AI can write an article, analyse a company, generate code, summarise a report, create an image or even help build an application.

But there is a catch.

The quality of the output is heavily influenced by the quality of the instruction you give the AI.

That instruction is called a prompt, and the process of designing better prompts is known as prompt engineering.

For investors, entrepreneurs, analysts and professionals using AI regularly, prompt engineering is becoming a practical skill rather than an obscure technical concept.

You don’t necessarily need to learn programming to become good at it.

You need to learn how to communicate with AI precisely.

This guide explains how prompt engineering works, the principles behind effective prompts, the major prompting techniques, common mistakes, and a practical framework you can use with ChatGPT, Claude, Gemini and other AI models.

What Is Prompt Engineering? 🤖

Prompt engineering is the process of designing and refining instructions given to an AI model to obtain a desired output.

A prompt can be as simple as:

“Explain inflation.”

Or considerably more sophisticated:

“Act as an Indian equity research analyst. Analyse the impact of persistent inflation on FMCG companies. Focus on pricing power, gross margins, volume growth and valuation. Present the analysis in a table and conclude with the key risks.”

Both prompts ask about inflation.

But the second prompt gives the AI considerably more information about:

• Who it should behave like

• What it should analyse

• Which factors matter

• Which market it should focus on

• What the output should look like

The result is likely to be much more useful.

This is the fundamental idea behind prompt engineering:

Don’t merely ask AI a question. Design the task.

Why Prompt Engineering Matters 🎯

Large language models are extraordinarily capable, but they do not automatically know exactly what you want.

Consider asking an AI:

“Analyse Reliance Industries.”

What exactly should it analyse?

Its stock price?

Its valuation?

Jio?

Retail?

Oil-to-chemicals?

Debt?

Cash flow?

Competitive position?

Long-term investment potential?

Different interpretations can produce completely different answers.

Now consider:

“Analyse Reliance Industries from the perspective of a long-term Indian equity investor. Evaluate business quality, revenue diversification, profitability, free cash flow, debt, competitive advantages, major risks and valuation. Separate factual observations from interpretation and conclude with five questions an investor should investigate before investing.”

The second prompt dramatically reduces ambiguity.

The basic relationship

Vague Prompt
     ↓
More Interpretation Required
     ↓
Higher Probability of Unwanted Output
     ↓
More Editing

Structured Prompt
     ↓
Less Ambiguity
     ↓
Better Alignment
     ↓
More Useful Output

Prompt engineering therefore isn’t about making prompts unnecessarily long.

It is about making them appropriately specific.

Start With Direction: The First Line Matters 🎯

One of the most useful practical lessons in prompt engineering is that the opening instruction establishes the direction of the task.

Think of the first line as giving the AI a destination.

Compare:

“Analyse this company.”

with:

“Act as a long-term equity analyst evaluating this company for a 10-year investment horizon.”

The second prompt immediately establishes a direction.

The AI now has an idea of:

• The perspective to adopt

• The type of analysis required

• The time horizon

• The likely audience

• The nature of the desired answer

You can then add the details:

“Act as a long-term equity analyst evaluating this company for a 10-year investment horizon. Analyse business quality, financial performance, competitive advantages, valuation and risks.”

The first line establishes the direction. The subsequent instructions provide the navigation.

A useful mental model is:

Opening Instruction
       ↓
Direction
       ↓
Context + Task
       ↓
Constraints
       ↓
Output Format
       ↓
Final Response

This doesn’t mean the first line is magically more important than everything else.

A poor task, missing context or inadequate data can still produce a poor result.

But when designing a complex prompt, start by making the objective and direction unmistakably clear.

The Anatomy of a Good Prompt 🧩

A useful prompt can be thought of as a collection of components.

One practical framework is:

ROLE
  ↓
CONTEXT
  ↓
TASK
  ↓
INPUT
  ↓
ASSUMPTIONS
  ↓
CONSTRAINTS
  ↓
OUTPUT FORMAT
  ↓
OUTPUT STYLE
  ↓
OUTPUT RESTRICTIONS

Not every prompt needs every component.

