12/08/2025, 20:57 Prompt Engineering for Equity Investors

Your email address Subscribe

Get 5 FREE AI Investing e-books direct to your inbox!

Prompt Engineering for Equity Investors:

Unlocking AI in Stock Research
Executive Summary

Artificial intelligence isn't just knocking on the door of stock analysis; it's already
remodeling the house. Large language models (LLMs) like the GPT series are stepping
up as tireless, 24/7 research assistants for investors like you. Imagine a tool that can sift
through mountains of financial news, regulatory filings, and market data at speeds no
human could dream of, pulling out key insights and potential investment signals. That's
the power on offer. But to truly harness this power and make smarter investment
decisions, there's a new skill to master: prompt engineering. Think of it as learning the
art of conversation with AI – giving it clear, concise instructions and the right context so
it delivers sharp, useful analysis instead of generic fluff or, worse, misleading answers.

In this e-book, we'll dive into why prompt engineering is becoming indispensable for
both everyday retail investors and seasoned professionals in the world of equities. We'll
explore how LLMs are already being put to work in equity research, walk through
essential prompting techniques, and introduce custom frameworks I've developed – 🟦
F.A.I.R., 🟩T.A.S.K., and 🟧C.O.M.P.S. – specifically tailored to the workflows of stock
analysis. We'll then level up with advanced strategies, like guiding the AI through step-
by-step reasoning and using few-shot examples, to elevate your analytical game.
Crucially, we'll tackle the thorny issues of compliance and data privacy, so you can use
these powerful tools responsibly and ethically in the highly regulated financial
landscape.

🔍 Real-world case studies – from a DIY retail investor digging into annual reports
to sophisticated systems like Morgan Stanley's internal AskResearchGPT.
🧰 Practical, copy-paste prompt library – ready-made templates to get you
started immediately.

https://davidwang95.github.io/Equities_Prompting_Ebook/ 1/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

🛠️ Troubleshooting playbook – solutions for common hiccups (like when the AI

serves up incorrect data or stubbornly refuses to answer).
🚀 Continued learning – key resources and tools to keep you at the forefront of
this rapidly evolving field.

⚠️ Note: Using an LLM for stock research will not, and should not,
replace your human judgment or the guidance of a professional financial
advisor. Making investment decisions based solely on an AI's output is a
risky proposition. These models, for all their strengths, have limitations:
they often lack real-time data, they can (and do) make errors (sometimes
called "hallucinations"), and they certainly aren't licensed to give
financial advice. However, armed with the right prompts and a critical
eye, an LLM can become an incredible productivity booster –
summarizing dense filings in minutes, sparking new insights, and even
suggesting questions you might not have thought to ask. This e-book is
your guide to unlocking that potential, step by step, so you can work
smarter with AI as your powerful ally, not a pitfall. Ready to begin? Let's
dive in!

Why Prompt Engineering Matters for Equity Investors

Imagine having an AI research assistant that's devoured every SEC filing, earnings call
transcript, news article, and financial textbook you could ever need. That's essentially
what today's large language models offer investors. They can digest and make sense of
massive volumes of unstructured financial data – far more than any single human or
even a team could realistically process – and distill it into remarkably human-like
analysis. This is a game-changer for equity research, an arena where an information
edge and timely insights are the coin of the realm.

But here's the catch: simply having access to a powerful LLM doesn't automatically
translate to high-quality, actionable outputs. The old adage "garbage in, garbage out"
applies with full force. Ask a vague question like, "What's the outlook for TechCorp?",
and you'll likely get a bland, generic answer filled with platitudes. However, if you learn
to ask the right questions, infused with the right context, you can unlock deeply
relevant and surprisingly accurate insights. Crafting these effective queries is the very
essence of prompt engineering.

https://davidwang95.github.io/Equities_Prompting_Ebook/ 2/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

So, why does prompt engineering really matter? It's the crucial bridge between your
expertise and the AI's vast capabilities. As an investor, you bring domain knowledge,
specific goals (like understanding why a stock plummeted after an earnings miss, or
how two competitors stack up), and critical thinking. The AI brings its enormous
knowledge base (trained on a significant portion of the internet, including a wealth of
financial texts) and incredible processing speed. The prompt is your way of clearly
communicating your goals and providing that vital context to the AI. Well-crafted
prompts unlock the AI's full potential, transforming it from a simple Q&A bot into a
sophisticated analytical partner. A good prompt can guide an LLM to produce a focused
valuation analysis instead of a Wikipedia-style summary. It can nudge the model to
consider important nuances (like distinguishing whether revenue growth was organic or
driven by acquisitions) rather than just skating on surface-level observations.

More than that, prompt engineering actually sharpens your own analytical and
communication skills. As one forward-thinking investment firm noted, to harness LLMs
effectively, analysts must amplify one of their most critical abilities: communication.
When you clearly specify what you need from the model, you're essentially formalizing
your own thought process. This not only leads to better AI output but often clarifies
your own understanding of the problem you're trying to solve. It forces you to think
precisely about the information you need and the questions you should be asking.

In a nutshell, prompt engineering is rapidly becoming a non-negotiable skill for

investors because it delivers:

Relevance: It empowers you to direct the AI to the most pertinent insights for your
specific investment thesis or question. Precise prompts yield answers that are
directly on-point for your needs.
Depth: It enables you to go beyond superficial facts and obtain deeper, more
nuanced analysis. You can instruct the model to adopt expert personas or delve
into the underlying causes and potential implications of events.
Efficiency: It saves you precious time by delivering higher-quality outputs faster. A
poorly phrased query might lead to a frustrating back-and-forth of clarifications; a
well-crafted prompt can often get you the answer you need in a single attempt.
Customization: You gain fine-grained control over the output. You can tailor the
format (bullets, tables, narrative), the tone (formal, informal, cautious), and the
level of detail to match your exact requirements. Need a concise five-bullet
summary for a quick read? Or a detailed report section for a deeper dive? Prompt
engineering makes it possible.
Risk Mitigation: Thoughtful prompts can help mitigate some of the inherent risks
associated with LLMs, such as bias or the tendency to "hallucinate" (invent facts). By
carefully constraining the model's focus and providing factual grounding, you can

https://davidwang95.github.io/Equities_Prompting_Ebook/ 3/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

improve the reliability of its outputs. (We'll explore this in more detail in the
troubleshooting section.)

These benefits are crucial for everyone, from retail investors (who often act as "one-
person research teams" juggling limited time) to professional analysts at large
institutions (who face a daily deluge of information and the pressure to cover more
ground, faster). It's no surprise that investment firms are actively training their staff in
prompt engineering, viewing it as a new core competency and, as some have put it,
"capital well spent" in the pursuit of better investment outcomes. Just as proficiency in
Excel or Bloomberg terminals became essential tools in previous decades, knowing how
to effectively leverage AI through skilled prompting is becoming critical in today's
investment landscape.

Reminder: No AI, no matter how advanced, can guarantee correct or profitable

investment decisions. Always, always use these tools as powerful aides to your thinking,
not as infallible oracles. They are here to augment your analytical process, not to
replace your due diligence or critical judgment. With that vital caveat in mind, let's
explore how LLMs are currently being used in equity research and how you can start
using them more effectively.

Large Language Models in Equity Research

Large language models have swiftly carved out a significant niche in equity research and
investing workflows. Their remarkable ability to understand natural language and
generate human-like text makes them incredibly versatile assistants for a wide array of
tasks throughout the investment process:

Information Summarization: LLMs can condense lengthy, dense documents like

10-K annual reports, 10-Q quarterly filings, and earnings call transcripts into
digestible summaries. For example, you could feed an LLM the Management
Discussion & Analysis (MD&A) section of an annual report and ask for a concise
summary of the company's strategy, performance, and key risks. This can save
hours of reading time, allowing you to quickly grasp the essentials.
Question & Answering: You can ask detailed, specific questions about a company,
an industry, or a financial concept, and the LLM will draw upon its vast training data
to provide answers. For instance: "What were the main factors cited for margin
improvement by RetailCo in its last earnings call?" The model might respond with a
summary referencing cost-cutting initiatives and enhanced pricing power, if those
were indeed mentioned in the transcript it processed (or was trained on).
Comparative Analysis: LLMs can be prompted to compare and contrast
companies or investment options across various metrics. You might ask: "Compare

https://davidwang95.github.io/Equities_Prompting_Ebook/ 4/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

BigAuto and FastCar Inc. in terms of their recent revenue growth, profit margins,
and debt levels." The model could then produce a side-by-side analysis, perhaps
even in a simple table format, highlighting which company appears stronger on
each specified metric (based on the information it has).
Sentiment & Trend Analysis: More advanced applications involve tasking the LLM
with gauging sentiment from news articles, social media (if it has access or is
provided with the data), or even analyzing the tone of management in earnings
calls. A prompt like, "Summarize the market sentiment surrounding FinTechCorp
over the past month based on these news headlines," (followed by the headlines)
could yield an indication of whether the prevailing tone has been positive, neutral,
or negative, potentially citing key events.
Drafting Narratives & Reports: An LLM can serve as a starting point for drafting
various written materials, such as investment theses, sections of research reports,
or even explanations of complex financial concepts in plain English. Some investors
even leverage the code generation abilities of models like GPT-4 to assist with
quantitative analysis, such as writing Python scripts to fetch and analyze stock data.
Domain-Specific LLMs: Recognizing the immense value of specialized financial
knowledge, we're seeing the emergence of LLMs trained specifically on financial
data. BloombergGPT, a 50-billion parameter model developed by Bloomberg, was
trained on a massive corpus of financial information (news, filings, transcripts,
proprietary data, etc.). It has demonstrated superior performance on many
financial tasks compared to general-purpose models, while still retaining strong
general language capabilities. Open-source initiatives like FinGPT are also gaining
traction, aiming to democratize financial AI by fine-tuning publicly available models
on financial datasets. These domain-specific models can often provide more
nuanced or accurate responses on finance-related topics (e.g., correctly parsing a
complex balance sheet item or understanding niche financial jargon in context)
than a general model. However, even general-purpose models have shown striking
competence in the financial domain.

