Few-Shot Prompting: One Example Beats Five Rules

Stop describing the style you want. Show the AI one example, and watch it match you.

You have a picture in your head of what you want. A certain tone. A certain shape. So you write more and more rules to describe it.

And the AI still misses.

There is a faster way. Instead of describing the result, you show one. This habit is called few-shot prompting, and it is the single biggest upgrade most people skip.

Why Words Fail and Examples Win

Try to describe your own writing voice in words. Go ahead.

"Friendly but professional. Warm, not stiff. Clear and direct, but not blunt." See the problem? Those words mean something different to everyone, including the AI.

A description points at a target. An example is the target.

When you show the AI a finished sample, it does not have to guess what "warm" means to you. It reads the rhythm, the word choices, the punctuation, the length. Then it copies the pattern.

This is how people learn too. You would not explain how to fold a paper airplane in five sentences. You would fold one and let them watch. Same idea here.

In Part 3 we gave the AI a role. Now we give it a model to copy.

See the Difference

Here is a real before-and-after. The task is writing a short, friendly reminder email.

The "before" version leaves "polite and professional but not stiff" up to the AI's imagination. You might get something formal and cold. You might get something too casual.

The "after" version removes the guessing. The AI sees the cheerful opener, the short lines, the "no rush" softener. It produces a new email that feels like the same person wrote it.

You did not write more rules. You wrote one good example.

How to Add an Example in 30 Seconds

The habit is simple. You sandwich an example between your task and your real request.

That label matters. It tells the AI what to do with the example. Without it, the AI might think the example is part of the content you want it to use.

Here is the pattern in a copy-paste shape:

Task: Write a product description for a coffee mug.

Match the style of this example:
"Meet your new morning ritual. This ceramic tumbler keeps coffee hot for hours and feels great in your hand. Dishwasher safe, drip free, and built to last."

Now write one for a stainless steel water bottle.

Notice the example is short. Three sentences. It shows the punchy opener, the benefit-first lines, the casual close. That is all the AI needs.

One Example for Style, Two for Tricky Cases

Most of the time, one example does the job. Reach for a second one when the task has more than one situation to handle.

Say you are sorting customer messages into "complaint," "question," or "praise." One example only shows the AI how to label one type. Two or three examples show it the full range.

Sort each message into: complaint, question, or praise.

Examples:
"My order arrived broken." → complaint
"Do you ship to Canada?" → question
"Love this product, thank you!" → praise

Now sort these:
1. "When will my refund show up?"
2. "Best purchase I made all year."

Each example covers a different case. Now the AI knows what each label looks like. This is where a second and third example earn their keep.

Use Your Own Past Work

Here is the best part of this habit. You already have examples sitting in your sent folder and your documents.

Want the AI to write in your voice? Paste two or three sentences you actually wrote. Then say "match this style." The AI copies your rhythm far better than it can follow a written description of your tone.

Here is how I write, for reference:
"Quick update on the project. We hit a small snag with the vendor, but nothing we can't handle. I'll have the fixed timeline to you by Friday."

Using that same voice, write a short message telling my team the launch is moving to next week.

This trick turns every good thing you have ever written into a teaching tool. Keep a few favorites handy.

One caution. Do not paste anything private into a public AI tool. Swap out client names, account numbers, or personal details first. We cover safe verification more in Part 7.

When Examples Are Overkill

This habit is powerful, but it is not for every prompt.

Skip the example when the task is common and the format is obvious. "List five gift ideas for a coffee lover" does not need a sample. The AI already knows what a list looks like.

The rule of thumb is easy. If you have asked twice and still get the wrong shape, stop adding rules. Add an example instead.

Build the Example Into Your Tools

Once you find an example that works, you do not want to retype it every time. That is wasted effort.

This is where saving your best prompts pays off. Our template builder lets you store a prompt with the example baked right in, so the pattern travels with you. You will turn your best prompts into reusable templates properly in Part 8.

You can also paste a prompt with examples into the free prompt scorer to see if it is clear enough before you rely on it.

The point is this. A great example is too valuable to use once and lose.

Try This Today

Pick one task you do often. An email, a summary, a caption, a label.

Find one finished example of it done right. It can be something you wrote or something you admire. Keep it short.

Then build the sandwich: task line, "match this example," your sample, your real request.

Notice how much shorter the example version is, and how much closer it lands. That gap is the whole lesson. Showing beats telling, almost every time.

Keep going

Next → Part 5: Control the Shape of the Answer

Or see the full Prompting Pro in 21 Days series.