The Skill That Will Matter More Than Any Degree: How to Prompt AI

AI models contain multitudes — they've been trained on writing from doctors, lawyers, novelists, marketers, engineers, teachers, and every other profession. Without a role, the model produces a generic average. With a role, it draws from the specific domain expertise, tone, and perspective you need. This is often the single highest-leverage thing you can add to any prompt.

AI can only work with what you give it. When you paste in an email without context, you get a response calibrated to the email text alone. When you also explain your relationship with the sender, your goal, the history, and what you're trying to avoid — you get something actually useful. Context is free. Omitting it is the most common reason AI outputs miss the mark.

Without format instructions, AI defaults to paragraphs — or whatever structure it thinks is appropriate. That almost never matches what you actually need. Specifying format is one of the easiest improvements you can make and produces immediately usable output instead of something you have to restructure after the fact.

AI models have been trained on every cliché ever published on the internet. They will default to "results-driven professional," "game-changing solution," "seamlessly integrated," and "passionate team" unless explicitly told not to. Negative instructions are often the difference between output that sounds like a person wrote it and output that reads like a press release from 2009. Banning what you don't want is faster and more effective than trying to describe the positive alternative.

AI generates text sequentially — each word is predicted from what came before. When you push it to a conclusion immediately, it shortcuts the reasoning process. Asking the model to think through a problem in steps before answering — chain-of-thought prompting — produces dramatically better results on any complex, analytical, or judgment-dependent task. This is especially important for decisions, evaluations, plans, and anything where nuance matters.

Describing a tone or style to AI is hard — the words you'd use to describe "warm but professional" or "punchy but not gimmicky" are ambiguous. Showing examples is almost always more effective than describing. Paste in two or three pieces of writing that match what you want and tell the AI to match them. This technique — called few-shot prompting in research contexts — is particularly powerful for brand voice, writing style, and any task where you have approved examples to reference.

Prompting is a conversation, not a one-shot command. The first output is a starting point. The single most common mistake is either accepting the first result or giving feedback so vague that the next version isn't better. "Make it better" tells the AI nothing. "The opening sentence is too weak, the second paragraph repeats the first, and the tone in the third section sounds corporate — fix those three things specifically" tells it exactly what to do.

Two iteration habits that accelerate good output: first, tell the AI to give you only the revised content (not an explanation of what it changed — that wastes your reading time). Second, limit each round of feedback to 3 specific changes at most. More than that and the AI tries to fix everything at once and often breaks something that was already working.