Show the AI What 'Good' Looks Like Before You Ask

Stop describing the answer you want. Show the AI one example of it, and watch the quality jump.

You ask for a "professional email." You get something stiff and generic. You ask for a "catchy headline." You get five that all feel flat.

The problem is not the AI. It is that "good" lives in your head, and the model cannot see it.

Today's habit fixes that. You learn to anchor the AI with a target before you ask. Show it what good looks like, and you stop rolling the dice on every reply.

Why "good" is invisible to the AI

When you ask for something "professional" or "engaging," you are using shortcut words. In your mind, those words carry a clear picture. You know the tone, the length, the vibe.

The AI does not share that picture. To a large language model, "professional" could mean a hundred different things. It guesses. Sometimes the guess lands. Often it misses.

This is the gap. You hold a clear target. The model gets a fuzzy word. So it fills the blank with the most average answer it can find.

The fix is not more adjectives. Piling on words like "polished, sharp, modern, clean" does not help. You are still describing a feeling, not showing a result.

What works is giving the AI something concrete to match. An example. Or a clear list of what a finished answer needs. We will cover both.

Anchor with one example

The strongest move is to show the AI a real example of the output you want. This is called anchoring. You give the model a target it can copy in shape, tone, and depth.

Here is the difference an example makes.

The first prompt gets a guess. The second gets something that sounds like you, because the AI has a model to follow.

Notice what the example does. It shows the greeting style, the warm-but-brief tone, the soft ask, and the friendly close. You did not describe any of that. You showed it. One example carried more than five sentences of instructions could.

When you use an example, label it clearly. Tell the model it is a target for style and structure, not content to summarize. A single line does it:

Here is an example of the style I want. Match the tone and format,
but write about my topic, not this one.

That line keeps the AI from copying your sample word for word. It borrows the shape and brings your content.

Define "done" when you have no example

Sometimes you do not have a sample handy. That is fine. You can still anchor the AI by writing a "definition of done."

A definition of done is a short checklist of what a finished answer must include. Think of how a good manager hands off a task. They do not say "make it good." They say what good looks like.

Here is a request with a clear definition of done built in:

Write a LinkedIn post announcing our new feature.

A good version of this post:
- Opens with a hook in the first line (no "We're excited to announce")
- Is 4 to 6 short paragraphs
- Explains one clear benefit, not a feature list
- Ends with a question to spark comments
- Uses a warm, plain tone with no buzzwords

Each line removes a guess. The model now knows the length, the structure, the tone, and the traps to avoid. It aims at your target instead of the average post.

See the pattern. The left side is a feeling. The right side is something the AI can check its work against.

A quick template you can reuse

You do not need to reinvent this every time. Here is a simple shape that works for most requests. Fill in the blanks.

Task: [what you want]

Definition of done — a good answer:
- [requirement 1]
- [requirement 2]
- [requirement 3]

[Optional] Example of the style I want:
"[paste a sample, then say: match the tone, use my topic]"

Use the example block when you have one. Use the definition-of-done block when you do not. Use both when you want the tightest control.

This habit pairs well with the next ones in the series. In Part 4 you will go deeper on examples, the "show, don't tell" move. For now, one example or one checklist is plenty.

How to find your example fast

The best example is one you already made. You just have to know where to look.

Keep a small stash of your strongest outputs. The email that got a quick reply. The summary your boss loved. The product description that converted. These are gold.

Over a few weeks, you build a personal library of "good." Each new request starts from a proven target, not a blank page. If you want a place to store and reuse these, the prompt library keeps your best work one click away.

Avoid these two anchoring mistakes

This habit is simple, but two slips can trip you up.

Mistake one: a messy or off-target example. The AI copies what you show it. If your sample is rambling or wrong in tone, you will get more of the same. Pick an example that truly reflects what you want. If you only have a rough one, say which parts to keep and which to ignore.

Mistake two: an example with no instructions. If you paste a sample and say nothing, the model may summarize it, copy it directly, or get confused. Always tell it the example's job.

When you anchor well, you also make it easy to judge the result. If you wrote a definition of done, you can score the answer against it. Did it hit every point? That check is the seed of a habit we cover later in the series.

Try this today

Pick one task you ask the AI for often. An email, a summary, a caption, anything routine.

You will feel the difference right away. The anchored version needs less fixing. That saved cleanup time is the whole point of this habit.

Once you have a version you like, run it through the prompt scorer to see where it still has slack. A higher score usually means fewer guesses left on the table.

Anchoring is the habit. Show the target, then ask. Tomorrow we give the AI something even more powerful than an example: a clear job to do.

Keep going

Next → Part 3: Give the AI a Job, Not Just a Question

Or see the full Prompting Pro in 21 Days series.