09 - Prompt Engineering – Part 2: Understanding Prompts: Components, Types & Characteristics

Understanding Prompts: Components, Types & Characteristics

Post 2 – Introduction to Prompt Engineering Series

Recap from the Previous Post

In our first post , we kicked off the series by introducing Prompt Engineering — the art and science of crafting effective instructions to guide AI models like LLMs. We saw how clear, structured prompts can significantly influence the quality of the responses. Real examples from Gemma 1B showed how even a small change in phrasing or context can completely shift the model’s output.

Now that we know why prompts matter, let’s explore what makes up a good prompt and how different styles and strategies come into play.

Today’s Focus: The Building Blocks of Prompts

In this post, we’ll cover three key areas that every prompt engineer (or GenAI user) should understand:

Components of Prompts – What elements make a prompt effective?
Types of Prompts – How can prompts vary based on goal or technique?
Characteristics of Good Prompts – What makes a prompt clear, reliable, and impactful?

1. Components of Prompts

While Prompt Engineering isn’t bound by a formal syntax like programming languages such as Python or C, it’s clear that some prompts consistently perform better than others. So, what makes a great prompt stand out?

In my experience, the difference lies in understanding how LLMs are trained. These models are not just fluent in language — they’re trained on massive corpora filled with patterns, structures, styles, and semantics from real-world data. That means they’ve learned not just words, but how information is typically organized and expressed.

So, the better we understand how these models think, the better we can communicate with them. And over time, through hands-on experimentation and observation, it becomes clear that certain prompt structures are more effective. These structures align better with how models process intent, context, and expected outcomes.

Prompt Engineering, then, becomes less about rigid rules and more about adapting your communication style to work with the model — using clarity, context, structure, and feedback to guide the interaction.

Below are components of prompt and every prompt will only use partial list of these components as appropriate.

i. Instruction

The core task or action you want the AI to perform. This should be clear, direct, and ideally start with a verb.

Example:
“Identify three types of credit risks involved in unsecured personal loans.”

ii. Context

Provides background or situational information to guide the model’s interpretation. Without this, the AI may respond too generically.

Example:
“Assume you're analyzing loans disbursed by an Indian retail bank during a period of rising interest rates.”

iii. Input Data

This is the raw content the AI is supposed to work with. It could be a paragraph, a list, a table, or even a dataset snippet.

Example:
“Customer A has missed 2 EMIs in the last 6 months but has a CIBIL score of 780 and no prior defaults.”

iv. Output Format

Tells the AI how to structure the response. This is especially useful when consistency or readability is important.

Example:
“Present the response in a table with columns: Risk Type, Description, Risk Level.”

v. Role / Persona

Assigns a specific perspective or expertise to the AI. This sets expectations for tone and detail level.

Example:
“Act as a senior credit risk analyst working for a private sector bank.”

vi. Examples (Few-Shot Learning)

You provide one or more example inputs and outputs to teach the model your desired format or reasoning approach.

Example:
“Example:
Risk Type: Credit Risk
Description: Risk due to borrower’s inability to repay
Risk Level: High”

vii. Constraints

Set limits or exclusions to avoid ambiguity or misinterpretation.

Example:
“Avoid technical jargon. Keep each description under 20 words.”

viii. Tone / Style

This can subtly control how formal or creative the AI’s response should be.

Example:
“Use a professional but easy-to-understand tone, suitable for a business presentation.”

ix. Goal / Intent

Clarifies the end purpose — what you’re ultimately trying to achieve from this prompt.
Example:
“This is for a risk review report being shared with senior management.”

Holistic Prompt Example

Now let’s bring it all together with a complete, structured prompt using all of the above components:

Prompt (Complete Example):
Act as a senior credit risk analyst working at a private sector bank. You are preparing a risk review report for senior management. Assume you're analyzing personal loan disbursals during a time of rising interest rates in India. Consider the following input:
Input:
“Customer A has missed 2 EMIs in the last 6 months, has a CIBIL score of 780, stable job history, and no previous defaults.”
Identify three potential credit risks for this customer. Present your answer in a table with the following columns: Risk Type, Description (under 20 words), and Risk Level (Low, Medium, High). Use a professional but easy-to-understand tone. Avoid technical jargon.
Example:
Risk Type: Credit Risk
Description: Risk due to borrower’s inability to repay

Risk Level: High

Response from ChatGPT as is below

With this kind of complete and structured prompt, you’re giving the model everything it needs to succeed — instructions, data, format, tone, and the ultimate goal.

2. Types of Prompts

Prompting isn’t one-size-fits-all. Based on your goal, context, and complexity, you might use different types of prompts. Here’s a closer look at each type, with use cases and a consistent banking risk example so you can see how they vary.

i. Zero-Shot Prompting

This is the simplest form — no examples, just a plain instruction. The model has to figure everything out on its own based on its training. Powerful model are good at proper response with zero-shot, but it is observed that responses are not very consistent.

Use When: Task is simple or well-understood by the model.
Pros: Fast, clean, minimal setup.
Cons: May lack precision or consistency.

