The difference between a useful AI answer and a useless one is almost always the question.

That's a simpler claim than it looks. Most recruiters — and most of everyone else — who try a tool like ChatGPT and come away unimpressed didn't get a bad answer. They asked a bad question. And because AI is polite, it gave them exactly what they asked for: generic, unspecific, hedged, vaguely helpful in the way a fortune cookie is vaguely helpful.

Getting good output is a skill, and it's the single highest-leverage skill a working recruiter can pick up in the next six months. Not because prompting is difficult — it isn't — but because it compounds. Every hour you spend getting better at prompting pays back in every subsequent hour of AI-assisted work, for the rest of your career.

Here's what the skill actually looks like.

The weak prompt

"Write me interview questions for a sales role."

This is how most people start. It seems reasonable. It's also the kind of prompt that produces the output people complain about.

Try it and you get six questions that could apply to any sales role at any company in any industry. "Tell me about your most successful sales year. How do you handle rejection? What's your approach to pipeline management?" They're not wrong. They're just not useful — you already knew those questions existed.

The AI didn't fail. You asked for generic sales interview questions. You got generic sales interview questions.

The strong prompt

"I'm interviewing for a Senior Enterprise AE role at a Series B SaaS company selling a $60K-ACV data platform to VPs of Engineering at mid-market tech companies. Candidates have 8+ years of experience and have closed $1M+ in pipeline per year. I want to assess three things: ability to run multi-threaded deals across 4–6 stakeholders, judgment on deal qualification (especially killing bad deals early), and ability to coach a technical buyer through a complex evaluation. Generate eight questions — four behavioral, four scenario-based — with the answer patterns I should listen for. Format as a table: question, what it's testing, signal in a strong response."

Read those two side by side. The second one probably feels excessive. It isn't. It's the amount of context a real human interviewer would already have in their head when they sat down to meet this candidate. You're just putting it on the page.

And the output you get back is dramatically better — not because the AI is working harder, but because you finally told it what you actually wanted.

The Four Things Every Good Prompt Contains

Good prompting isn't a formula to memorize. It's four things to include whenever you're asking an AI to do something nontrivial. If your output is weak, you're almost always missing one of them.

Context. Who is this for? What's the situation? What do you already know? "Senior Enterprise AE at a Series B SaaS company" is context. "Sales role" is not.

Constraints. What are the boundaries? "Eight questions, four behavioral and four scenario-based" is a constraint. "Write questions" is not. Constraints narrow the possibility space and force specificity. Vague asks produce vague answers.

Examples. When possible, show what good looks like. "Similar to Google's structured behavioral interviews" or "in the style of First Round's talent playbook." Examples give the AI a target. Without one, it averages across everything it's ever seen, which is the recipe for a generic output.

Format. How should the answer be structured? A table. Bullets. A one-page summary. A conversation. Specifying format is the single fastest way to go from "this is kind of useful" to "I can use this immediately".

That's the whole skill. Context, constraints, examples, format. Miss one and the output slips toward the generic. Include all four and you usually get something you can actually work with.

Prompts are Drafts, not Commands

One more thing, and it matters.

A common mistake is treating the first output as the deliverable. It isn't. The first output is a draft. You read it, notice what's weak, and respond — "make the second question more scenario-specific", "cut the fourth one, it's too generic", "add a question that tests for pipeline discipline specifically". The conversation is the product, not the first answer.

Recruiters who are good at prompting don't write one prompt and submit the output. They iterate. Usually three or four rounds. The final output rarely looks anything like what the first prompt produced — and that's the point. Prompting isn't a single instruction. It's a way of thinking out loud with a tool that can keep up.

The Real Skill Underneath

Here's the thing nobody tells you about prompting. The hard part isn't the AI.

The hard part is being specific about what you want.

Most people don't know what they want until they see it. Weak prompts come from not having clarified, in your own head, what a good answer would even look like. You can't ask the AI to write you a good outreach message if you can't articulate — to yourself — what makes an outreach message good for this candidate, this role, this moment.

Prompting forces that clarity. Which is why getting good at it tends to make people better at their actual jobs, not just their AI-assisted ones. The prompt is the thinking. The output is the proof.

When you get stuck, the problem usually isn't the tool.

If you want 80+ ready-to-use prompts for every hiring task, skip ahead to "Lesson 21 — The AI Prompt Pack"

Next: Lesson 07 — The New Hiring Stack