A simple question may require only a task.

A complex business or research task may benefit from almost all of them.

Let’s examine each component.

1. Role — Who Should the AI Be? 👤

A role establishes the perspective from which the AI should approach the task.

For example:

“Act as a financial analyst.”

or:

“Act as a senior Python developer.”

or:

“Act as an investor educator explaining concepts to an intelligent beginner.”

Role instructions can be useful when the task requires a particular professional perspective.

Example

Instead of:

“Analyse this company.”

Try:

“Act as a long-term equity analyst evaluating this company for a 10-year investment horizon.”

The second prompt establishes a lens.

However, role prompting should not be treated as magic.

Simply saying “You are the world’s greatest financial analyst” doesn’t automatically make the answer better.

The role should be connected to the actual task.

2. Context — Give AI the Information It Needs 📚

Context is often one of the most important components of a complex prompt.

AI cannot reliably infer information that you haven’t provided.

Suppose you ask:

“Which stock is better?”

Better for whom?

A retiree?

A 25-year-old?

A trader?

A dividend investor?

Someone investing for 20 years?

The AI needs context.

For example:

“The investor has a 15-year investment horizon, prefers dividend-paying businesses, wants relatively low leverage and is willing to accept moderate volatility.”

Now the question becomes much more meaningful.

Context can include:

• Background information

• User objectives

• Existing data

• Previous decisions

• Business requirements

• Target audience

• Investment horizon

• Risk tolerance

• Technical environment

• Relevant documents

A useful rule is:

If the AI needs to know something to make a good decision, consider putting it in the prompt rather than expecting it to guess.

3. Task — Tell AI Exactly What You Want Done 🎯

The task is the action you want the model to perform.

Examples:

• Analyse

• Compare

• Summarise

• Classify

• Extract

• Rewrite

• Generate

• Explain

• Critique

• Rank

• Translate

• Develop

• Recommend

Compare:

“Indian banking stocks.”

with:

“Compare the five largest Indian private-sector banks on profitability, asset quality, capital adequacy, valuation and long-term growth potential.”

The second prompt contains an actual task.

A strong task answers:

What should the AI do?

Not merely:

What topic should the AI talk about?

This distinction is surprisingly important.

4. Input — Give the AI Something to Work With 📊

The input is the information on which the AI should perform the task.

For example:

“Analyse the following quarterly results: [data]”

Or:

“Rewrite the following paragraph for an investment audience: [text]”

Or:

“Classify these companies into cyclical, defensive and structural-growth businesses: [list]”

For complex tasks, clearly separating the instructions from the input can reduce confusion.

For example:

<task>
Analyse the companies based on profitability and valuation.
</task>

<input>
Company A...
Company B...
Company C...
</input>

Structured delimiters such as XML-style tags can help models distinguish instructions from supplied information, particularly in longer prompts.

5. Assumptions — Define What AI Should Assume 💡

Sometimes information is missing.

Instead of allowing the AI to make arbitrary assumptions, specify them.

For example:

“Assume the investor has a 10-year investment horizon.”

or:

“Assume the company operates primarily in India.”

or:

“Assume the reader understands basic financial statements but has limited knowledge of valuation.”

This becomes particularly important for financial analysis.

Suppose an AI is asked:

“Is this company expensive?”

Expensive relative to what?

You could specify:

• Historical valuation

• Industry peers

• Earnings growth

• Free cash flow

• Replacement cost

• Expected future earnings

The assumption or comparison framework dramatically changes the answer.

6. Constraints — Tell AI What It Must Respect ⚠️

Constraints define the boundaries of the task.

Examples:

“Keep the answer below 1,000 words.”

“Use only the information provided.”

“Do not invent financial figures.”

“Use Indian examples.”

“Return the result as a table.”

“Do not recommend a stock without explaining the risks.”

Constraints are especially useful when the task has multiple requirements.