Example Prompt:

“List three types of financial risks in retail banking.”

ii. Few-Shot Prompting

You give the model a few examples to guide its output. It learns the pattern and follows suit. It is observed that every model performed better with single shot or few shots. It also setup tone of responses and output format.

Use When: The task is complex or format-sensitive.
Pros: Produces more accurate and structured responses.
Cons: Needs careful example selection.

Example Prompt:

“Identify financial risks using the format shown below.
Example:
Risk Type: Credit Risk
Description: Risk from borrower default
Now continue with two more examples.”

iii. Chain-of-Thought Prompting

Here you ask the model to explain its reasoning step-by-step instead of jumping to the final answer. Great for analytical tasks.

Use When: You want transparent reasoning, multi-step logic, or decisions.
Pros: Improves interpretability and correctness.
Cons: Longer, sometimes more verbose output.

Example Prompt:

“Explain the steps you would follow to assess credit risk in a new customer applying for a personal loan.”

iv. Role-Based Prompting

You assign the AI a specific persona or role, like a domain expert or decision-maker. This helps the model align with the tone, depth, and logic expected from that role.

Use When: The task needs subject-matter expertise or perspective.
Pros: More realistic and relevant output.
Cons: May need context to behave accurately.

Example Prompt:

“You are a senior risk manager in a private bank. Identify key operational risks in loan disbursement processes.”

v. Instructional Prompting

A direct, task-based command that often combines clarity and conciseness. Similar to zero-shot, but more structured.

Use When: You have a clear goal and don’t need detailed setup.
Pros: Fast and effective when well-written.
Cons: Needs to be very precise.

Example Prompt:

“Summarize key risks in digital loan approval workflows in 3 bullet points.”

Each type of prompt frames the same topic differently, guiding the model in how it thinks and responds. The best type depends on your objective: speed, detail, accuracy, or style.

3. Characteristics of a Good Prompt

A “good” prompt isn't just functional — it’s clear, focused, and effective. Think of it as giving instructions to a smart intern who needs the right framing to do their best work. Here’s what makes a prompt powerful:

i. Clarity

The prompt should be easy to understand — no vague language, double meanings, or hidden assumptions. Tell the AI exactly what you want.

Bad: “Tell me about risk.”
Good: “List three common types of risk in unsecured lending.”

ii. Specificity

Define the scope, depth, and format. Ambiguity leads to poor results.

Bad: “Write a report.”
Good: “Write a 100-word summary of credit risk in retail lending, in bullet format.”

iii. Context-Rich

Providing relevant background helps the model tailor its answer.

Bad: “What is operational risk?”
Good: “In the context of digital loan disbursement by Indian fintechs, what is operational risk?”

iv. Structured Format

Specify how the answer should be delivered — table, bullets, paragraph, etc.

Bad: “List key risks.”
Good: “Present key risks in a table with columns: Risk Type, Description, Mitigation.”

v. Persona or Role

Give the model a role to play — it helps with tone and relevance.

Bad: “Explain credit risk.”
Good: “Act as a risk manager at a mid-size bank and explain credit risk to junior staff.”

vi. Goal-Oriented

Make the purpose clear — is it for a report, a chatbot, or decision support?

Bad: “What are the risks here?”
Good: “This analysis is for a risk committee review — summarize top risks with their impact.”

vii. Concise

Be detailed but avoid overloading with too many instructions. Balance is key.

Bad: Overloading the prompt with every possible instruction.
Good: Use a few clear sentences and examples instead of a wall of text.

viii. Testable / Refined

Prompts should be tweakable and measurable. You should be able to improve them based on the AI’s output.

Tip: Run, observe, revise — prompting is iterative.

Holistic Prompt Example (Bringing it All Together)

Here’s a complete example using all these characteristics, in the banking risk context:

Prompt:
“Act as a senior credit risk analyst at a private bank in India. You are preparing a short report for the risk committee on unsecured personal loans. Focus on digital loan applications during rising interest rates.

Identify the top 3 financial risks, describe each in under 25 words, and present your answer in a table with columns: Risk Type, Description, Severity (Low/Medium/High). Avoid jargon and keep the tone professional. This summary will be used in a board-level review.”

This prompt includes:

Clarity: Direct task: identify risks
Specificity: 3 risks, 25-word limit
Context: Digital loans, rising interest rates, Indian bank
Structure: Table with 3 defined columns
Persona: Senior credit risk analyst
Goal-Oriented: Used in a board report
Concise: Clear, short sentences
Refinable: Can easily tweak severity scale or output length

Wrapping Up

These foundations — components, types, and characteristics — are essential to building effective prompts that deliver precise and useful results. Whether you’re analyzing loan defaults, detecting fraud, or writing policy summaries, prompt quality defines output quality.

In the next post, we’ll start breaking down each prompting strategy with deep dives, hands-on examples, and industry-specific use cases.

Which of these areas would you like to explore in more detail first? Drop a comment!

Stay tuned — we’re just getting started.

✍️ Author’s Note

This blog reflects the author’s personal point of view — shaped by 22+ years of industry experience, along with a deep passion for continuous learning and teaching.
The content has been phrased and structured using Generative AI tools, with the intent to make it engaging, accessible, and insightful for a broader audience.