L

Learn AI

Track progress · learn offline

Open

Lesson 2 of 8

The example ladder

6 min read

You've described exactly what you want and still get the wrong shape back. There's a move that skips describing altogether: instead of explaining the pattern, show it — one, two, a few finished examples — and let the AI copy it.

From telling to showing

Every prompt sits somewhere on a ladder. [Zero-shot](glossary://zero-shot) is what most people do — just a description, no examples. One-shot adds a single worked example. [Few-shot](glossary://few-shot) gives two or more. Each rung trades a little of your effort for a lot more control: the AI stops guessing the format and copies the pattern it can see in your examples.

Examples are a dial, not a switch. Zero-shot describes, one-shot shows once, few-shot shows several — and more clean examples means less guessing about the shape you want.

How many, and how clean

Two rules of thumb. First, you don't need many — around two to five usually does it; past a handful the gains fade. Second, the AI copies your examples faithfully, surface and all: if they're inconsistent, or their style doesn't match the real input, it imitates the wrong thing. Clean, on-pattern examples beat a pile of sloppy ones.

A few clean, consistent examples is the sweet spot. The AI imitates exactly what you show — including the flaws — so a mismatched or sloppy example teaches it the wrong pattern.

One example is one-shot, several is [few-shot](glossary://few-shot), none is [zero-shot](glossary://zero-shot). Same idea, dialled up or down: how many worked examples you hand over before the real task.

The gist

You need every support ticket tagged in one exact set of categories, in a strict format. What's the most reliable prompt?

Continue in the app

Take the whole Prompt Engineering course — tracked

Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.