A good constraint prevents the AI from wandering outside the intended scope.

7. Output Format — Tell AI What the Answer Should Look Like 📋

One of the easiest ways to improve AI output is to specify the format.

Compare:

“Analyse this company.”

with:

“Analyse this company using the following structure:

1. Business quality

2. Financial performance

3. Competitive advantages

4. Valuation

5. Risks

6. Final assessment”

The second prompt gives the AI a destination.

Common output formats

The output format is not merely cosmetic.

It can change how the model approaches the task.

8. Output Style — Define How It Should Communicate ✍️

The same information can be presented in dramatically different ways.

You could ask for:

• Beginner-friendly

• Professional

• Academic

• Conversational

• Analytical

• Concise

• Detailed

• Technical

• Investor-focused

For example:

“Explain this like a financial analyst speaking to an intelligent retail investor.”

is very different from:

“Explain this to a 12-year-old.”

Both can be correct.

The desired audience determines the appropriate communication level.

9. Output Restrictions — Define What You Don’t Want 🚫

Sometimes you need to explicitly control the output.

For example:

“Do not include an introduction.”

“Do not repeat the question.”

“Use no more than five bullet points.”

“Do not make unsupported claims.”

“Do not include citations in the final published article.”

However, there is an important nuance.

Positive instructions are often more useful than simply listing prohibitions.

Instead of:

“Don’t be vague.”

Try:

“Support each conclusion with a specific reason or example.”

Instead of:

“Don’t write a long answer.”

Try:

“Limit the answer to approximately 500 words.”

This gives the model a positive target.

Let AI Ask Clarifying Questions 🤔

Sometimes the problem isn’t that your prompt is poorly written.

The problem is that you haven’t provided enough information to answer it properly.

In such cases, don’t force the AI to guess.

Tell it that it is allowed to ask questions.

For example:

“If any information required to complete the task is missing or ambiguous, ask clarifying questions before proceeding. Do not make important assumptions without confirmation.”

Now the interaction can become:

Prompt
   ↓
AI identifies ambiguity
   ↓
Clarifying Questions
   ↓
User Provides Information
   ↓
AI Performs Task
   ↓
Answer

Example

Instead of:

“Analyse this stock.”

AI could ask:

“Before I analyse it, could you clarify:

1. What is your investment horizon?

2. Are you primarily interested in business quality or valuation?

3. Should I compare it with peers?”

This can produce a much more relevant analysis.

But don’t overdo it

You don’t want AI asking questions simply because it can.

A good instruction is:

Ask clarifying questions only when missing information would materially affect the quality or correctness of the answer. Otherwise, proceed using reasonable assumptions and state them clearly.

This gives AI a useful balance between asking too much and assuming too much.

Tell AI What to Do When Information Is Missing ⚠️

Another useful prompting technique is explicitly defining how the AI should handle missing information.

For example:

“If reliable information is unavailable, state that it is unavailable rather than guessing.”

Or:

“If the supplied data is insufficient to reach a conclusion, explain what additional information is required.”

This is particularly valuable in financial analysis, research and data extraction.

Compare:

“Analyse the company’s debt.”

with:

“Analyse the company’s debt using the supplied financial statements. If the required figures are unavailable, identify the missing information rather than estimating it.”

The second prompt establishes a failure behaviour.

That’s an underrated part of prompt engineering.

A good prompt doesn’t only describe what the AI should do when everything goes well.

It also tells the AI what to do when things go wrong.

The Power of Examples: Zero-Shot vs Few-Shot Prompting 🧪

One of the most useful prompting techniques is providing examples.

Zero-shot prompting

You give the model an instruction without an example.

“Classify the following companies as defensive or cyclical.”

Few-shot prompting

You provide examples first.

“Classify companies as defensive or cyclical.

Example:
Nestlé India → Defensive
Tata Motors → Cyclical

Now classify:
Hindustan Unilever
Mahindra & Mahindra”

The examples demonstrate the pattern.