Despite these impressive strengths, it's crucial to be aware of the notable limitations of
LLMs in equity research:

Knowledge Cutoff and Real-Time Data: Most publicly accessible LLMs (unless
specifically designed with live internet access or updated via plugins) have a
"knowledge cutoff" date. This means their training data, and thus their knowledge
of events and data, ends at a certain point in the past. Consequently, the model
might be unaware of the latest quarter's earnings, recent market-moving news, or
current stock prices. This limitation means you often need to provide the most
current information within your prompts if you want the model to consider it in its
analysis.

https://davidwang95.github.io/Equities_Prompting_Ebook/ 5/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Numerical Accuracy: LLMs are language models, not calculators or spreadsheet

programs. While they can perform basic arithmetic, they can struggle with complex
calculations or large datasets. More importantly, they can "hallucinate" numbers or
make numerical errors – for example, misplacing a decimal, confusing millions with
billions, or simply inventing a figure that sounds plausible. If you ask for financial
ratios, always double-check the math yourself or use a reliable financial data
source.
Hallucinations and Factual Errors: An LLM may sometimes generate information
that sounds perfectly plausible but is, in fact, partially or entirely false – a
phenomenon known as "hallucination." It might "recall" a non-existent SEC filing,
attribute a quote incorrectly, or make up a statistic because it fits the pattern of a
typical answer. This is particularly dangerous in finance, where factual accuracy is
paramount. While prompt engineering techniques can help reduce this risk, user
vigilance and verification are absolutely key. Never make a trading decision or
investment commitment without verifying critical facts from original, reliable
sources.
No True Understanding or Causality: These models operate by predicting the
next word in a sequence based on patterns learned from their training data. They
don't possess true understanding, consciousness, or the ability to reason about
causality in the way a human analyst does. They might be excellent at summarizing
what's explicitly stated in the data they process, but they are less reliable when it
comes to drawing novel conclusions that require deep, intuitive reasoning or
specialized domain expertise not present in their training. It's often helpful to treat
their analysis as you would the work of a very well-read but inexperienced junior
analyst: potentially useful and insightful, but always requiring review, critical
assessment, and a sanity check from you.
Compliance and Financial Advice: LLMs like ChatGPT are typically programmed
not to give explicit financial advice (e.g., "Should I buy this stock?"). They will usually
deflect such questions or provide a generic, non-committal response. This is a
deliberate and important safeguard. Providing personalized investment advice is a
regulated activity that requires licensing and a deep understanding of an
individual's financial situation and risk tolerance – capabilities an AI does not
possess. Therefore, think of LLMs as powerful information retrieval and analysis
tools, not as advisory tools.

All that said, the adoption of LLMs in the financial industry is undeniably real and
accelerating. Major financial institutions are increasingly integrating these technologies.
For example, Morgan Stanley developed an internal GPT-4-powered assistant for its
wealth management advisors, allowing them to quickly query the firm's extensive
research library and intellectual capital in natural language, receiving summarized
answers with source citations in seconds. This significantly augments analyst workflow,
https://davidwang95.github.io/Equities_Prompting_Ebook/ 6/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

saves time, and enables quicker, more informed client service. Many investment firms,
both large and small, are actively prototyping similar internal tools or establishing clear
guidelines for their analysts to use public tools like ChatGPT with appropriate oversight
and data protection measures. As these models continue to improve, gain access to
more real-time data through secure integrations, and become more adept at
specialized financial tasks, their role in equity research is set to expand even further.

For an individual investor, this is incredibly empowering. You essentially have a tool at
your fingertips that can function like a personal research assistant – one that has "read"
a vast library of financial information (up to its knowledge cutoff) and can communicate
its findings reasonably well. But to truly unlock the value from this sophisticated
assistant, you need to learn how to ask it the right way. In the next section, we'll get into
the practical nuts and bolts of prompt engineering essentials – the core techniques you
can start using immediately to dramatically improve the relevance, accuracy, and
quality of the AI outputs you receive in your stock research endeavors.

Prompt Engineering Essentials

Prompt engineering is often described as both an art and a science. The good news?
You don't need a degree in computer science or a deep understanding of the model's
internal architecture to become effective at it. At its heart, it's about clear
communication, logical thinking, and a bit of experimentation – skills you likely already
use in your daily analysis and writing. Let's break down the essential principles for
getting better results from an LLM:

🎯 Be Specific and Provide Context: Clarity is King

The golden rule of prompt engineering is precision. The more precise, detailed, and
unambiguous your prompt, the more focused, relevant, and useful the AI's answer will
be. Always operate under the assumption that the AI knows nothing about your true
intent or the specific nuances of your situation; it's your job to provide that clarity. Here
are key ways to inject vital context:

📋 Include Relevant Details

If you're asking about a specific company, always use its full name and perhaps its
industry to avoid ambiguity. Instead of a vague, "What's the outlook for Delta?", a much
better prompt would be, "What's the outlook for Delta Air Lines (DAL) over the next 12
months, considering the recent trends in fuel costs and passenger demand?" Just saying
"Delta" could lead the model down a rabbit hole, perhaps confusing the airline with the

https://davidwang95.github.io/Equities_Prompting_Ebook/ 7/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Greek letter used in options trading, or another company entirely. Clarifying the scope
(company, industry), timeframe, and the particular aspect you're interested in is crucial.

🎭 Frame the Situation

If necessary, add a sentence or two of background to set the scene. For example: "I'm a
retail investor analyzing TechCorp after its Q3 earnings release. The stock dropped 15%
immediately following the announcement. Could you provide some potential insights
into why the market might have reacted so negatively, based on common investor
concerns after such an event?" Here, you've given the model crucial context about the
event (earnings release, stock drop) and clearly stated your objective (insights into the
negative reaction). This is far more effective than a simple, context-free, "Why did
TechCorp stock go down?"

👥 Define the Audience or Perspective (Role-Playing)

If you want the answer tailored to a certain style, depth, or level of technicality, explicitly
state it. For instance, compare:

"Explain the concept of price-to-earnings ratio as if I have a basic understanding of

finance"
"Provide a detailed analysis of the limitations of the P/E ratio in valuing growth-
stage technology companies, suitable for an equity research report."

The first will yield a simpler, more accessible explanation, while the second will invite a
more technical and nuanced deep-dive. In fact, explicitly telling the LLM to adopt a
specific role can significantly sharpen its responses. If you preface your query with, "Act
as a seasoned portfolio manager. Analyze the key risks and opportunities for this stock
in the current macroeconomic environment," the AI will attempt to frame its answer
with the tone, considerations, and insights characteristic of that professional persona.
This "role-playing" technique is surprisingly effective.

📊 Example of Context in Action

Let's look at an example of context in action. Compare these two prompts:

Prompt A (Too vague): "Should I be worried about MacroAuto's debt?"

Prompt B (Superior): "MacroAuto Inc. (NYSE: MAI) recently announced its debt
levels doubled to $10 billion after acquiring a key competitor last year. As an equity
analyst focused on industrial companies, analyze whether this increased debt load
poses a significant danger to MacroAuto's financial health and its ability to fund
future growth. Consider its current cash flow, interest coverage ratios (if known, or
discuss typical healthy ranges), and the strategic rationale for the acquisition."
https://davidwang95.github.io/Equities_Prompting_Ebook/ 8/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Prompt (B) is far superior. It sets the scene with concrete information (debt doubled to
$10B, acquisition context, ticker for clarity), assigns a clear task (analyze the danger to
financial health and growth funding), and specifies a viewpoint (equity analyst). You can
easily imagine that prompt (B) will elicit a much more informative and structured
answer than the ambiguous prompt (A).

💬 Use Clear, Concise Language: Get to the Point

While LLMs are wizards with language, they don't truly understand it in a human sense;
they are sophisticated pattern-matchers. If you use convoluted, ambiguous, or overly
idiomatic phrasing, the model might get confused, misinterpret your intent, or generate
an irrelevant tangent. Strive for straightforward wording and avoid unnecessarily
complex sentence structures in your prompt. Keep it direct and unambiguous.

For example:

Clear and Effective: "Explain the impact of rising interest rates on BigBank's net
interest margin in simple, direct terms."
Muddled and Ineffective: "So, I was kind of wondering, you know, with the rates
going up and all, how that might, like, maybe, affect BigBank, 'cause they do a lot
with mortgages and loans, or something? What's the deal there?"

The second version is far more likely to lead to an off-target or unhelpful answer
because the model is struggling to parse your actual question from the noise.

