How to Use AI to Review and Improve Your Own Prompts

The Prompt That Made Me Feel Dumb

A few months into using AI tools seriously, I noticed a pattern: I'd write a prompt, get a mediocre response, tweak the prompt slightly, get another mediocre response, and eventually just give up and accept whatever I got. It felt like I was always one small adjustment away from something great, but I couldn't figure out what that adjustment was.

Then I tried something that felt almost too obvious: I asked Claude to critique my prompt before answering it. And the feedback was genuinely humbling. It pointed out that my prompt was vague, lacked context, and buried the actual ask under a pile of background information. It rewrote the prompt in about 45 seconds. The new version got me a response that was three times more useful.

That was the moment I started treating AI as a prompt coach, not just a prompt responder. Here's exactly how to do it.

Why This Works (And Why Most People Skip It)

Most people treat prompts like search queries — you type something in, you get something back, and if it's wrong you just try again. But AI models actually "understand" what makes a good prompt. They've been trained on enormous amounts of text about communication, instruction-following, and task completion. That means they can articulate why a prompt is weak, not just respond to it.

The reason people skip this step is simple: it feels inefficient. You're adding an extra round-trip before you even get to your real task. But once you start doing it regularly, two things happen. First, you get dramatically better results on complex or high-stakes tasks. Second, you start internalizing the feedback — and your prompts get better even when you don't ask for a critique.

Think of it like having a writing coach read your draft before you send an important email. Yes, it takes an extra few minutes. But the output quality difference is worth it.

The Three-Step Prompt Review Method

Here's the exact system I use. It takes about two minutes and consistently improves my results on anything that matters.

Step 1: Write your prompt as you normally would. Don't overthink it. Just write what you'd naturally type. This is your baseline.

Step 2: Ask the AI to critique it before responding. Use this wrapper prompt around your original prompt:

Step 3: Compare the original response to what the improved prompt would have gotten. The AI will answer its own improved version — so you'll see the difference immediately. Read the critique, internalize it, and save both versions if the improved one was significantly better.

A Real Example: Before and After

Let me show you this with an actual prompt I used recently. I was trying to get help writing a performance review for a team member. My original prompt was:

Claude's critique flagged: no role context, no specific accomplishments, no specifics about what "areas to improve" means, no tone guidance (formal vs. warm), and no target length. Then it rewrote the prompt as:

The output from the improved prompt was genuinely ready to use with minor edits. The original would have given me a generic template I'd have had to rewrite almost entirely. The critique took less than 30 seconds to read. Completely worth it.

Prompt Scoring: A Variation That's Even More Useful

Once you're comfortable with the basic critique method, try asking for a score. This gives you a quick benchmark and trains you to think about prompts more systematically.

I started using this for prompts I was going to reuse — things like templates for weekly recaps, code documentation requests, or client email drafts. Getting a score helps me understand whether I'm at a "good enough" level or whether there's a real gap I'm missing.

When to Use This (And When to Skip It)

I don't run every single prompt through a critique. That would be overkill and would slow me down on simple tasks. Here's how I decide:

Use prompt review when: The task is complex or multi-part. You're going to reuse this prompt as a template. The stakes are higher than usual (client work, job applications, important decisions). You've tried a prompt twice and gotten disappointing results both times.

Skip it when: It's a quick one-off question. You're brainstorming and the quality bar is intentionally low. You already have a proven prompt template for this task.

The rough rule I follow: anything that would take me more than 15 minutes to act on manually is worth spending 2 minutes getting the prompt right first.

The Meta-Skill You're Actually Building

Here's the thing I didn't expect when I started doing this: the AI's critiques started to sound like my own internal voice after a while. Now when I write a prompt, I naturally think "wait, did I say what format I want?" or "am I giving enough context for this to make sense without any background?" — before I even submit it.

That's the real payoff. You're not just getting better responses today. You're building an intuition for clear communication that makes you better at working with AI tools long-term. And honestly, it makes you better at writing instructions and briefs for humans too. The skills transfer.

The AI is one of the best feedback loops you'll ever have access to — completely judgment-free, always available, and genuinely useful at articulating what's missing. Most people never take advantage of it this way. Now you know how.