Golden Prompt Lab

Techniques That Get Better Results

The previous lessons covered the basics: be specific, include context, say what you want the answer to look like. That alone will take you far. But there are a handful of techniques that regularly get even better results — especially when you're working on something that matters.

None of these are complicated. They're just small adjustments to how you ask.

Ask it to think out loud

When you're trying to make a decision or work through something with a lot of moving parts, ask the AI to walk through its reasoning before giving you an answer. You'll get a much more thoughtful response — and you'll be able to see exactly where you agree or disagree.

Task: Help making a decision

Vague

Should I take this job offer?

Specific

I have a job offer I'm trying to decide about. It pays $8,000 more per year but the commute is an hour each way instead of 20 minutes. I have two young kids and I am already feeling stretched thin on time. Can you think through the tradeoffs with me before giving me your take?

Asking it to think through the tradeoffs first means you get reasoning you can actually evaluate, not just a quick yes or no.

Try the specific version:

The phrase "think through this with me before giving me an answer" or "walk me through your reasoning" is a reliable way to trigger this. Try it with something real:

Prompt

I need to decide: [decision]. Before you give me a recommendation, think through [considerations] out loud. Then give me your take.

Show it what you want

Sometimes it's easier to show the AI what a good answer looks like than to describe it. This is called "few-shot prompting" — a fancy name for a simple idea: give it an example or two before you ask for the real thing.

This is especially useful when you want something to sound like you, not like a generic AI response.

Task: Write something in your voice

Vague

Write a thank-you note for my neighbor who watched my dog.

Specific

Here are two thank-you notes I have written in the past so you can see how I write. Note 1: Thanks so much for grabbing my mail last week — you are a lifesaver and I owe you one. Note 2: Seriously, thank you. Having someone I trust nearby means everything. Now write a thank-you note for my neighbor Janet who watched my dog Biscuit for three days while I was at my sister's wedding.

Showing two examples of how you write takes about 30 seconds and the result sounds like you, not like a form letter.

Try the specific version:

Give it a role to play

The AI responds differently depending on who you ask it to be. A vague role like "you are an expert" doesn't do much. A specific role that matches what you actually need makes a real difference.

Some useful ways to frame this:

  • Explain it like a patient teacher would: "Explain what Medicare Part B covers as if I have never dealt with health insurance before and I am easily confused by jargon."
  • Give it the perspective of someone I trust: "Read this email I'm about to send to my landlord and tell me if it sounds too aggressive — be honest like a good friend would be."
  • Imagine the person I'm writing to: "Read this the way my 75-year-old mother would. Would she understand it? Would anything confuse or worry her?"
  • Put it in the shoes of the other side: "Respond as someone who is skeptical of this idea. What would they push back on?"

The more specific the role, the more specific and useful the answer.

Ask it what's wrong

One of the most useful things you can do is ask the AI to look critically at something — either something it just wrote, or something you wrote yourself.

Prompt

I am about to [action]. Before I do, I want a gut check. Does anything here seem [concern]? Be honest — I would rather know now.

Before you send anything important — a complaint, a request, a message to someone you are nervous about — paste it in and ask: "Is there anything here that might land wrong?" It takes thirty seconds and has saved many a sent-message regret.

Break big tasks into steps

If you have something complicated to do, asking the AI to do all of it at once usually produces a mediocre result across the board. Breaking it into steps — where each step builds on the last — produces much better results at each stage.

Instead of: "Help me plan my mother's 80th birthday party — figure out the venue, the guest list, the food, and write the invitations."

Try it in steps:

  1. "My mom is turning 80 in April. She loves her garden, hates loud music, and has about 30 close friends and family. What kind of party setting would suit her best?"
  2. "I like the garden party idea. Given that it's April in Colorado and weather can be unpredictable, what should I think about for backup plans?"
  3. "Good points. Now help me write a guest list worksheet — what information do I need to collect for each person?"
  4. "Now draft a warm, simple invitation for a garden party at my house on April 20th at 2pm. Mention it is a celebration of her 80th birthday. RSVP by April 10th."

Each answer is better because it has the context from everything that came before. And if one step goes sideways, you fix just that step — not start over from scratch.

Use constraints on purpose

Telling the AI what not to do, or what limits to stay within, often improves the answer as much as telling it what to do.

Useful things to add:

  • Audience: "My reader is 70 years old and not comfortable with technology"
  • Length: "Please keep this to two short paragraphs" or "Just give me three bullet points"
  • Tone: "Keep it friendly but firm" or "Don't be too formal"
  • What to leave out: "Don't include medical advice — I just want general information" or "Don't suggest anything that costs money"

Piling on too many constraints at once can backfire. If you tell the AI to be short, formal, avoid jargon, include three examples, and stay positive — all at the same time — it will try to satisfy everything and probably satisfy nothing well. One or two clear constraints work better than five competing ones.

Putting it together

You don't need to use all of these every time. For a quick question, just ask. But when you're working on something that matters — a difficult email, a big decision, a message you're nervous about — these techniques are in your back pocket. Pick the one that fits and give it a try.