📝 Pro Tips for Clear Communication

Keep prompts concise but comprehensive

Use bullet points or numbered lists for complex queries
Structure multi-part questions clearly
Avoid unnecessary fluff or rambling preambles

🔄 Iterate and Refine: Prompting is a Conversation

Don't expect perfection on your first try, especially with more complex queries.
Prompting is inherently an interactive and iterative process. Even seasoned prompt
engineers rarely craft the absolute perfect prompt for a complex task on their very first
attempt. Treat the AI's initial output as a draft, a starting point, or a step in an ongoing
conversation that you can guide and refine.

📋 Quick Checklist: When to Refine Your Prompt

https://davidwang95.github.io/Equities_Prompting_Ebook/ 9/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Too broad or generic: Add more specific details, constraints, or context. Ask for
concrete examples.
Missed the point: Try rephrasing using different keywords or a clearer structure.
Incorrect information: Provide correct information in your next prompt.
Wrong tone/style: Explicitly instruct the desired tone or format.

🎭 Instruct Roles and Formats: Take Control of the Output

As we touched upon earlier, you have the power to ask the model to adopt a specific
role or to follow a particular output format. This is a surprisingly powerful yet simple
technique, even for basic prompting:

👥 Role-Playing Examples

"You are a veteran stock analyst with deep expertise in the semiconductor sector.
Explain the primary competitive advantages and disadvantages of ChipMaker Inc. in
the current market."
"Explain the concept of 'economic moat' like I'm a high school student just
beginning to learn about stocks."

📊 Format Specification Examples

"Provide the answer as a table comparing Company A and Company B..."

"Present a numbered list of the 5 most critical risk factors..."
"Write a concise one-paragraph summary followed by 3 bullet points..."
"Answer in the style of an executive memorandum..."

✅ Your Pre-Prompt Checklist

📌 Objective Clearly Stated? Did I specify exactly what I want the model to do?
🔍 Sufficient Context Provided? Have I given all the necessary background
details?
💬 Unambiguous Language Used? Is my wording clear and direct?
👥 Role or Perspective Defined? Would the answer be improved with a specific
viewpoint?
📋 Format Instructions Included? Did I indicate the desired output format?
🔒 No Sensitive Data Included? Am I including any confidential information?
🎯 Realistic Expectation Set? Am I asking something the model can reasonably
do?

With these essential principles covered, you now have a solid foundation for effective
prompt engineering. Next, we'll build upon this by introducing some structured prompt

https://davidwang95.github.io/Equities_Prompting_Ebook/ 10/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

frameworks specifically designed for common stock analysis workflows. These

frameworks – F.A.I.R., T.A.S.K., and C.O.M.P.S. – act like templates or mental formulas,
helping you systematically structure your prompts for different analytical scenarios in
equity research. They encapsulate many of the principles we've just discussed in a
handy, memorable format. Let's explore those next.

Prompt Frameworks Tailored to Stock Workflows

To make prompt engineering more systematic and less like guesswork, it's incredibly
helpful to use frameworks – essentially, reusable patterns or templates for structuring
your prompts. In this chapter, I'll introduce three custom frameworks I've developed
and found particularly effective for typical equity research tasks. These frameworks
serve as both mnemonics (easy ways to remember key components) and practical
checklists, ensuring you cover all the essential elements in your prompt for a given
analytical scenario. Think of them as mental models for different types of stock analysis:

🔍 The Three Core Frameworks

F.A.I.R. – Designed for crafting comprehensive Fundamental Analysis prompts.
T.A.S.K. – Ideal for structuring step-by-step, Task-oriented prompts, especially when
you need the AI to solve an analytical problem or explain a process.
C.O.M.P.S. – Perfect for Comparables (Comps) analysis prompts that involve
comparing multiple companies, stocks, or data points.

Each framework is tailored to a common workflow in stock analysis. Let's break them
down one by one, with clear examples.

📈 F.A.I.R. Framework: For In-Depth Fundamental Analysis

When you need a well-rounded, insightful analysis of a single stock, a specific financial
event (like an earnings release), or a particular situation, the F.A.I.R. framework helps
ensure you don't miss key analytical components. It stands for:

F = Facts 📋

Begin your prompt by clearly establishing the relevant facts or the essential context.
These could be key financial figures, recent noteworthy events, specific data points
from a report, or any other foundational information the model should factor into its
analysis. Essentially, you're feeding the AI the raw material or drawing its attention to
the most pertinent information.

https://davidwang95.github.io/Equities_Prompting_Ebook/ 11/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Example: "Fact: XYZ Corp's revenue grew by 5% in the last reported quarter, but its net
income surprisingly fell by 10%. The company also announced the appointment of a
new CEO in June."

A = Analysis 🔍

Next, ask the model to analyze or interpret these facts. This is the core analytical task –
the "why" or "how" behind the stated facts. You're prompting the AI to connect the dots
and explain the dynamics at play.

Example: "Analyze the potential reasons why XYZ Corp's net income could have
dropped despite revenue growth. Consider factors such as changes in gross or
operating margins, increased operating expenses, one-time charges, or shifts in
product mix."

I = Insights 💡

Now, push the model to deliver higher-level insights or discuss the broader implications
of its analysis. This moves the answer beyond a simple interpretation of facts into the
"so what does this mean?" territory. Insights could relate to the company's future
outlook, its competitive positioning, investor sentiment, or strategic direction.

Example: "Provide insights into what these financial results and the CEO change might
indicate about XYZ Corp's current operational efficiency, its strategic priorities, and its
overall business health."

R = Risks ⚠️

Finally, instruct the model to identify and discuss any potential risks, uncertainties, or
red flags related to the scenario or its analysis. No thorough analysis is complete
without considering what could go wrong, what challenges lie ahead, or what key
unknowns remain.

Example: "Also, discuss any significant risks or red flags that investors should be
mindful of in light of these developments (e.g., potential impact of rising input costs,
challenges related to the CEO transition, competitive pressures, or evolving market
conditions)."

📝 F.A.I.R. Example Prompt:

"Act as a financial analyst. Fact: CleanTech Inc. (NASDAQ: CLNT) reported $200 million in
Q4 2024 revenue, representing a 15% year-over-year increase. However, it also
reported a net loss of $5 million for the quarter, its first quarterly loss in the past three
https://davidwang95.github.io/Equities_Prompting_Ebook/ 12/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

years. Analysis: Explain the primary factors that likely drove this net loss despite the
solid revenue growth. Consider potential reasons such as increased research and
development expenses, higher sales and marketing costs associated with market
expansion, supply chain disruptions impacting cost of goods sold, or one-time
restructuring charges. Insights: What does this combination of strong revenue growth
and a net loss suggest about CleanTech's current business model, its investment
priorities, and the competitive dynamics in the renewable energy market? Risks:
Highlight any significant risks or concerns for investors that emerge from these Q4
results, such as potential ongoing cash burn, the sustainability of its growth strategy if
profitability doesn't improve, or increased vulnerability to financing conditions if losses
continue."

🎯 T.A.S.K. Framework: For Step-by-Step Task Execution

Sometimes, you have a specific analytical problem to solve, a calculation you want the
AI to reason through (even if it doesn't do the math itself), or a process you want it to
explain systematically. The T.A.S.K. framework is designed to help structure prompts for
these kinds of procedural or stepwise analytical tasks. It stands for:

T = Target 🎯

Clearly define the target outcome, the specific question you're trying to answer, or the
precise objective you want to accomplish. This focuses the model squarely on the end
goal.

Example: "Target: Determine whether Company ABC's current stock valuation appears
justified by its historical growth performance and future growth prospects."

A = Approach 📋

Suggest or ask for a specific approach or methodology to tackle the target. This could
involve naming a particular analytical method (like Discounted Cash Flow analysis,
comparables analysis, or ratio analysis) or simply instructing the model to adopt a
logical, step-by-step approach. Essentially, you're guiding how the AI should go about
addressing the task.

Example: "Approach: First, assess Company ABC's historical revenue and earnings
growth rates over the past 3-5 years. Then, compare its key valuation multiples (e.g., P/E
ratio, P/S ratio) to those of its closest industry peers. Finally, consider any significant
qualitative factors that might influence its future growth trajectory."

S = Steps 👣

https://davidwang95.github.io/Equities_Prompting_Ebook/ 13/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Encourage or instruct the model to break down the problem into distinct steps (if the
chosen approach didn't already explicitly lay them out). For more complex questions,
you can even explicitly number the steps within your prompt. LLMs generally handle
such structured instructions well and will often mirror the stepped format in their
answer.

Example: "Steps:

1. Calculate or retrieve Company ABC's average annual earnings growth rate for the
last three fiscal years.
2. State Company ABC's current Price-to-Earnings (P/E) ratio and compare it to the
average P/E ratio of its primary industry competitors.
3. Evaluate whether ABC's historical growth rate and any stated future growth
initiatives can reasonably justify its current P/E multiple relative to its peers."

K = Key Takeaways (or Knowledge) 📚

Instruct the model to conclude with the key takeaways, the main findings, or the answer
derived from executing the specified steps. This ensures you get a clear, concise
bottom-line summary of the analysis.