Why examples are powerful

Instructions tell the AI:

“What I want.”

Examples show:

“What good looks like.”

That distinction is extremely useful.

One-Shot, Few-Shot and Multi-Shot Prompting

These terms simply describe how many examples you provide.

More examples are not automatically better.

If one excellent example completely communicates the required pattern, ten examples may simply consume context unnecessarily.

Use enough examples to remove ambiguity — not enough to create noise.

XML Tags: A Surprisingly Useful Prompting Technique 🏷️

If you have worked with AI extensively, you may have noticed that structured prompts often use tags such as:

<role>
You are a financial analyst.
</role>

<context>
The investor has a 15-year horizon.
</context>

<task>
Analyse the company.
</task>

<constraints>
Use Indian market data.
</constraints>

<output_format>
Return the analysis in five sections.
</output_format>

This isn’t because the AI is literally executing XML code.

The tags act as semantic boundaries.

They make the prompt easier to organise and can help distinguish instructions, context, inputs and expected outputs.

When should you use XML?

XML-style prompting is particularly useful when the prompt contains:

• Multiple instructions

• Large amounts of context

• Examples

• Constraints

• Multiple input sections

• Complex output requirements

For a simple question such as:

“What is ROE?”

XML would probably be overkill.

Prompt Chaining: Don’t Make One Prompt Do Everything 🔗

Complex tasks can sometimes be divided into multiple prompts.

Imagine you want to analyse 100 annual reports.

Instead of asking an AI to:

Read all reports → extract data → analyse → compare → rank → write report

you could create a pipeline:

Annual Reports
      ↓
Data Extraction
      ↓
Data Validation
      ↓
Financial Analysis
      ↓
Company Comparison
      ↓
Ranking
      ↓
Investment Report

This is called prompt chaining.

Each stage performs a specific task.

The output of one stage becomes the input for the next.

This approach can make complicated workflows easier to control, test and debug.

Prompt Engineering vs Context Engineering 🧠

There is an important evolution happening in AI.

Early discussions focused heavily on:

“What is the perfect prompt?”

Modern AI applications increasingly involve something broader:

What information, tools and instructions should the model receive at each stage of the task?

Consider an AI investment analyst.

It may receive:

System Instructions
       +
Investor Profile
       +
Company Financials
       +
Annual Reports
       +
Recent News
       +
Market Data
       +
Tool Results
       ↓
      LLM
       ↓
Investment Analysis

The prompt is only one component.

The quality of the context supplied to the model can be just as important.

This is one reason modern AI applications increasingly combine models with search, databases, APIs, file retrieval and tools rather than relying solely on the model’s internal knowledge.

Prompt Engineering and RAG 🔎

This is particularly relevant to investors.

Imagine asking an AI:

“What happened to Indian banking stocks this week?”

A model’s internal knowledge may not contain the latest events.

A Retrieval-Augmented Generation system, or RAG, can retrieve current information and provide it to the model before generating the answer.

The architecture becomes:

User Question
      ↓
Retrieve Relevant Information
      ↓
Documents / Database / Web / API
      ↓
Relevant Context
      ↓
LLM
      ↓
Answer

Prompt engineering determines how the retrieved information should be used.

For example:

“Using only the retrieved documents, identify the three major factors affecting Indian banking stocks this week. Separate reported facts from your interpretation.”

That is much more controlled than simply:

“Tell me what happened.”

The Difference Between a Prompt and a Workflow ⚙️

This distinction becomes important as AI systems become more sophisticated.

A prompt might say:

“Analyse this annual report.”

A workflow might say:

1. Retrieve annual report.
2. Extract financial statements.
3. Identify revenue and profit trends.
4. Calculate relevant ratios.
5. Compare with previous years.
6. Identify risks.
7. Retrieve recent company developments.
8. Produce investment analysis.
9. Validate numerical consistency.
10. Generate final report.

The prompt controls one model interaction.

The workflow controls the entire process.

