Handling projects with ChatGPT: from brief to delivery
Use ChatGPT as a project partner: write the brief, break down tasks, find the critical path, record decisions, produce in pieces and review against a definition of done.
In this lesson you will
- Write a project brief with scope and done criteria
- Break a project into dependent tasks
- Keep decisions, facts and versions organised
- Review deliverables against success measures
About 18 minutes
A project is more than a prompt
Noor decides to launch a small line of candles. She asks ChatGPT for a business plan and receives eight confident pages. She feels productive for an afternoon, then realises she has no idea what to do on Monday morning. The plan was text; what she needed was a project: a goal, a scope, a schedule, decisions, checks and a record. This lesson shows how to use ChatGPT as a project partner from the first idea to the finished result, with you firmly in the driver’s seat. The method works for a business launch, a research paper, a wedding, a website, a school event or a home move.
The project loop
- Define. The goal, the deadline, the budget, the audience, what counts as done.
- Break down. Phases, tasks, owners, dependencies.
- Decide. Options, trade-offs, risks, choices you record.
- Produce. Draft the deliverables in small checked pieces.
- Review. Test against the definition of done.
- Deliver and learn. Ship, then note what you would change.
Step 1: the project brief
Ask the assistant to interview you, then produce a one-page brief that you correct. A brief is the anchor that stops drift.
I want to launch a small candle line online in 8 weeks with a Rs 150,000 budget. Interview me with up to 7 questions, one at a time, to build a project brief. Then write the brief with: goal, scope (in and out), audience, deadline, budget, success measures, risks, and open questions.
Step 2: break it down and find the critical path
Tasks depend on each other. Some can happen in parallel; the longest chain of dependent tasks sets the minimum project length, called the critical path. Ask ChatGPT for a task list with durations and dependencies, then check it yourself. You can even test the logic with a short script. This example was run for real:
tasks = [
{"name": "Research suppliers", "days": 2, "needs": []},
{"name": "Choose three scents", "days": 1, "needs": ["Research suppliers"]},
{"name": "Order samples", "days": 5, "needs": ["Choose three scents"]},
{"name": "Draft product pages", "days": 2, "needs": ["Choose three scents"]},
{"name": "Photograph samples", "days": 1, "needs": ["Order samples"]},
{"name": "Launch", "days": 1, "needs": ["Draft product pages", "Photograph samples"]},
]
finish = {}
def end_of(task):
if task["name"] in finish:
return finish[task["name"]]
start = max([end_of(t) for t in tasks if t["name"] in task["needs"]], default=0)
finish[task["name"]] = start + task["days"]
return finish[task["name"]]
for t in tasks:
print(f"{t['name']:<22} finishes on day {end_of(t)}")
print("Whole project:", max(finish.values()), "days")
Research suppliers finishes on day 2
Choose three scents finishes on day 3
Order samples finishes on day 8
Draft product pages finishes on day 5
Photograph samples finishes on day 9
Launch finishes on day 10
Whole project: 10 days
Reading the result: writing the product pages can happen while samples ship, so only the chain research, scents, samples, photos, launch decides the finish date of 10 days. Ordering samples (5 days) is the bottleneck, so ordering them a day earlier saves the whole project a day. Spotting that is the kind of thinking you must do yourself. The assistant can draft the list; you test whether it makes sense.
Step 3: decisions and risks
For the supplier decision, give me 3 options with costs, advantages, risks and what I would need to find out. Then list the 5 biggest risks to the whole project, each with a warning sign and a prevention step. Mark every number as ESTIMATE unless I gave it.
Keep a short decision log: date, decision, reason, who decided. It saves arguments later and trains your judgment.
Step 4: produce in small pieces
- Work deliverable by deliverable, not everything in one chat.
- Give the brief and the relevant facts every time you start a new piece.
- Use a project or saved instructions feature, if your version has one, to keep the brief and style rules available (availability varies).
- Keep a running facts file: confirmed numbers, names, dates. Paste it into prompts to prevent invented facts.
- Version your files: product-pages-v1, v2, so you can go back.
Step 5: review against ‘done’
Here is my definition of done and my current deliverables. Act as a strict project reviewer. For each success measure, say met, partly met or not met, with the evidence from the text. List gaps and the smallest fix for each. Do not praise.
Staying in control
| Danger | What happens | Counter |
|---|---|---|
| Scope creep | The assistant keeps suggesting extras | Check every idea against the scope in the brief |
| Plausible nonsense | Neat plans built on invented numbers | Label estimates, verify real figures |
| Drift in long chats | Early decisions are forgotten | Restate the brief, summarise state regularly |
| Loss of ownership | You can’t explain your own plan | Rewrite key sections in your own words |
| No record | You cannot show why you chose | Keep the decision log and versions |
Working with other people
- Tell collaborators which parts used AI and what has been checked.
- Agree one place for the facts file and the current versions.
- Use AI to prepare agendas and summaries, but let people confirm decisions and owners.
- Remember that others may have different comfort levels and rules about AI.
Treating the first plan as the plan. It is a draft to challenge. Ask: what is missing, what is wrong, what would a sceptic say?
Skipping the definition of done. Without it the project never ends, or ends arguing about whether it did.
Ask for a summary of decisions, open questions and the next three actions. Paste it at the top of the next session, and you can resume a project in two minutes, even after a month away.
Check what you learned
Five quick questions. Pick an answer and you will see straight away whether it is right and why.
What anchors a project against drift?
- More adjectives
- A longer chat
- A random schedule
- A clear brief with scope and definition of done
The brief is the reference point.
What is the critical path?
- The shortest task
- The longest chain of dependent tasks
- The cheapest task
- The first task
It sets the minimum project length.
Why keep a decision log?
- It makes chats shorter
- It deletes risks
- It records what, why and who decided
- It replaces the plan
It prevents later confusion.
What helps stop invented facts in a project?
- A facts file you paste into prompts
- Longer prompts without facts
- Ending the chat early
- Asking for confidence
Confirmed facts keep drafts grounded.
How should a session end?
- With nothing
- By deleting notes
- With a hand-off summary of decisions and next steps
- By changing the goal
It makes resuming easy.
Summary
- A project needs a brief, tasks, decisions, checks and a record.
- Find the critical path yourself.
- Keep a facts file and a decision log.
- End sessions with a hand-off summary.
Questions
Can ChatGPT replace a project manager?
It can draft plans and checklists, but it cannot own accountability, negotiate with people or know your real constraints.
How do I stop the plan from being generic?
Give real constraints, real numbers and ask for risks and alternatives specific to your situation.
What if the chat gets too long?
Ask for a state summary and start a fresh chat with the brief and summary pasted at the top.