Example: "Key Takeaways: Summarize whether Company ABC's stock appears to be

undervalued, overvalued, or fairly valued based on this step-by-step analysis, and
briefly explain the primary reasons for your conclusion."

📊 C.O.M.P.S. Framework: For Effective Comparables Analysis

Comparables analysis (often called "comps") is a cornerstone of equity research. It
involves comparing a company to its peers or to industry benchmarks across various
financial metrics and qualitative factors. The C.O.M.P.S. framework is designed to help
you craft clear and effective prompts for any comparative task, whether you're
comparing two specific stocks, a stock against a market index, or evaluating multiple
investment options. It stands for:

C = Context 🌍

Set the context by clearly specifying what is being compared. List the companies, assets,
time periods, or scenarios you want the AI to focus on.

Example: "Context: Compare AutoCo (NYSE: ACO), TruckCo (NASDAQ: TKO), and
MotoCo (OTC: MCO) – three publicly traded companies operating within the broader
vehicle manufacturing sector."

https://davidwang95.github.io/Equities_Prompting_Ebook/ 14/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

O = Objective 🎯

State the primary objective or the main goal of the comparison. What are you trying to
achieve or find out? Are you looking to identify which entity is "best" on a particular
metric, highlight key differences, rank them according to certain criteria, or understand
relative positioning?

Example: "Objective: Determine which of these three vehicle manufacturing companies

has demonstrated the strongest financial performance and operational efficiency over
the last reported fiscal year."

M = Metrics 📊

Specify the exact metrics or criteria that should be used for the comparison. This is a
crucial step. Without clear metrics, the model might guess what to compare (e.g.,
revenue size? stock price performance? number of employees?), leading to an irrelevant
analysis. You should clearly state the financial ratios, growth rates, qualitative factors,
or other data points that matter for your specific analytical objective.

Example: "Metrics: For each company, compare their (a) year-over-year revenue growth
percentage, (b) gross profit margin, (c) operating profit margin, and (d) debt-to-equity
ratio for the most recent fiscal year."

P = Perspective (or Presentation) 📋

Indicate any particular perspective, timeframe, or desired presentation format for the
comparison.

Example: "Perspective/Presentation: Focus on the data from the last completed fiscal
year (e.g., FY2024). Please present the comparison in a table with columns for each
company and rows for each of the specified metrics."

S = Summary 📝

Ask the AI to provide a summary of its findings or a concise conclusion drawn from the
comparison. This ensures the model not only lists numbers or facts but also interprets
them and offers a bottom-line assessment based on the requested analysis.

Example: "Summary: Based on the comparison of these metrics, highlight which

company appears to be the strongest overall performer in the last fiscal year. Also, note
any significant outliers (either positive or negative) among the three companies and
briefly explain their potential implications."

https://davidwang95.github.io/Equities_Prompting_Ebook/ 15/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

💡 Using Frameworks in Practice

You don't need to rigidly adhere to the acronyms or use them in every single prompt.
However, they are incredibly handy mental checklists, especially when you're
formulating a more complex query or if an initial, less structured prompt yielded an
unsatisfactory output. If an AI response feels lacking, you can mentally check it against
these frameworks:

For a fundamental analysis, did I cover all the F.A.I.R. elements to get a thorough
view?
For a complex problem or process explanation, did I guide the AI through the
necessary steps using T.A.S.K.?
For a comparison, did I clearly specify all the C.O.M.P.S. components, especially the
metrics and objective?

Often, you'll quickly identify a missing piece in your original prompt and can then re-
prompt with a more complete and structured query, leading to a significantly better
outcome.

We've now covered some custom frameworks to help you systematically structure your
prompts for common stock analysis workflows. Next, we'll explore some advanced
prompting techniques that go beyond basic Q&A. These techniques can further
enhance the depth, reliability, and sophistication of the AI's responses, especially when
you're tackling more complex equity analysis tasks.

Advanced Techniques for Equity Analysis

🔍 Introduction to Advanced Techniques

Once you're comfortable with the essentials and frameworks, you can start leveraging
more advanced prompt engineering techniques. These can significantly boost the
quality of your analysis, help overcome some of the inherent limitations of LLMs (like
multi-step reasoning or ensuring factual accuracy), and unlock more sophisticated
applications in your finance work. Here are some powerful approaches particularly
relevant for equity analysis:

🧠 Chain-of-Thought Prompting (CoT): Guiding Step-by-Step Reasoning

A common challenge with LLMs is that they aren't naturally adept at complex, multi-
step reasoning or intricate logical deductions – tasks often required in financial analysis
(e.g., building a forecast, conducting scenario analysis, or understanding the second-
order effects of an event). Chain-of-Thought (CoT) prompting directly addresses this by
https://davidwang95.github.io/Equities_Prompting_Ebook/ 16/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

explicitly guiding the model to "think out loud" or articulate its reasoning process step
by step before arriving at a final answer.

📝 How to Use It:

You simply include a phrase in your prompt like:

"Think through this step by step."

"Explain your reasoning process before giving the final answer."
"First, list out the logical steps you will take to analyze this, then proceed with the
solution."

💡 Why It Helps:

By forcing the model to enumerate and work through intermediate steps, CoT
prompting often leads to more accurate, logical, and transparent answers. It can reduce
the likelihood of "hallucinations" or nonsensical conclusions because the model is
constrained to follow a more structured thought path, making it easier for you to verify
each part of its logic.

📊 Example Prompt Using Chain-of-Thought:

"ABC Corp reported a 15% increase in net earnings for the last quarter, but its operating
cash flow decreased by 10% year-over-year. Explain step-by-step how such a divergence
between earnings and operating cash flow can occur. In your explanation, consider the
potential impacts of changes in working capital (accounts receivable, inventory,
accounts payable), non-cash revenues or expenses, and significant one-time items.
After detailing the possible reasons, provide a concluding thought on why investors
might not need to panic if this divergence is due to temporary factors."

🎯 Few-Shot Prompting: Learning by Example

Few-shot prompting is a technique where you provide the LLM with one or more
examples of the desired task and output directly within your prompt. This allows the
model to "learn" the pattern, style, format, or level of nuance you're looking for on the
fly, from the examples you provide.

📝 How to Use It:

Suppose you want an LLM to summarize a stock's quarterly earnings results in a

particular structured format that you use for your own research notes. You could
include a concise example of how you analyzed another company's results as a guide
within the prompt.
https://davidwang95.github.io/Equities_Prompting_Ebook/ 17/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

💡 Why It Helps:

Writing in a specific analyst tone or using a particular level of financial jargon

Performing simple, pattern-based calculations
Ensuring a consistent output format
Extracting specific pieces of information in a structured manner

⛓️ Prompt Chains and Task Decomposition: Breaking Down

Complexity

For very complex analytical tasks, trying to achieve everything in a single, massive
prompt can be counterproductive and lead to suboptimal results. Instead, you can
often achieve better outcomes by decomposing the complex task into a series of
smaller, more manageable sub-tasks, addressing each with a separate, focused prompt.

📝 How to Do It:

Identify the distinct sub-tasks within your larger analytical goal. Address them
sequentially, potentially using the output from one prompt as an input or context for
the next. For example, if you want to assess the long-term viability of a company:

1. Prompt 1 (Industry Analysis): "Provide a concise overview of the key growth

drivers, competitive landscape, and major risks for the [Specific Industry] industry
over the next 5-10 years."
2. Prompt 2 (Company's Competitive Positioning): "Thank you. Now, considering
those industry dynamics you just outlined, analyze how well BigTech Co. is
positioned to capitalize on the growth drivers and mitigate the risks within the
electric vehicle industry."
3. Prompt 3 (Financial Health Assessment): "Okay, that's insightful. Next, please
assess BigTech Co.'s financial health based on its latest annual report."
4. Prompt 4 (Synthesizing into a Conclusion): "Finally, based on our discussion of
the industry, BigTech Co.'s competitive positioning, and its financial health, provide
a summarized investment thesis."

💡 Benefits:

Focus and Manageability

Leverages Conversational Context
Error Isolation and Correction
Mimics Human Workflow
Complexity Management

https://davidwang95.github.io/Equities_Prompting_Ebook/ 18/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

🛠️ Using Tools and External Data (Hybrid Prompting & RAG)

While not strictly a "prompting" technique in isolation, the integration of LLMs with
external tools, live data sources, or private document repositories is a powerful
advanced strategy that significantly enhances their utility, especially in finance.

📝 What is RAG?

In a RAG system, when you ask a question, the system first retrieves relevant
information from a specified knowledge base (e.g., your firm's internal research
documents, recent SEC filings, a curated set of news articles, or even live web search
results). This retrieved information is then provided to the LLM as part of the context
along with your original prompt.

💡 Benefits of RAG and Providing External Data:

Reduces Hallucinations
Improves Factual Accuracy
Overcomes Knowledge Cutoff
Enables Analysis of Private/Proprietary Data

🤖 Meta-Prompting: Asking the AI to Help You Prompt

This is a slightly "meta" but surprisingly useful trick: you can actually ask the AI to help
you craft better prompts. If you're unsure how to best phrase a question to get the
information you need, or if your initial prompts aren't yielding good results, you can
describe your goal to the AI and ask for its suggestion on how to prompt for it.

