In your experience with GenAI, how has refining a prompt drastically changed the output quality?

I’ve taken a slightly different approach to improving prompt quality: I built a custom GPT that writes better prompts for us.

Instead of spending hours teaching colleagues how to structure effective prompts, it now takes minutes. The model uses different prompt frameworks based on the situation and even asks clarifying questions if context is missing. All the user has to do is describe their need in plain language, the AI determines the best structure and builds the optimized prompt. As a bonus it also explains what it did and why to help educate on prompt engineering as well.

I am seeing higher-quality outputs without requiring everyone to become a prompt engineer.

I shared this with a client this week who had almost no AI experience, and she laughed, saying: “So… we’re asking AI to figure out what to ask itself?” Exactly! And it works, it only took 25 minutes and she was completing a task that took her hours, and she didn't like to do, in minutes. This shift lets teams focus on outcomes instead of struggling with prompt mechanics.

Refining a prompt transforms vague outputs into precise, actionable insights. A broad prompt may miss context, but with iterative clarity—like adding scope, tone, or format—the results become sharper and more useful. Small tweaks often lead to big improvements in quality, especially for project plans, reports, or stakeholder communications. Look up the RTF (Request, Type, Format) and CREATE (Character, Request, Example, Adjustment, Type, Evaluation) frameworks in PMI Infinity. It's a great AI resource for the PM community.

I've had some real "wow" moments where a small tweak completely transformed what I got back. Let me share a couple of examples that really stick out.

The stakeholder communication disaster that became a win Early on, I asked AI to "write an email about project delays." What I got was this generic, corporate-speak disaster that would've made stakeholders panic. But when I refined it to "write a confident, solution-focused email to the steering committee explaining a 2-week delay in Phase 2 due to vendor integration issues, emphasizing our mitigation plan and revised delivery date," suddenly I had something I could actually send. The difference was night and day.

Risk assessment that went from useless to invaluable I once asked for "project risks" and got this generic list that could apply to any project anywhere. Completely useless. But when I refined to "identify technical, resource, and timeline risks specific to a cloud migration project for a 200-person financial services company with strict compliance requirements, including probability ratings and mitigation strategies," the output became genuinely insightful and actionable.

The meeting agenda transformation Simple request: "create a meeting agenda" gave me boring bullet points. But "create a 90-minute project kick-off agenda for 8 stakeholders from IT, Finance, and Operations, including time allocations, expected outcomes for each segment, and prep materials needed" - suddenly I had something I could actually use to run an effective meeting.

The pattern I've noticed is that the more I treat the AI like I would brief a senior consultant - with context, constraints, and clear expectations - the more professional and usable the output becomes. It's really changed how I approach delegation in general, honestly. Specificity and context are everything!

What's your experience been? Have you had any similar "before and after" moments?