For serious AI applications, thinking in terms of workflows can be more powerful than endlessly refining a single prompt.

AI Can Improve Your Prompts 🤖🔄🤖

Here is another important development in prompt engineering:

You can use AI to improve the prompts you give to AI.

You don’t have to design every sophisticated prompt entirely by yourself.

For example, you can start with:

“I want AI to analyse an Indian company for a long-term investor.”

Then ask AI:

“Improve this prompt. Identify missing information, suggest useful constraints and create a structured prompt that will produce a more rigorous investment analysis.”

The AI can then help you:

• Identify ambiguity

• Add missing context

• Suggest useful metrics

• Improve the task definition

• Design an output structure

• Add constraints

• Create examples

• Remove unnecessary instructions

• Critique an existing prompt

This creates an iterative loop:

Human Objective
      ↓
Initial Prompt
      ↓
AI Improves Prompt
      ↓
Human Reviews
      ↓
Refined Prompt
      ↓
AI Produces Output
      ↓
Human Evaluates
      ↓
Further Refinement

This is perhaps one of the most practical lessons of modern prompt engineering.

You don’t have to become an expert prompt writer before using AI effectively.

You can make AI your prompt-engineering assistant.

The human still provides the objective, judgement and desired outcome.

AI can help translate that intention into better instructions.

Prioritise Your Instructions 🎯

Complex prompts sometimes contain many requirements.

For example:

“Analyse the company, compare peers, discuss valuation, include financial ratios, provide risks, keep it under 1,000 words, use a table and explain everything for beginners.”

What happens if some of these requirements conflict?

Tell the AI which requirements matter most.

For example:

“Accuracy and completeness are more important than keeping the response below 1,000 words. If necessary, exceed the word limit slightly to avoid omitting important analysis.”

This gives the AI a hierarchy.

You can even specify:

Priority:
1. Accuracy
2. Completeness
3. Clear reasoning
4. Output format
5. Conciseness

This can be particularly useful for complicated workflows.

Ask AI to State Its Assumptions 💡

When a task involves uncertainty, ask AI to make its assumptions visible.

For example:

“State any material assumptions you make before providing the analysis.”

This can reveal that the AI has assumed:

• A particular investment horizon

• A particular definition of “cheap”

• A particular peer group

• A particular interpretation of the data

Visible assumptions are much easier to evaluate than hidden assumptions.

For financial analysis, this is especially valuable.

Prompt Iteration: Your First Prompt Probably Won’t Be Perfect 🔄

One of the biggest misconceptions about prompt engineering is that experts somehow know the perfect prompt immediately.

In reality, prompting is often iterative.

Prompt
  ↓
AI Output
  ↓
Evaluate
  ↓
Identify Problem
  ↓
Modify Prompt
  ↓
AI Output
  ↓
Evaluate Again

Suppose your prompt produces overly generic investment analysis.

Don’t simply write:

“Give me a better answer.”

Diagnose the problem.

Perhaps the model lacks:

• Financial metrics

• Time horizon

• Peer comparison

• Valuation framework

• Risk requirements

• Output structure

Then modify the prompt accordingly.

This is much closer to engineering than to writing a magic sentence.

Prompt Evaluation: Treat Prompts Like Software 🧪

If you are building an AI application, manually looking at one impressive response isn’t enough.

You need to test the prompt repeatedly.

For example, suppose you create a prompt for extracting financial metrics from annual reports.

You could test it against:

• 10 companies

• Multiple sectors

• Different report formats

• Missing data

• Restated figures

• Unusual accounting disclosures

Then measure:

This moves prompt engineering from art toward engineering.

For production AI systems, this matters even more because model behaviour can change between model versions or snapshots. Evaluation helps determine whether a prompt continues to produce acceptable results.