📝 How to Use It:

"I want to analyze Company ABC's competitive advantages compared to its main
peers, focusing on its technology, brand, and market share. What would be an
effective way to prompt an AI like you to get a detailed and well-structured analysis
of this?"
"I'm trying to get an LLM to summarize the key financial takeaways from an
earnings call transcript, but my prompts are giving me very generic answers. Can
you suggest a more specific prompt structure that would encourage a more
insightful summary?"

📋 Recap of Advanced Methods: The Power of Intentional Guidance

https://davidwang95.github.io/Equities_Prompting_Ebook/ 19/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

These advanced techniques can often be combined for even greater effect. For
instance, you might use a few-shot prompt that includes an example of a chain-of-
thought reasoning process. Or you could use a tool (like code execution) to perform a
calculation as one step in a prompt chain, then feed that calculated result into a
subsequent prompt that asks for an explanation or interpretation.

💡 Core Principles:

Tell the model how to think: Guide it step by step, encourage logical deduction
Show the model what "good" looks like: Provide clear examples of the desired
output format, style, or content
Break big problems into smaller, manageable pieces: Don't try to do everything in
one go; decompose complex tasks
Bring in real, relevant data whenever needed: Don't rely solely on the model's
internal knowledge if you can provide factual, up-to-date information

By applying these more sophisticated approaches, you can elicit surprisingly nuanced,
detailed, and reliable outputs from LLMs. You can essentially co-pilot with the AI
through complex tasks like stress-testing your investment assumptions, performing
rudimentary scenario analysis, or even drafting substantial portions of a research
report.

However, always remember the golden rule: you remain the human in charge. You are
the analyst, the critical thinker, the decision-maker. Oversee each step of the AI's
process, rigorously verify any critical facts or figures, and ultimately use your own
judgment and expertise to draw the final conclusions and make investment decisions.

Now that we've armed you with a range of prompting techniques from basic to
advanced, we must turn our attention to a critical aspect of using these tools in the
financial world: compliance and data privacy. The next section focuses on how to use AI
responsibly and ethically, ensuring you operate within regulatory boundaries, company
policies, and protect sensitive information.

Compliance and Data Privacy for Equity Investors

🔍 Introduction to Compliance in AI-Powered Investing

The financial industry operates under a heavy blanket of regulations, and for very good
reasons – it deals with sensitive information, people's livelihoods, and the stability of
markets. Whether you're an individual retail investor or, especially, if you work within a
financial institution, navigating compliance and data privacy is paramount when using
Large Language Models. Ignoring these aspects can lead to serious consequences.
https://davidwang95.github.io/Equities_Prompting_Ebook/ 20/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Here's a breakdown of key considerations and best practices to keep you on the right
side of the line:

🛡️ Protect Confidential and Non-Public Information: The Cardinal

Rule

Never, ever input sensitive non-public information into a public AI service like the free
versions of ChatGPT or other general-purpose LLMs accessible via the web. This is the
most critical rule. Such information includes:

Material Non-Public Information (MNPI): This is any information about a publicly

traded company that has not been disclosed to the public and could affect its stock
price (e.g., unpublished financial results, impending merger or acquisition plans,
significant product breakthroughs or failures). As a retail investor, you're unlikely to
possess MNPI legally, but professionals in finance might encounter it and must
handle it with extreme care. Sharing MNPI inappropriately can lead to severe legal
penalties, including insider trading charges.
Confidential Client Data or Personally Identifiable Information (PII): This
includes any data related to your clients (if you're an advisor) such as their financial
holdings, investment goals, contact details, or any other personal information.
Proprietary Research or Internal Firm Documents: Any unique research,
analytical models, trading strategies, or internal communications developed by
your firm should not be fed into public LLMs unless your firm has an explicit policy
and a secure, approved environment for doing so.

⚠️ Why is this so critical?

When you use a public AI tool via its website, your inputs (the prompts and any data
you paste) might be stored on the AI provider's servers. Depending on the provider's
terms of service and your settings, this data could potentially be reviewed by humans or
even used to further train future versions of their models (though many providers like
OpenAI now offer options to opt out of data usage for training, especially for their paid
API services). In 2023, several major companies, including some prominent banks,
temporarily banned or severely restricted employee use of public AI tools like ChatGPT
precisely because of the risk of accidental leakage of confidential corporate or client
information.

🛠️ Solutions and Best Practices

📊 Anonymize and Aggregate

https://davidwang95.github.io/Equities_Prompting_Ebook/ 21/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

If you need to explore a scenario that is based on a real but sensitive case, you must
anonymize the data thoroughly. Change company names, alter specific figures to be
representative rather than exact, and remove any details that could identify the specific
entity or individuals involved. For instance, instead of typing, "Our client, ACME Corp.,
based in Springfield, has an EBITDA of $15.7 million and total debt of $50.2 million.
Analyze its debt service capacity," you might generalize it to: "Consider a hypothetical
mid-sized manufacturing company in the Midwest with an EBITDA of approximately
$15-20 million and total debt around $50 million. Discuss the typical factors an analyst
would consider when assessing its debt service capacity."

🔐 Use Secure, Enterprise-Grade Solutions

If you work for a financial institution, your firm may provide access to LLM capabilities
through secure, internal platforms or private instances hosted by approved vendors
(e.g., using Azure OpenAI Service or similar enterprise solutions). These environments
are typically designed with data security and privacy controls that prevent your data
from being exposed publicly or used for training general models. Always prioritize using
firm-approved tools for any work involving sensitive information. Morgan Stanley's
internal AskResearchGPT, which operates on their proprietary data within a controlled
environment, is a prime example of this approach.

📋 Understand Data Usage Policies

For any AI tool you use, carefully review its data usage and privacy policies. Pay
attention to whether your data is used for training, how long it's retained, and what
security measures are in place. For API usage, data is often not used for training by
default, but it's crucial to verify.

🚫 "When in Doubt, Leave It Out"

This is the safest mantra. If you have even the slightest concern that information might
be sensitive or confidential, do not include it in a prompt sent to a public AI service.

⚠️ No Financial Advice from AI: Maintain the Boundary

It's a critical point for both compliance and practicality: you should not ask an AI for
direct, personalized financial advice, and the AI should not (and generally will not)
provide it. As we've noted, mainstream LLMs like ChatGPT are usually programmed to
refuse explicit questions like, "Should I buy or sell X stock?" or any query that could be
construed as providing personal financial recommendations. This is by design. Giving
personalized investment advice is a regulated activity that requires licensing and a deep
understanding of an individual's financial situation and risk tolerance – capabilities an AI
https://davidwang95.github.io/Equities_Prompting_Ebook/ 22/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

does not possess. Therefore, think of LLMs as powerful information retrieval and
analysis tools, not as advisory tools.

📝 From your side, as the user:

Avoid phrasing queries as if you're seeking personal advice. Instead, focus your
prompts on eliciting analysis, information, or explanations.
Avoid ❌: "Tell me which of these five semiconductor stocks is the best one for me
to buy for long-term growth." (This is asking for a prediction and a personal
recommendation – not appropriate.)
Prefer ✅: "Compare the recent financial performance (revenue growth,
profitability) and current valuation multiples (P/E, P/S) of these five semiconductor
stocks: [List them]. Then, list the potential pros and cons of each as a long-term
investment, considering common market perspectives." (This requests objective
analysis and a summary of different viewpoints, not a directive for your action.)

🔍 Accuracy and Verification: Your Due Diligence is Non-Negotiable

Compliance also means not misleading others (or yourself) with incorrect or unverified
information. As we've stressed, LLMs can "hallucinate" – confidently presenting
numbers, facts, or even quotes that are inaccurate or entirely fabricated. If you plan to
use any AI-generated analysis in a formal context (such as a client report, a published
article, a blog post, or even just as a basis for your own significant trading decisions),
you must rigorously verify all key facts, figures, and critical assertions against reliable,
primary sources. Treat the AI's output as a first draft from a very fast but sometimes
error-prone assistant – a draft that always needs careful fact-checking and critical
review. You are the senior analyst in this relationship; the AI is the junior.

📋 Record Keeping and Disclaimers: Transparency is Key

If you use AI-generated content in any professional capacity, maintaining good records
and practicing transparency is crucial.

📝 Documentation

Depending on your firm's policies and any applicable regulatory requirements, you
might need to document when and how AI was used in the preparation of research,
reports, or client communications. Some compliance regimes are beginning to
formulate guidelines around AI usage, and internal firm policies are often ahead of
formal regulations.

⚠️ Disclaimers
https://davidwang95.github.io/Equities_Prompting_Ebook/ 23/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Always include clear and conspicuous disclaimers when AI-assisted content is shared,
especially externally. A good disclaimer might read something like:

"This analysis was prepared with the assistance of artificial intelligence (AI) Large
Language Models. While efforts have been made to ensure accuracy, AI models may
have limitations, produce errors, or generate incomplete information. All information,
analysis, and opinions presented should be independently verified from reliable
sources before making any investment decisions. This content is for informational
purposes only and does not constitute financial advice or a recommendation to buy or
sell any security. This covers both the AI assistance factor and the standard no-advice
stipulation. Many reputable financial education websites and research providers are
adopting similar transparency practices.

