“Treat the Prompt as a Requirement”
Most project managers have an AI assistant within reach these days. It might be in a browser tab, a chat window, or built into one of the tools we already use for planning and reporting.
So the question is no longer whether we have access to AI. The more interesting question is why two project managers can use the same AI tool and get completely different results.
Usually, the problem isn’t the model. It’s the brief.
Think of a large language model as a very capable analyst who works incredibly fast but knows absolutely nothing about your project until you tell it. Give it a vague request, and you will probably get a vague answer. Give it a clear brief, and you are much more likely to get something you can actually work with.
That is where prompt engineering comes in.
Despite how the name sounds, the basic idea should feel familiar to experienced project managers. We already spend our careers defining objectives, clarifying requirements, explaining constraints, and making sure people understand what “done” looks like.
The same thinking works with AI.
Prompt Engineering Is Requirements Engineering
I like to think of a prompt as a small requirement.
We all know what happens when a requirement is ambiguous. The team interprets it differently, the deliverable misses the mark, and eventually someone has to do the work again.
The same thing happens with AI.
A simple prompt structure covers most situations:
- Role. Who do you want the AI to act as? “Act as a senior risk manager” gives it a different perspective and level of detail than simply asking a question.
- Context. Give it the project information it needs: the project stage, constraints, relevant data, dependencies, or documents. Without enough context, the answer may be too generic to be useful.
- Task. Be specific about what you want. “Identify 15 to 20 risks across these categories” is much clearer than “list some risks.”
- Constraints. Tell it what not to do. For example: “do not invent facts,” “flag assumptions,” or “stay within a specific word count.”
- Output. Tell it what the answer should look like: a table, a short briefing, a memo, or a list.
Instead of asking:
“Analyze my project risks.”
try:
“Act as a senior project risk manager. Review the project information I provide and identify the three risks most likely to affect the critical path. Do not invent information. Separate facts from assumptions. Return a table showing Risk, Probability, Impact, Early Warning Indicator, and Response.”
The difference is small, but the result can be significant. You have given the AI a much clearer definition of the job.
Three Ways to Put This to Work
The examples below are situations most project managers will recognize. You can adapt the prompts to your own projects.
I have also included a question to ask next for each one because, in my experience, that is where AI becomes much more useful than simply asking it to produce a first draft.
1. Pressure Test a Risk Before the Steering Committee
Imagine a project where a critical subassembly is coming from a single external supplier. The supplier has already missed two milestones. Integration testing is four weeks away, and the delivery date is contractually fixed.
The risk is fairly obvious.
The harder part is making sure you have thought through the situation before you walk into the steering committee.
Instead of asking AI what you should do, give it the situation and ask it to help you structure the analysis:
Prompt
| Act as a senior project risk manager. Project context: • A critical component is being delivered by a single external supplier. • The supplier has missed two milestones. • Integration testing starts in four weeks. • The final delivery date is contractually fixed. Identify: 1. The primary risk, stated as cause, event, and impact 2. Likely causes 3. Early warning indicators 4. Preventive actions 5. Contingency actions 6. Decisions that require sponsor involvement Do not invent project facts. Mark every assumption clearly. Present the result as a table. |
I would not copy the AI’s output directly into the risk register. I would use it to challenge my own thinking and see whether there are gaps I have not considered.
Then ask:
“What information am I missing before I present this risk to the steering committee?”
That second question can be surprisingly useful. It may uncover an assumption or missing piece of information before it becomes an uncomfortable question in the meeting.
What to check: Every assumption the AI flagged, and whether the probability and impact make sense for your organization and project.
2. Turn Raw Project Data Into an Executive Update
Most project managers have no shortage of data. The problem is finding the time to turn all that information into something a sponsor can understand in a minute or two.
Suppose you have:
- Planned progress: 70%
- Actual progress: 62%
- One major milestone is 10 business days late
- Budget is 4% over
- The contractual deadline is currently not affected
You could simply ask AI to “write a status report.”