A Practical Prompt Engineering Framework 🎯

For most moderately complex tasks, this framework works well:

The 9-Part Prompt Framework

<role>
Who should the AI act as?
</role>

<context>
What background does the AI need?
</context>

<objective>
What outcome are we trying to achieve?
</objective>

<task>
What exactly should the AI do?
</task>

<input>
What information should it work on?
</input>

<assumptions>
What should it assume?
</assumptions>

<constraints>
What rules or limitations apply?
</constraints>

<output_format>
What should the answer look like?
</output_format>

<output_style>
How should the answer communicate?
</output_style>

You don’t need to use every section every time.

The framework is a checklist for thinking, not a requirement to create enormous prompts.

A Before-and-After Example 📈

Weak prompt

“Analyse Tata Motors.”

Better prompt

“Analyse Tata Motors as a long-term investment for an Indian retail investor. Evaluate business quality, revenue and earnings trends, profitability, debt, cash generation, competitive advantages, major risks and valuation. Compare the company with relevant peers where appropriate. Distinguish factual observations from interpretation. Present the analysis under clear headings and conclude with five questions an investor should investigate before making an investment decision.”

Notice what changed.

The second prompt defines:

• Role

• Context

• Objective

• Task

• Analytical dimensions

• Comparison requirement

• Reasoning requirement

• Output structure

That’s prompt engineering.

A Real-World Investor Scenario 👨‍💼

Imagine Ravi is investing ₹10 lakh for the next 15 years.

He wants a portfolio of financially strong Indian companies.

A weak prompt might be:

“Give me 10 good stocks.”

The AI has almost unlimited freedom to interpret the question.

A better prompt might say:

“Act as a long-term Indian equity analyst. The investor has a 15-year horizon and prefers financially strong companies with sustainable cash generation. Identify 10 NSE/BSE-listed companies for further research. Evaluate profitability, leverage, cash flow, competitive advantages, earnings growth and valuation. Do not simply rank companies by recent share-price performance. For every company, provide the investment thesis, two major risks and three metrics the investor should monitor.”

Now the AI has a much more defined job.

But there is an important caveat:

A better prompt does not turn AI into a crystal ball.

The output still needs verification and independent investment judgement.

Common Prompt Engineering Mistakes ⚠️

1. Being Too Vague

“Tell me about AI.”

This can produce almost anything.

Better:

“Explain generative AI to an Indian equity investor, focusing on how it could affect IT services companies over the next five years.”

2. Giving Conflicting Instructions

For example:

“Give me a comprehensive analysis in exactly 100 words.”

The instructions may technically be possible, but they create unnecessary tension.

Better prompts prioritise requirements.

3. Overloading the Prompt With Unnecessary Instructions

Longer does not automatically mean better.

A 5,000-word prompt full of repetitive instructions may be worse than a carefully structured 500-word prompt.

Prompt quality is not measured by word count.

4. Asking for Everything at Once

“Analyse the company, compare competitors, forecast earnings, value the stock, analyse management, read all news and give me a buy/sell decision.”

This may work with powerful models, but complex workflows often benefit from decomposition.

5. Providing Poor Examples

If you give the model examples containing mistakes, inconsistent formatting or undesirable behaviour, the examples can work against you.

Examples should demonstrate the behaviour you actually want.

6. Forgetting the Output Format

If you want a table, say so.

If you want JSON, say so.

If you want five bullets, say so.

If you want a concise executive summary followed by detailed analysis, define that structure.

7. Assuming AI Knows Your Objective

Two investors can ask:

“Is this stock attractive?”

and mean completely different things.

One may want a dividend investment.

Another may want a 10-year growth stock.

Another may be looking for a short-term trade.

The objective matters.

8. Not Telling AI What to Do When Information Is Missing

If you don’t tell AI how to handle missing information, it may attempt to fill the gap.

Better:

“If reliable information is unavailable, say so rather than guessing.”

9. Expecting the First Prompt to Be Perfect

Prompt engineering is iterative.

The first useful prompt is often the beginning of the process rather than the end.

Common Misconception ⚠️

“The Longer the Prompt, the Better the Result”

Not necessarily.