🌐 The Evolving Regulatory Landscape: Stay Informed

Regulators worldwide are actively monitoring the rapid advancements and adoption of
AI in the financial sector. While specific, detailed rules for "prompt engineering" per se
may not exist yet, a host of existing general regulations and principles clearly apply:

Data Protection and Privacy Laws (e.g., GDPR in Europe, CCPA in California): If
you are handling personal data, especially of clients or individuals in jurisdictions
with strong data privacy laws, be extremely cautious. Sending such data to an
external AI service could be considered data processing and might require explicit
consent and adherence to strict data handling protocols. For personal investing
research not involving others' data, this is less of an issue, but for any professional
context involving client information, it's a major red flag unless done through
secure, compliant channels.
Research Objectivity and Disclosure Rules (e.g., SEC, FINRA rules in the U.S.): If
you publish equity research or make recommendations, there are stringent rules
about ensuring the information is fair, balanced, not misleading, and that any
potential conflicts of interest are disclosed. If AI contributes to your research, you
are still responsible for ensuring the final output meets all these regulatory
standards. The AI won't understand these nuances; you must.
Intellectual Property (IP) and Copyright: Be mindful if you're prompting an AI
with large verbatim chunks of text from copyrighted materials, such as paid
research reports, books, or proprietary databases. You are essentially sharing that
potentially protected content with the AI service provider, which could raise IP
infringement concerns. It's generally safer to use publicly available information,
your own original writing, or concise summaries (rather than extensive verbatim
copies) for inputs, especially when dealing with third-party copyrighted content.
Anti-Manipulation and Market Integrity Rules: Ensure that your use of AI
doesn't inadvertently lead to the creation or dissemination of false or misleading
https://davidwang95.github.io/Equities_Prompting_Ebook/ 24/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

information that could manipulate markets. This ties back to the critical need for
accuracy and verification.

🏢 Adhere to Firm Policies: Your First Line of Compliance

If you work for an investment firm, bank, or any regulated financial institution, your
company's internal policy on AI usage is your primary guide and absolute requirement.
Many firms have already developed, or are in the process of developing, specific
guidelines and policies regarding the use of AI tools, including LLMs. These policies will
dictate which tools are approved, what types of data can (and cannot) be used with
them, any required disclaimers or documentation, and internal approval processes for
certain use cases. Always consult and strictly adhere to your firm's internal policies.
They are designed to protect both you and the firm from legal, regulatory, and
reputational risks. Some firms may provide access to internally vetted and secured AI
tools – prioritize using those for any work-related tasks.

🎯 Summary: Key Principles for Compliance and Data Privacy

In summary, the guiding principles for compliance and data privacy when using LLMs in
finance are: be cautious, be transparent, be secure, and be responsible. The last thing
you or your firm wants is to inadvertently breach client confidentiality, violate
regulatory requirements, or distribute inaccurate information due to an oversight in
how AI was used. Fortunately, by focusing on using AI for the analysis of public
information, rigorously verifying its outputs, protecting sensitive data, and adhering to
established policies, you can effectively and responsibly leverage these powerful tools.

We've covered the essential cautionary tales and guidelines – but don't let this
discourage you. The potential benefits of using LLMs in equity research are immense,
and many investors and institutions are already finding ways to harness this potential
safely and effectively. It's about smart, informed usage.

Next, let's bring all this theory to life by looking at some real-world and illustrative case
studies. We'll see how different types of investors (from retail to professional) have
applied prompt engineering in their workflows, and what practical lessons we can draw
from their experiences.

Case Studies: Prompt Engineering in Action

Theory is one thing, but seeing prompt engineering applied in real or realistic scenarios
can truly illuminate its power and pitfalls. In this section, we'll walk through a few
illustrative case studies. These are a mix of scenarios inspired by actual uses and

https://davidwang95.github.io/Equities_Prompting_Ebook/ 25/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

publicly discussed examples, demonstrating how prompt engineering techniques are

being employed in equity analysis today.

📈 Case Study 1: The DIY Retail Investor Summarizing an Annual

Report

🎯 Background

Jane is a diligent retail investor who holds shares in AlphaTech Corp. (fictional). The
company has just released its annual report (10-K), a dense 120-page document. Jane
wants to quickly grasp the key takeaways – the company's strategic direction, any
significant new risk factors, and a summary of its financial performance – but she
doesn't have several hours to meticulously comb through the entire filing.

🔍 Prompt Engineering Approach

Jane decides to leverage an LLM like ChatGPT to get a head start. She identifies the
most critical sections: "Business Overview," "Management's Discussion and
Analysis (MD&A)" for performance, and "Risk Factors." She carefully copies key
excerpts from these sections (being mindful of prompt length limits, perhaps taking the
introductory paragraphs and a few key sentences from major risk descriptions). She
then crafts the following prompt:

"Context: Below are selected excerpts from AlphaTech Corp.'s recently released 10-K
annual report. I've included parts of the Business Overview, MD&A, and Risk Factors
sections. [<< Jane pastes the selected text from the 10-K here >>]

Task: Based only on the provided text excerpts:

Summarize AlphaTech's core business model and its stated strategy for the upcoming
year in 2-3 sentences.
Identify and list the top 3 most significant risk factors highlighted in the
provided text, explaining each in simple terms.
Briefly mention any key financial performance trends (e.g., revenue growth,
profitability changes) discussed in the MD&A excerpt.

Format: Please provide the answer in three distinct sections corresponding to the
tasks above."

✅ Result: The AI processes the provided text and produces a structured summary. For
example:

https://davidwang95.github.io/Equities_Prompting_Ebook/ 26/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Business & Strategy: AlphaTech designs and sells specialized semiconductor

testing equipment primarily for the consumer electronics industry. Their strategy
for the next year focuses on expanding into the automotive and industrial sectors
while investing in next-generation 5nm chip testing solutions.
Top Risk Factors:
Supply Chain Disruptions: The company relies heavily on a small number of
key component suppliers, and any disruption could significantly impact
production and delivery schedules.
Intense Competition: AlphaTech faces strong competition from larger, more
established players, which could pressure pricing and market share.
Technological Obsolescence: The semiconductor industry evolves rapidly;
failure to innovate and adapt to new technologies could render AlphaTech's
products uncompetitive.

Financial Performance Trends: The MD&A excerpt indicates a 15% increase in

annual revenue driven by strong demand, but also a 5% decrease in net income
due to higher R&D investments and increased raw material costs.

💡 Outcome & Lesson: Jane quickly gets a coherent overview of the crucial points from
the dense 10-K. This allows her to focus her subsequent manual review on the areas
flagged by the AI. Importantly, during her own skim-through, she notices the AI's
summary of one risk factor was slightly oversimplified. She then refines her approach
for future summaries by prompting the AI to "quote the exact sentence fragment that
best describes each risk before explaining it in simpler terms." This iterative learning
improves her prompting skill. The AI provided an excellent first-pass summary, saving
significant time, but her verification and iterative refinement were key to maximizing its
utility and accuracy.

📊 Case Study 2: Professional Analyst Leveraging an Internal, Secure

LLM
Inspired by Morgan Stanley's AskResearchGPT

🎯 Background

Maria is an equity analyst at a large investment bank that has deployed an internal,
secure LLM assistant (let's call it "AskFinanceGPT"). This tool is fine-tuned on the firm's
proprietary research reports, internal market commentary, and approved financial data
sources. Ahead of an urgent client meeting, Maria needs to quickly get the firm's latest
consolidated view on BioPharma Co. (fictional), a company covered by multiple analysts
across different therapeutic areas within the firm.

https://davidwang95.github.io/Equities_Prompting_Ebook/ 27/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

🔍 Prompt Engineering Approach

Maria opens AskFinanceGPT (which, crucially, only accesses the firm's internal, vetted
knowledge base, ensuring compliance and data security). She types a natural language
query:

Query: "Provide a consolidated summary of our firm's current investment outlook on

BioPharma Co. (Ticker: BPHM). Include:

The current official rating (e.g., Buy, Hold, Sell) and 12-month price target.
The key bullish drivers and bearish risks highlighted in our latest research.
Any recent significant updates or changes to our thesis.
If there are differing views or scenario analyses (e.g., bull vs. bear case
valuations) from different analysts covering aspects of BPHM, please highlight
them.
Ensure all information is sourced from reports published in the last 3 months and
provide direct links or references to the source documents within our system."

📈 Result

AskFinanceGPT processes the query against the firm's internal research database and
swiftly returns a synthesized summary:

Firm Outlook: Our firm's current outlook on BioPharma Co. (BPHM) is 'Overweight' with
a 12-month price target of $45.00 (as of March 15, 2025, Report #XYZ123).

Key Bullish Drivers:

Strong Phase 3 trial data for its lead oncology drug, OncoVance
Expanding market share in autoimmune therapies
Robust pipeline with 3 upcoming catalysts in H2 2025 (Report #ABC789)

Key Bearish Risks:

Upcoming patent expiration for its blockbuster drug ArthriBloc in 2026

Potential FDA regulatory hurdles for OncoVance's approval
Increasing competition in the autoimmune space from generics and new entrants
(Report #DEF456)

Recent Updates:

Dr. AnalystA recently noted increased optimism on OncoVance's market penetration

speed (Analyst Note #GHJ007)
Dr. AnalystB highlighted potential margin pressures from rising input costs in
their latest sector review (Sector Report #KLM333)

https://davidwang95.github.io/Equities_Prompting_Ebook/ 28/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Scenario Analysis:

Base Case: PT $45

Bull Case: PT $55 (assuming early OncoVance approval and stronger-than-expected
sales)
Bear Case: PT $30 (factoring in potential FDA delay and faster erosion of
ArthriBloc sales)

All statements would ideally be followed by internal links to the cited reports.