A better approach is to give it the numbers, the audience, and the purpose:
Prompt
| Act as an experienced project manager preparing an executive update. Project data: • Planned progress: 70% • Actual progress: 62% • One major milestone is 10 business days late • Budget: 4% over budget • The contractual deadline is currently not affected Use only the data provided. Do not invent explanations for the variance. Write the update with these sections: • Overall status (one line: Green, Amber, or Red, with one sentence of reasoning) • Progress against plan • Key achievement • Main issue • Business impact • Corrective action • Decisions or support required • Next milestone Audience: Project Sponsor and Steering Committee. Maximum length: 200 words. Tone: Factual and direct. |
This gives you a useful first draft very quickly.
But do not stop there.
Ask AI to switch perspectives:
“Read this as the project sponsor. What five questions would you ask me?”
Now you are preparing for the meeting, not just preparing the report.
What to check: The status color and the reasoning behind it. That is your management judgment, not the AI’s decision.
3. Structure a Change Request Tradeoff
A change request is rarely just about adding something to scope. A seemingly small change can affect the schedule, resources, quality, dependencies, and risk.
That is why Perform Integrated Change Control exists.
AI can help you lay out the tradeoffs clearly so the Change Control Board can start with a structured analysis rather than an argument about preferences.
Imagine:
- The launch date is fixed.
- A requested feature needs roughly six additional weeks of work.
- The development team is already near capacity.
- Testing is on the critical path.
Try:
Prompt
| Act as a senior project manager preparing a change request analysis. Situation: • The launch date is fixed. • The requested feature needs roughly six additional weeks of work. • The development team is near capacity. • Testing is on the critical path. Compare three options: A. Add the feature without moving the deadline. B. Add the feature and move the deadline. C. Deliver the feature in a later release. For each option, assess scope, schedule, resources, quality risk, delivery risk, and stakeholder impact. Identify the key tradeoffs and the information still missing. Do not make the decision for me. |
The value here is not that AI makes the decision.
It does not.
Its value is that it can help organize the decision, make the tradeoffs visible, and highlight questions that deserve attention.
The recommendation still belongs to the board, and the accountability still belongs to you.
The Real Skill Is What You Ask Next
Prompt engineering is sometimes presented as the search for one perfect prompt.
I do not think that is how most project work actually happens.
Good project managers ask questions, look at the answer, challenge it, and ask another question. Working with AI can follow exactly the same pattern.
Start broad:
“Identify the major risks on this project.”
Then narrow it:
“Which three could affect the critical path?”
Challenge the answer:
“What evidence supports that assessment?”
Look for gaps:
“What information is missing?”
And finally, put yourself in the stakeholder’s position:
“What would a skeptical sponsor challenge?”
The pattern is simple:
Prompt, Review, Challenge, Refine, Validate, Decide.
AI can help you move through the early stages faster.
You still own the last one.
Guardrails That Do Not Transfer
There is one rule I would keep in mind regardless of how good your prompt is:
AI supports the analysis. The project manager validates the facts.
Generative AI can produce an answer that sounds confident and professional while being based on a wrong assumption or invented detail.
In a project environment, that can quickly become a problem. An incorrect statement generated by AI could find its way into a schedule, budget, contract discussion, compliance decision, or stakeholder communication.
For important decisions, check AI output against authoritative sources: the project plan, contract, organizational policies, and the people who actually own the work.
Data governance matters too. Before putting confidential customer, financial, contractual, or personnel information into an AI tool, understand your organization’s AI policy and how the tool handles your data.
And accountability does not transfer. Consistent with the PMI Code of Ethics, the decision and its outcome stay with the project manager, not the tool.
Where This Is Heading
AI is increasingly appearing inside the tools project managers already use: documents, email, scheduling, reporting, and collaboration platforms.
As access becomes more common, simply knowing how to use an AI tool will become less of a differentiator.
The more valuable skill will be knowing how to frame a problem.
What context does the AI need? What assumptions should it avoid? What constraints matter? What question are you actually trying to answer?
Those are not new project management questions. AI has simply given us a new place to apply them.
Closing
You do not need to become an AI engineer to get value from prompt engineering.
You need to become better at translating your professional judgment into clear instructions, and better at deciding what to do with the answer that comes back.
The advantage is not knowing how to ask AI everything. It is knowing what to ask, why to ask it, and what to do with the response.
The best project managers will not be the ones who ask AI to do their jobs. They will be the ones who know how to use AI to improve the quality of their thinking.
