Prompt Engineering

ELI5 (Explain like I’m 5)

Imagine a skilled actor given a vague instruction: “Perform this scene.”

They could interpret it in many ways. But when a director adds guidance like, “The character is going through a personal crisis. Make it tense. Speak slowly. Show hesitation.” the performance becomes more aligned with the story.

Prompt engineering works similarly.

The AI model already has ability, but clear direction helps shape the result into something focused and useful.

What is Prompt Engineering?

Prompt engineering involves iteratively designing, testing and improving prompts given to LLMs to minimize errors, hallucinations, and irrelevance.

Key Techniques

There are multiple techniques to refine a prompt. Some of them are discussed below.

1. Give structured instructions

This technique involves 

• giving the right level of detail
• adding context
• specifying constraints (tone, length of response, format). 

This is particularly important in business workflows where brand consistency is important. 

Example: Generate a 100-word product description for a bamboo mechanical keyboard using a sophisticated tone, specifically highlighting the silent switches and formatting the output as a headline followed by three bullet points.

Constraints: 100-word, sophisticated tone, headline followed by three bullet points.

Adding context: product description for bamboo mechanical keyboard

Right level of Detail: specifically highlighting the silent switches 

2. Giving Examples (few-shot patterns)

Include 2-5 examples to demonstrate desired output format.

Example:

"Convert the given boring headline into 'Clickbait' style.

'Stock market up' ->

Follow the pattern given in the below examples.

1. 'It rained today' -> 'You Won't Believe This Massive Rainstorm!'
2. 'The cat ate' -> 'This Hungry Feline Just Did Something Incredible!'
3. 'New phone launch' -> 'The Secret Feature of the New Phone Everyone is Talking About!'

Here 3 examples have been given to the model and it is being asked to convert the given headline on similar lines.

3. Role and audience framing

In this method, you ask the model to think like someone and answer from that person’s perspective. The responses will vary depending on the role you specify.

Below are 2 different roles with the same prompt. The answers will vary significantly. (Try these prompts on Geekflare Connect).

You are a professional fitness coach. Explain the importance of eating vegetables.

You are a school teacher. Explain the importance of eating vegetables.

Common misconceptions about Prompt Engineering

• Prompt engineering is not retraining the model. It does not change weights or add new knowledge.
• It is not about tricking the model. It is about clarity and consistency.
• It does not guarantee correctness. Even a well-designed prompt cannot override training gaps or model limitations.