💡 Outcome & Lesson

In minutes, Maria receives a comprehensive, up-to-date, and internally consistent

summary that would have taken her considerable time to compile by manually
searching and reading multiple lengthy research reports. The system's ability to cite
sources allows her to quickly dive deeper into specific points if needed. This
demonstrates how a well-prompted, domain-specific LLM, operating within a secure
and compliant environment, can dramatically boost analyst productivity and enable
more agile client service. The key was a clear, structured prompt that specified exactly
what information was needed and from what sources. Maria still uses her expertise to
frame this information for the client, but the information retrieval and synthesis were
massively accelerated.

🎓 Case Study 3: Academic Research Highlighting LLM Capabilities

Inspired by the University of Chicago Study

🎯 Background

A widely discussed 2023 study from the University of Chicago explored GPT-4's ability to
analyze standardized financial statements to predict future earnings changes. The
researchers effectively prompted the model with financial data and asked it to make a
directional earnings prediction.

🔍 Prompt Engineering Approach (Conceptual)

While the exact prompts aren't always public, the methodology involved providing GPT-
4 with structured financial statement data (e.g., revenue, various expense items, assets,
liabilities from income statements and balance sheets for several periods) for a given
company. The prompt would have then asked the model to:

Financial Analysis Prompt

https://davidwang95.github.io/Equities_Prompting_Ebook/ 29/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

"Analyze the provided historical financial statement data for Company X. Based on
trends in revenues, expenses, margins, assets, and liabilities, predict whether the
company's earnings per share are more likely to increase or decrease in the next
reported period. Explain your reasoning step-by-step, highlighting the key financial
indicators that support your prediction."

Note: This is a conceptual representation; actual research prompts would be highly

standardized and systematically applied across many companies.

📊 Result

The study found that GPT-4, even without industry-specific context or news (relying
solely on the patterns in the financial statements), achieved a notable accuracy in
predicting the direction of next-period earnings, in some cases outperforming human
financial analysts who were also given the same limited information. The LLM appeared
to identify subtle patterns and relationships within the financial data that were
indicative of future earnings trends. Follow-up research through 2024 and early 2025
has continued to explore these capabilities, with some studies focusing on the LLM's
ability to interpret the nuances in MD&A sections or identify early warning signs of
financial distress from footnote disclosures when prompted appropriately.

💡 Outcome & Lesson

This type of research underscores the potential of LLMs, when skillfully prompted with
relevant data, to perform sophisticated pattern recognition and analytical tasks that
were previously the exclusive domain of human experts. It doesn't mean LLMs will
replace analysts (real-world analysis involves far more qualitative factors, industry
knowledge, and forward-looking insights). However, it shows that LLMs can be a
powerful tool for generating initial hypotheses, identifying anomalies, or providing a
"second opinion" on financial data. An individual investor could adapt this by feeding a
few quarters of a company's key financial data into an LLM and asking for an analysis of
trends and potential future direction, always to be cross-referenced with their own
broader research. The key is providing structured data and asking for reasoned
explanations.

Case Study 4: Iterative Troubleshooting – Correcting "Hallucinated" Financials

Background

Sam, an investor, is trying to use a general-purpose LLM to quickly compare the Price-
to-Earnings (P/E) ratios of two companies in the same sector, CompanyX and

https://davidwang95.github.io/Equities_Prompting_Ebook/ 30/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

CompanyY. He asks a simple prompt: "Which company has a higher P/E ratio,
CompanyX or CompanyY, and what does their P/E suggest about their valuation?"

Initial Problematic Result

The LLM responds: "CompanyX has a P/E ratio of 25, while CompanyY has a P/E ratio of
18. This suggests that CompanyX's stock is relatively more expensive compared to its
earnings than CompanyY's." Sam is puzzled because he doesn't recall those exact P/E
figures, and upon checking a reliable financial data source (like Yahoo Finance or his
brokerage platform), he finds that CompanyX's current P/E is actually 30, and
CompanyY's is 22.

What Went Wrong

The LLM likely hallucinated the P/E numbers or retrieved outdated information from its
training data. It tried to be helpful by providing specific figures, but in doing so, it
presented incorrect data. This is a common pitfall when asking LLMs for precise, real-
time, or frequently changing numerical data without providing it directly.

Prompt Engineering Fix (Iterative Refinement)

Sam learns from this experience. He realizes he needs to supply the factual data
himself to ensure accuracy. He looks up the current P/E ratios and then crafts a new,
improved prompt:

"I am comparing the valuations of CompanyX and CompanyY. According to recent market
data (as of May 2025), CompanyX has a Price-to-Earnings (P/E) ratio of approximately
30, and CompanyY has a P/E ratio of approximately 22. Task:

Acknowledge these provided P/E ratios.

Analyze what the difference in these P/E ratios (30 for CompanyX vs. 22 for
CompanyY) suggests about market expectations for each company's future growth,
risk profile, or profitability.
Discuss potential reasons why CompanyX might command a higher P/E multiple.
Important: Please base your analysis only on the P/E ratios I have provided. Do
not introduce or guess any other financial figures for these companies."

Improved Result

Now, grounded with the correct factual data, the LLM produces a much more reliable
and useful analysis: "Acknowledged: CompanyX P/E is ~30, CompanyY P/E is ~22. The
higher P/E ratio of 30 for CompanyX compared to 22 for CompanyY indicates that

https://davidwang95.github.io/Equities_Prompting_Ebook/ 31/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

investors are currently willing to pay more for each dollar of CompanyX's earnings than
they are for CompanyY's. This often suggests that the market has higher expectations
for CompanyX's future earnings growth, perceives it as having a lower risk profile, or
believes it possesses stronger competitive advantages or profitability prospects.
CompanyX might command this premium if, for example, it operates in a faster-growing
sub-sector, has a more innovative product pipeline, or has demonstrated more
consistent historical growth than CompanyY..."

Outcome & Lesson

By taking the extra step to provide the critical factual data directly in the prompt and
explicitly instructing the AI not to guess, Sam transformed a misleading output into a
valuable analytical insight. This case study highlights a crucial general rule for using
LLMs in finance: whenever precise, current data is essential, provide that data to the AI.
Don't assume it knows the latest statistics. This approach significantly improves the
reliability of the output and ensures that the AI's analytical capabilities are applied to an
accurate factual basis. It might involve a little more upfront work for the user (looking
up the numbers), but the improvement in output quality and trustworthiness is well
worth the effort.

These case studies illustrate that effective prompt engineering is often an iterative and
context-dependent process. Different scenarios – whether summarizing lengthy texts,
querying a specialized knowledge base, performing data analysis, or correcting an AI's
error – benefit from different techniques and levels of user input. The common threads
are:

Provide clear context and specific, accurate data whenever possible.

Ask for structured, well-defined outputs.

Leverage internal, secure tools when dealing with proprietary or sensitive information.

Always, always verify important outputs and use your own critical judgment.

With these real-world and illustrative examples in mind, you might be eager to start
trying out similar approaches yourself. To help you hit the ground running, the next
section provides a library of example prompts – a kind of "cheat sheet" that you can
copy, paste, and adapt for a variety of common stock investing tasks.

Copy-Paste Prompt Library

One of the quickest ways to get proficient with prompt engineering is to see and use
well-crafted examples. This section offers a library of ready-made prompt templates
https://davidwang95.github.io/Equities_Prompting_Ebook/ 32/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

designed for various common tasks in stock analysis and investing. You can copy, paste,
and then modify these prompts to suit the specific companies, situations, or data you're
working with. They are organized by common analytical categories.

Think of these as starting points or inspiration. Feel free to tweak the phrasing, add
more context, combine elements from different prompts, or adjust the level of detail to
match your needs.

Fundamental Analysis Prompts

(Use these to analyze company financials, earnings reports, and core business
fundamentals.)

Earnings Report Summary (When you provide the text):

"Context: Below is the 'Management's Discussion and Analysis' (MD&A) section from
[Company Name]'s Q[X] [Year] earnings report. [Paste MD&A text here]

Task: Based only on the provided text:

Summarize the key revenue and profit figures and their year-over-year changes.
List the top 2-3 reasons management cited for the company's performance in this
quarter.
What was mentioned about the company's outlook or guidance for the next
quarter/full year?"

Quick Earnings Snapshot (General Knowledge, verify numbers):

"Provide a brief overview of [Company Name]'s (Ticker: [Symbol]) most recently

reported quarterly earnings. Include reported EPS vs. estimate, revenue vs. estimate,
and any key highlights or management commentary regarding future guidance. State the
approximate date of this earnings report if known."

(Self-correction: Always verify numbers from a reliable financial source after using
this.)

Understanding Financial Ratios (When you provide the ratios):

"[Company Name] currently has a Debt-to-Equity ratio of [X.X] and a Return on Equity
(ROE) of [YY]%. Explain in simple terms:

https://davidwang95.github.io/Equities_Prompting_Ebook/ 33/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

What these two ratios indicate about [Company Name]'s financial health and
efficiency.
How these figures might compare to typical ranges for companies in the [Specific
Industry, e.g., 'mature software'] industry.
One potential positive and one potential negative implication of these ratio
levels for an investor."

