Prompting 104: ตัวอย่างชนะคำอธิบายPrompting 104: examples beat explanation

If you cannot describe the style, show it.

AI is good at pattern matching. One useful example can replace a long explanation.

This is called few-shot prompting when you give the model a few examples of input and output.

Why examples help

Instructions explain rules.

Examples show taste.

Compare these two prompts:

Write in a friendly professional tone.

and:

Use this tone:
"Thanks for flagging this. I checked the file and found one issue we should fix before sending it out."

The second prompt is clearer because the model can imitate the pattern.

Use before and after

For rewriting tasks, before-and-after examples are powerful.

Rewrite the next message using the same kind of improvement shown below.

Before:
Please be informed that the meeting has been postponed due to scheduling conflicts.

After:
The meeting has been moved because several people are unavailable at the original time.

Now rewrite:
[paste text]

The example teaches what "better" means.

Use format examples

If you need a table, checklist, or structured answer, show the exact shape.

Output format:

| Issue | Why it matters | Suggested fix |
|---|---|---|
| [short issue] | [impact] | [specific fix] |

This is often better than saying "make a table" because the model knows the columns.

Use positive and negative examples

Sometimes the model needs to know what not to do.

Good:
"Could you send the updated file by Friday?"

Too harsh:
"You need to send this immediately."

Too vague:
"Please handle this soon."

Positive and negative examples teach the boundary.

Keep examples diverse

If all examples are too similar, the model may copy superficial details.

Weak examples:

• all three are about email
• all three use the same length
• all three have the same sentiment

Better examples:

• one short email
• one announcement
• one customer reply
• different lengths
• same style principle

The model should learn the rule, not the accident.

Practice: style sample

Pick one writing style you use often.

Create a prompt with:

Task:

Style examples:
Good:

Too formal:

Too casual:

Text to rewrite:

Then test it with two different texts. If it only works for one, improve your examples.

References

• OpenAI Prompt Engineering describes few-shot learning as steering a model with input/output examples rather than fine-tuning.
• Google Gemini Prompt Design Strategies recommends examples and prefixes to make output easier to guide and parse.
• Sahoo et al., 2024: Systematic Survey of Prompt Engineering covers example-based prompting as a core prompt engineering family.