statement with what's out
Most scope creep starts with something nobody said was out.
Write a scope statement for {{project:migrating our billing system}}.
Goal: {{goal}}
Deadline: {{deadline:end of Q1}}
Team: {{team:4 engineers and 1 designer}}
List what is in scope, what is explicitly out of scope, and what is
undecided. For each undecided item, write the question that settles it
and who should answer it. Flag anything in scope that looks too big for
the team and deadline.
3. Clarifying questions before you plan
Run this before any plan. It finds the gaps in a brief.
I've been asked to run this project. Read the brief and list the
questions I should ask the sponsor before planning, grouped by: goals and
success, scope, deadline and budget, people and approvals, and risks.
Put the 5 most important first. Don't answer them.
Brief:
{{brief}}
Which prompts help you plan the work?
4. Work breakdown structure
Splits a big deliverable into pieces small enough to estimate and assign.
Create a work breakdown structure for {{deliverable:a new customer onboarding flow}}.
Break it into phases, then work packages, then tasks of no more than
{{max_task:3 days}} each. Return a table: ID, phase, task, description,
depends on (IDs), and skills needed. Include tasks people forget:
reviews, testing, documentation, sign-offs and launch prep.
Context:
{{context}}
5. Three-point estimates
Ranges are more honest than single numbers and show where the risk is.
For each task below, give an optimistic, most likely and pessimistic
estimate in {{unit:days}}, and the expected value using
(optimistic + 4 x likely + pessimistic) / 6. Explain the biggest
uncertainty behind each pessimistic number. Mark tasks where the range is
so wide we should investigate before committing.
Team context: {{team:4 engineers and 1 designer}}
Tasks:
{{tasks}}
6. Dependencies and critical path
Shows which tasks decide the end date.
Here are tasks with durations and dependencies:
{{tasks}}
Identify the critical path, the total duration, and the tasks with
slack and how much. List the 3 tasks where a delay hurts most, and
suggest which work could run in parallel to shorten the schedule. Show
your working so I can check it.
7. Milestone timeline
A plan stakeholders can read in one glance.
Turn this plan into a milestone timeline from {{start:1 November}} to
{{deadline:end of Q1}}.
Return a table: milestone, date, what "done" means for it, owner, and
what it depends on. Add buffer before the final date and say how much.
Flag holidays or known busy periods: {{blackout:the last two weeks of December}}.
Plan:
{{plan}}
8. RACI chart
Settles who decides and who just needs to know.
Create a RACI chart for {{project:migrating our billing system}}.
People and roles: {{people}}
Key activities and decisions: {{activities}}
Return a table with activities as rows and people as columns, using R
(responsible), A (accountable), C (consulted) and I (informed). Each row
gets exactly one A. Flag rows with no R, too many Cs, or one person
carrying most of the Rs.
Which prompts help manage risk?
9. Risk register
Turns worries into owned, tracked risks.
Build a risk register for {{project:migrating our billing system}} from
these notes and concerns:
{{concerns}}
Return a table: risk, cause, impact, likelihood (low, medium, high),
impact level (low, medium, high), early warning sign, mitigation, owner
(leave blank if unknown), and review date. Add up to 5 common risks for
this kind of project that the notes miss, marked "suggested".
10. Pre-mortem
Atlassian's Team Playbook calls the premortem "a project management strategy that will help you prepare for every twist and turn": the team lists what could go wrong and right before the project starts, votes, and gives each mitigation an owner and a deadline. This prompt drafts the list your team then votes on.
Imagine it is {{future:six months from now}} and
{{project:migrating our billing system}} has failed badly.
Based on the plan below, write the 10 most likely reasons it failed,
ranked. Then list 5 things that went well and made the biggest
difference. For the top 3 failure reasons, suggest one mitigation with
an owner role and a deadline.
Plan:
{{plan}}
Which prompts help you report and adjust?
11. Weekly status report
Readers want the status and the asks, not a diary.
Write a weekly status report for {{audience:the steering group}} from
these updates.
Format: overall status (green, amber or red) with one sentence why; a
milestone table with planned date, forecast date and days of slip;
what's next; risks and issues; decisions or help needed, with a date.
Under {{words:250}} words. Don't soften bad news, and don't call it
green if a milestone has slipped.
Updates:
{{updates}}
12. Change request impact
For when someone asks for "one small addition" mid-project.
A stakeholder has asked for this change: {{change}}
Current plan and timeline:
{{plan}}
Assess the impact on scope, schedule, effort, cost, risk and quality.
Give 3 options (accept and move the date, accept and drop something,
defer to a later phase) with the trade-off of each. End with a short
message I can send to the requester.
13. Escalation email
When a blocker needs someone with more authority.
Write an escalation email to {{recipient:the VP of Engineering}} about
this blocker: {{blocker}}
Impact if not resolved by {{date}}: {{impact}}
What we've already tried: {{tried}}
Lead with the decision or help I need. Keep it under 150 words, factual,
with no blame. Give 2 options for resolving it and my recommendation.
14. Retrospective and lessons learned
Turns a project's history into changes for the next one.
Here are notes, timeline slips and feedback from {{project:migrating our billing system}}:
{{notes}}
Summarize: what went well, what didn't, and the root causes behind the
3 biggest problems (not just symptoms). Then list up to 5 concrete
changes for the next project, each with an owner role. Use only what's
in the notes and quote the evidence for each point.
How do you get better project plans from AI?
Supply the constraints. A model doesn't know your team size, the holidays, the other projects pulling on the same people, or which stakeholder needs extra care. Put those facts in the prompt, or the plan will assume an ideal world.
Ask it to show its working. For estimates and critical path, ask the model to show the numbers so you can spot a wrong dependency or an arithmetic slip.
Check every fact against your sources. Anthropic's guide to reducing hallucinations suggests asking the model to quote its sources for each claim and to retract anything it can't support. That matters most in status reports and retrospectives, where a made-up detail damages trust.
Review with the team. AI drafts the plan; the people doing the work confirm it.
What should a project manager keep out of the prompt?
Follow your company's rules on AI tools. Leave out client names, contract values, personal data about team members and anything under NDA unless your workspace is approved for it. OpenAI's prompt engineering guide for everyday work recommends giving the model background and acceptance criteria, and you can usually give that background with roles and placeholders instead of real names.
Performance judgments about people also stay with you. A model can summarize what happened, but it shouldn't decide who is to blame.
Make your favorite one a skill
If you write the same status report every Friday, turn the prompt into a skill so Claude, Claude Code or Codex follows your format every time. Paste it into the free SKILL.md generator. For goal-setting and PRDs, see the product manager prompts.
To keep these ready for the next kickoff, save them in Kipt with their {{blanks}}, tag them by project phase, and press ⌘K to find one. The library lives as Markdown in a private GitHub repo you own.