Two technicians ask an AI the same question. One gets a generic essay; the other gets a working script with sensible error handling and a note about the edge case that would have bitten them. Same tool, same day. The difference was entirely in how they asked — and "how to ask" turns out to be a learnable, improvable skill, not a knack.
Why phrasing matters mechanically
Remember the engine from earlier on this trail: the model continues text based on patterns. Your prompt is the pattern being continued. A vague prompt matches millions of vague documents, so you get the average of them — generic mush. A precise prompt with context narrows the pattern-space to exactly the kind of document you actually want. You're not talking to a person who fills gaps with common sense; you're steering a continuation. Steer it.
The four moves that do most of the work
- Give context like you'd brief a contractor. "Write a PowerShell script to clear temp files" versus "Write a PowerShell script for Windows 11 machines managed by Intune, to clear user temp folders older than 7 days, log actions to C:\Logs, and never touch anything in use." The second prompt contains the answer's shape.
- State constraints and the audience. "Explain in plain English for a new starter." "No third-party modules." "Under 200 words." Constraints are steering.
- Show an example of what good looks like. Paste a sample of your ticket-note format and ask it to match. One example beats three paragraphs of description — the model is a pattern engine, so hand it a pattern.
- Iterate instead of restarting. The first answer is a draft. "Good — now add error handling." "Shorter." "Explain line 12." Refinement is where the value compounds; treat it like code review, not a slot machine.
The professional guard-rails, non-negotiable: never paste passwords, keys, customer data, or anything confidential into a public AI tool — treat the chat box like a public forum unless your organisation has an approved private deployment. And everything from the hallucination waypoint applies: verify before production, always.
Where this fits in your career
Prompting well won't replace knowing your field — the technician who understood Intune wrote the better prompt because they knew what to specify. That's the honest relationship: AI multiplies existing knowledge rather than substituting for it. But between two equally knowledgeable people, the one who can brief an AI precisely now works measurably faster — drafting scripts, summarising tickets, explaining errors, producing documentation. It's become a baseline tool skill, like search-engine fluency was twenty years ago. Learn it the same way: deliberately, on low-stakes tasks first.