One-Shot Prompting

Best way to prompt (and way easier)

This post is a very pragmatic piece on how to get better results from AI without the usual prompt engineering. I included examples you can copy and use. It's focused more on the framework and less on the theory of why LLMs work.

After spending months writing elaborate prompts with bullet points, constraints, and step-by-step instructions, I discovered something simpler. One-shot prompting. Show the AI one perfect example. Ask for more like it. That's it. I discovered this through this tweet from Mich Lieben.

The Method

Step 1: Find one perfect example of what you want.

Step 2: Paste it with this instruction: "Write one like this, but for [new context]."

That's the entire process.

Why This Works

Large language models are pattern-matching systems. They excel at imitation. They struggle with complex instructions.

Traditional prompting forces the AI to juggle multiple requirements. One-shot prompting gives it a concrete pattern to follow.

How I use it

There is a thing called swipe files. This is a curation of mainly marketing/sales/creative examples used for (future) inspiration. And for the one-shot prompting you need a strong source file/example. So I go to swipefile.com for good marketing copy, ads, and email examples. But I also check my own work for successful emails, proposals, descriptions, check industry leaders for their best content in your niche and competitors for stuff I’m jealous of. .

Pick examples that performed well. There has to be evidence of success: high open rates, conversions, engagement.

Build your own

Start building your own swipe file.

• Save every good email you receive

• Screenshot effective social posts

• Bookmark landing pages that convert

• Collect sales pages that work

Testing: Start with small requests to test the pattern.

• Use the example for similar contexts first

• Gradually push the boundaries

• Note which elements the AI preserves vs. changes

How to Start

1. Find one perfect example in your domain.

2. Test it with a simple request.

3. Ask for 3-5 variations.

4. Refine for specific use cases

5. Scale the approach across different content types.

My own examples

This is the example of a ‘great’ ad I use as input.

Prompt: ‘create 3 more ads like this:’

Results:

1. "I used to spend half my day resizing charts pixel by pixel. Beautiful.ai made me feel useless—and then incredibly relieved."

2. "I used to tweak font sizes like my career depended on it. Now Beautiful.ai handles it while I pretend I’m still essential."

3. "My slides used to take hours, a stiff neck, and three existential crises. Beautiful.ai ruined that tradition in under a minute."

Prompt: ‘create 3 more ads like this:’

Results:

1. What would you prefer:

   a. A full-price SaaSGrow ticket

   or

   b. $900 off a SaaSGrow ticket?

   If you picked a, you can stop reading.

   Otherwise, we thought you’d want to know: Super Early Bird tickets are only available until Thursday, April 20. After that, prices go up. No complaints if you miss it.

   [Snag your ticket now]

2. What sounds better:

   a. Waiting for the replay

   or

   b. Joining live, asking questions, and getting bonus materials free?

   If you answered a, feel free to close this tab.

   If not, register for AI Trends Live before Monday, July 10 to get access to the exclusive live Q&A and prompt packs.

   [Reserve your spot now]

3. Your pick:

   a. Pay £1,250

   or

   b. Pay £850 (and use the rest for your flight)?

   Early access tickets are only live for 6 more days. After Friday, August 18, prices jump.

   So if you’re serious about scaling pipeline this year, don’t wait.

   [Get your discount ticket]

This works for almost any task. Want a better cold email? Share a good one. Need a clearer job description? Paste one and say, “Now write one like this for [X].” The AI doesn’t need long instructions, just a solid reference. One good example is often enough to get useful, accurate output, quickly and without much back and forth.

Good luck!