A long prompt can be excellent when the task is genuinely complex.

But unnecessary instructions create noise.

Consider these two prompts:

Prompt A

“Analyse this stock carefully and give me everything you know.”

Prompt B

“Evaluate this stock for a 10-year investor. Focus on business quality, earnings growth, free cash flow, leverage, competitive advantages, valuation and risks. Separate facts from interpretation and present the result in a structured table.”

Prompt B is much shorter.

Yet it provides a better-defined task.

The goal of prompt engineering is therefore not:

Maximum prompt length.

It is:

Maximum useful clarity.

Advanced Prompt Engineering Techniques 🚀

Once you understand the basics, several advanced techniques become useful.

Structured Delimiters

Use tags or headings to separate different types of information.

<context>
...
</context>

<data>
...
</data>

<task>
...
</task>

Few-Shot Examples

Show the model examples of desired behaviour.

Input: ROE = 25%
Classification: Strong

Input: ROE = 8%
Classification: Weak

Now classify:
ROE = 18%

Task Decomposition

Break a complicated task into smaller stages.

Extract → Validate → Analyse → Compare → Conclude

Self-Checking

Ask the model to validate its output against explicit requirements.

For example:

“Before providing the final answer, check that every company has a valuation assessment and at least two risks.”

This does not guarantee correctness, but it can improve adherence to a defined checklist.

Tool Use 🔧

For tasks involving current information, calculations or external data, the model may need access to appropriate tools rather than relying exclusively on generated text.

For example:

Question
   ↓
Search / Database / API
   ↓
Retrieved Data
   ↓
AI Analysis
   ↓
Validated Output

Modern AI systems can combine model generation with tools such as web search, file search, code execution and APIs.

Prompt Engineering for Financial Analysis 📊

Financial applications deserve additional caution.

A prompt should ideally specify:

Time horizon

“Analyse for a 10-year investment horizon.”

Investment objective

“Prioritise capital appreciation.”

or:

“Prioritise sustainable dividend income.”

Metrics

For example:

• ROE

• ROCE

• ROA

• EBITDA margin

• PAT margin

• Free cash flow

• Debt-to-equity

• Interest coverage

• Revenue growth

• Earnings growth

• Valuation

Comparison

“Compare with three relevant listed peers.”

Risk

“Identify the three most important risks that could invalidate the investment thesis.”

This transforms:

“Is this a good stock?”

into a much more useful analytical exercise.

Prompt Engineering Is Becoming a Core AI Skill 🧠

There was a time when using computers effectively meant knowing software.

Then spreadsheets became essential for analysts.

Then search engines changed how people researched information.

AI is creating another layer:

The ability to communicate effectively with intelligent systems.

Prompt engineering sits at that intersection.

For ordinary users, it can improve:

• Research

• Writing

• Learning

• Coding

• Data analysis

• Productivity

For businesses, it can become part of:

• AI workflows

• Customer support

• Document processing

• Research systems

• RAG applications

• AI agents

• Automated analysis

The skill therefore extends well beyond writing clever ChatGPT questions.

The Future: From Prompt Engineering to AI System Design 🔮

Prompt engineering is likely to remain important, but its role is evolving.

The future of sophisticated AI applications is unlikely to be:

“Find the perfect prompt.”

It is more likely to be:

        USER
          ↓
      AI SYSTEM
          ↓
 ┌────────┼────────┐
 ↓        ↓        ↓
Prompt   Context  Tools
 ↓        ↓        ↓
 └────────┼────────┘
          ↓
        LLM
          ↓
      Evaluation
          ↓
       Output

The strongest AI applications will combine:

Good instructions + good context + good tools + good data + good evaluation.

Prompt engineering is one component of that larger system.

And there is another important development:

The AI itself can increasingly become part of the prompt-engineering process.

Instead of spending hours trying to perfect a prompt manually, you can describe your objective to an AI and ask it to critique, restructure and improve your prompt.

The human remains responsible for the objective and final judgement.

