Most PMs Are Using AI Like a Search Engine. Here’s What You’re Missing.

When I first started experimenting with large language models in my project management work, I was treating them the way most people do: ask a question, get an answer, move on. Useful, but not transformative. The shift happened when I realized that the quality of what the model gives you is almost entirely determined by how you ask. That’s prompt engineering in practice — and once you understand it, the same tools you’ve been using start producing dramatically different results.

LLMs like Claude, ChatGPT, and Gemini don’t just retrieve information. They generate responses shaped by context, framing, and instruction. Give a vague prompt and you get a generic response. Give a structured, specific prompt — with clear role, clear task, and clear expectations for the output — and you get something you can actually use. The RTF method captures this simply: define the Role, the Task, and the Format before you ask anything. A prompt like “act as a senior project manager reviewing a risk register, and flag the top three risks most likely to impact the delivery timeline, in a bullet format under 100 words” will outperform “what are the risks here?” every time.

For more complex outputs, the CREATE method gives you more control. You define the Character (who the AI is playing), the Request (exactly what you need), Examples of what good looks like, any Adjustments or constraints, the Type of output you want, and space for the model to Evaluate its own response. It sounds like more work, but once you build a few of these prompts as templates, you reuse them constantly. I have prompt templates for stakeholder update drafts, risk summaries, retrospective facilitation, and scope change analyses. Each one took twenty minutes to build and saves me hours every week.

The practical applications in delivery work are broader than most PMs realize: generating first drafts of project plans and status reports, brainstorming risk scenarios for a new workstream, decomposing ambiguous requirements into structured task breakdowns, and drafting stakeholder communications tailored to different audiences. None of this removes your judgment from the process — you still review, refine, and decide. What it removes is the blank page problem and the time spent on the first 70% of a document that doesn’t require your unique insight anyway.

The one habit that matters most: always treat AI output as a first draft, never a final one. The model doesn’t know your stakeholders, your organizational politics, or the context that lives in your head. You do. The combination of what it produces quickly and what you refine deliberately is where the real value lives. Start simple, iterate on your prompts, and resist the urge to use the output unedited. The PM who learns to prompt well doesn’t just work faster. They think more clearly, because the process of writing a good prompt forces you to define exactly what you actually need.