AI becomes the assistant that helps turn that objective into better instructions.

A Practical Prompt Template 📝

For everyday use, you can start with this:

<role>
Act as a [ROLE].
</role>

<context>
Here is the relevant background:
[CONTEXT]
</context>

<objective>
The objective is to:
[OBJECTIVE]
</objective>

<task>
Perform the following task:
[TASK]
</task>

<input>
Use the following information:
[INPUT]
</input>

<assumptions>
Assume:
[ASSUMPTIONS]
</assumptions>

<constraints>
Follow these constraints:
[CONSTRAINTS]
</constraints>

<output_format>
Return the answer as:
[FORMAT]
</output_format>

<output_style>
Use a [STYLE] style suitable for [AUDIENCE].
</output_style>

Don’t blindly copy this template into every conversation.

Instead, use it as a mental framework for constructing better instructions.

And remember: you can also ask AI to help you improve the template itself.

A Five-Question Prompt Checklist 🎯

Before pressing Enter, ask yourself:

1. Did I explain what I want?

Task

2. Did I provide enough background?

Context

3. Did I define the boundaries?

Constraints

4. Did I explain what a good answer looks like?

Examples / Output Format

5. Did I tell the AI who the answer is for?

Audience / Style

If the answer to all five is yes, you are already doing meaningful prompt engineering.

And if you’re unsure whether your prompt is good enough, ask AI to review it before you use it.

Risks & Limitations ⚠️

Prompt engineering has limits.

A beautifully structured prompt cannot compensate for:

• Incorrect source data

• Missing context

• Outdated information

• Weak models

• Poor retrieval

• Ambiguous objectives

• Incorrect assumptions

• Inadequate evaluation

There is also no guarantee that an AI will always follow every instruction perfectly.

Model behaviour can vary between models and model versions. Even apparently identical prompts can produce different outputs.

This is why important AI workflows should be tested rather than trusted blindly.

For financial applications, this becomes even more important.

AI-generated analysis should be treated as an analytical input—not automatically as investment advice or a substitute for independent verification.

Conclusion 🎓

Prompt engineering is often described as the art of talking to AI.

A better description is:

Prompt engineering is the discipline of translating human intent into instructions that an AI system can execute effectively.

The fundamentals are straightforward:

Start with a clear direction.
Provide context.
Specify the task.
Give relevant inputs.
Set assumptions and constraints.
Ask clarifying questions when necessary.
Show examples when useful.
Define the output.
Evaluate the result.
Iterate.

And remember something that is increasingly important:

You don’t have to engineer every prompt alone.

AI can help you analyse your prompt, identify ambiguity, suggest missing context, improve its structure and refine the expected output.

The biggest mistake is searching for a magical prompt that works for everything.

There isn’t one.

The better approach is to understand why a prompt works, diagnose where an AI response goes wrong, and systematically improve the instructions, context and workflow.

That is when prompting stops being a collection of tricks and starts becoming a genuine AI skill.

Key Takeaways 💡

1. The opening instruction establishes direction. Start complex prompts with a clear objective or role so the AI understands where the task is heading.

2. Good prompts reduce ambiguity by clearly defining the task, context and desired outcome.

3. Let AI ask questions when necessary. If missing information materially affects the answer, asking for clarification is often better than allowing AI to guess.

4. Context and assumptions matter. Give AI the information it needs and make important assumptions explicit.

5. Examples can be extremely powerful. Few-shot prompting can demonstrate the desired pattern, format or behaviour.

6. You can use AI to improve your prompts. Ask AI to critique, restructure and refine your instructions instead of trying to perfect every prompt manually.

Call to Action 🚀

AI is becoming an increasingly important tool for investors, analysts and businesses.

But the quality of the questions we ask determines how much value we extract from it.

And today, you don’t even have to be an expert prompt engineer.

You can use AI to become better at prompting AI.

Explore more practical AI, investing and financial education insights on Smart Investing India.

Invest smartly, India! 🇮🇳📈