Analyzing Revenue Streams:

"Based on publicly available information up to your last knowledge update, how does
[Company Name] (Ticker: [Symbol]) primarily generate its revenue? Describe its main
business segments or product/service categories and, if possible, give an approximate
percentage breakdown of revenue contribution from each. What are the key growth
drivers for its largest segment(s)?"

Investment Thesis Element - Economic Moat:

"Analyze the concept of an 'economic moat' as it applies to [Company Name] (Ticker:

[Symbol]).

What are its primary competitive advantages (e.g., brand, network effects,
patents, cost advantages)?
How sustainable do these advantages appear to be in the face of current and
potential competition in the [Specific Industry] sector?
Identify one key threat that could erode its moat over the next 5 years."

Qualitative DCF Thought Exercise (Focus on Logic, Not Calculation):

"Imagine we are trying to conceptually value [Company Name] (Ticker: [Symbol]) using a
Discounted Cash Flow (DCF) approach. Assume the company is projected to grow its free
cash flow at an average of [e.g., 12%] per year for the next 5 years, and then at a
terminal growth rate of [e.g., 3%] thereafter. Assume a discount rate (WACC) of [e.g.,
9%]. Without performing precise calculations, explain the key sensitivities in such a
DCF model. For instance, how would the valuation be impacted if:

a) The high-growth period was only 3 years instead of 5?

b) The discount rate was 11% instead of 9%?
c) The terminal growth rate was 2% instead of 3%?

https://davidwang95.github.io/Equities_Prompting_Ebook/ 34/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Focus on explaining the directional impact and the underlying logic."

Next up, we'll tackle the inevitable: what to do when the AI doesn't quite give you what
you want. Our troubleshooting playbook will provide solutions for common issues,
ensuring you can navigate these powerful tools with confidence and get the information
you need.

Troubleshooting Playbook: Navigating Common LLM

Hiccups

Even with the best-laid prompts, you'll occasionally encounter responses from LLMs
that are off-target, incorrect, confusing, or simply unhelpful. Don't get discouraged! This
is a normal part of interacting with these complex systems. This troubleshooting
playbook outlines common problems and offers practical solutions to help you quickly
get back on track and coax the desired information from your AI assistant.

Problem 1: Hallucinated Facts, Figures, or Sources

Symptom: The AI's output includes a financial figure, a specific fact (e.g., a date of an
event, a product detail), or even a cited "source" or "CEO quote" that you know to be
incorrect, seems suspicious, or you simply can't verify. For example, it states,
"CompanyX's Q3 revenue was $5.2 billion," when you believe (or later verify) it was $4.8
billion.

Solution(s):

Direct Correction & Re-Prompt: If you know the correct information, provide it
and ask for a revised analysis. "Actually, CompanyX's Q3 revenue was $4.8 billion,
not $5.2 billion. Please re-evaluate its revenue growth based on the correct figure."
The model will usually acknowledge the correction and adjust.
Constrain to Provided Data: This is the most robust solution for data-sensitive
queries. Explicitly instruct the model to use only the information you provide.
"Using only the financial data provided in the table below, calculate the gross profit
margin for CompanyY. [Paste table here]. Do not use any external knowledge or
make assumptions beyond this data."
Request Sources (and be skeptical): You can ask, "Can you provide the source for
that statement/figure?" Sometimes, this prompts the AI to admit uncertainty or
even retract the claim if it can't substantiate it from its reliable training data.
However, be aware that LLMs can also hallucinate plausible-sounding sources, so
always verify any critical source if possible.
https://davidwang95.github.io/Equities_Prompting_Ebook/ 35/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Employ Chain-of-Thought (CoT): As discussed earlier, asking the model to "think

step by step" can reduce the likelihood of it jumping to an unsupported factual
conclusion.
Lower the "Temperature" (if using an API): Some API access allows you to adjust
generation parameters like "temperature." A lower temperature (e.g., 0.2) makes
the output more deterministic and factual, while a higher temperature encourages
more creativity (and potentially more hallucination). This is more for developers but
good to be aware of.

Problem 2: The Answer is Too General, Vague, or Superficial

Symptom: You receive a fluffy, high-level answer that lacks specific details, actionable
insights, or concrete examples. E.g., "CompanyX is a dynamic company operating in a
competitive market. It has several strengths and also faces some challenges. Its future
prospects could be positive if it executes its strategy effectively." – This tells you virtually
nothing.

Solution(s):

Increase Specificity in Your Follow-Up: Ask for concrete details, numbers,

examples, or deeper explanations. "That's too general. Could you please identify
three specific strengths of CompanyX and provide a brief example for each? Also,
list two specific challenges it currently faces."
Refine Your Original Prompt: Your initial question might have been too broad.
Instead of "Tell me about CompanyX," try a more targeted query like, "What are
CompanyX's main competitive advantages in the [Specific Niche] market, and how
do these compare to its primary rival, CompanyY?"
Request a Different Perspective or Role: "Explain that from the perspective of a
skeptical credit analyst focusing on potential risks."
Break Down the Request (Prompt Chaining): If a broad question yields a vague
answer, break it into smaller, more specific sub-questions and tackle them one by
one.

Problem 3: The Answer is Too Detailed, Long-Winded, or Off-Topic

Symptom: The AI produces an exhaustive essay when you needed a few quick bullet
points, or it starts to ramble and deviate from your core question.

Solution(s):

Politely Interrupt and Refocus/Summarize: In a chat interface, you can often just
start typing your next prompt. You can then say, "Okay, thank you. Can you
summarize the main points of your previous response in just three bullet points?"
https://davidwang95.github.io/Equities_Prompting_Ebook/ 36/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

or "Let's pause that line of thought. Can we focus specifically on [the original core
topic]?"
Specify Length and Format Upfront: The best prevention is to clearly state your
desired output length and format in your initial prompt. Use directives like:
"Provide a concise summary in no more than 100 words." or "List the key factors in
5 bullet points." or "Keep the explanation to one paragraph."
Define the Audience and Purpose: Framing the request for a specific audience
with limited time can also imply brevity. "Explain this concept as if you're briefing a
busy CEO who has only two minutes before a meeting."

Problem 4: Refusal to Answer or "As an AI..." Disclaimers

Symptom: The AI responds with a message like, "I'm sorry, but I cannot fulfill that
request," or "As an AI, I cannot provide financial advice," possibly because it
misinterpreted your analytical query as a request for prohibited content (like
investment advice, predictions of future stock prices, or illegal activities).

Solution(s):

Reframe the Request for Analysis, Not Advice: This is the most common fix. If
you asked, "Should I invest in Stock X?" it will likely refuse. Change it to: "What are
the potential pros and cons of investing in Stock X from a fundamental analysis
perspective?" or "Analyze the investment case for Stock X, considering its growth
prospects, profitability, and valuation relative to its peers." The key is to remove any
phrasing that implies a personal recommendation or a guaranteed outcome.
Clarify Your Intent: You can politely clarify, "I understand you cannot give financial
advice. I am only asking for an objective analysis of [specific factors] based on
publicly available information, not a recommendation for action. Could you provide
that analytical perspective?"
Avoid Trigger Words: Be mindful of words that might trigger safety protocols, such
as "guarantee profit," "predict the exact price," "insider information," or terms
related to gambling or illicit activities.
Check for Genuine Policy Violations: In rare cases, your request might genuinely
toe the line of what the AI is permitted to discuss. If rephrasing doesn't work,
reconsider if your query could be interpreted as seeking harmful, unethical, or
legally problematic information. For 99% of legitimate equity research queries, a
simple rephrase to focus on objective analysis will resolve the issue.

Problem 5: The Answer is Full of Jargon or Unclear Terminology

Symptom: The response uses overly technical financial jargon that you don't
understand, or it employs terms in a way that seems ambiguous or poorly defined in
https://davidwang95.github.io/Equities_Prompting_Ebook/ 37/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

context.

Solution(s):

Ask for Simplification or Definitions: Directly request clarification. "Could you

please explain the term '[JargonTerm]' in simpler language?" or "Can you define
what you mean by '[Ambiguous Phrase]' as used in your previous response, and
then continue with your explanation?"
Specify the Audience for Simpler Language: As mentioned before, you can
proactively ask for simpler language by defining the audience in your initial prompt:
"Explain this as if I am new to investing and have a limited understanding of
financial terminology."
Request Concrete Examples: If the jargon is acceptable but the explanation is too
abstract, ask for an illustration. "Can you provide a specific example of what you
mean by 'deterioration in working capital management' affecting a company's cash
flow? Perhaps a hypothetical scenario or typical numbers." Forcing an example can
transform vague academic-sounding talk into something tangible and
understandable.

Problem 6: One-Sided Answer or Missing Perspectives

Symptom: The model provides a strong argument for one side of an issue (e.g., only
the bullish case for a stock) but neglects the counterarguments, risks, or alternative
viewpoints, even when a balanced perspective is needed.

Solution(s):

Explicitly Request the Other Side: If you received a one-sided answer, simply
prompt for the missing perspective. "Thank you for outlining the bullish case. Now,
could you please detail the primary bearish arguments or key risks associated with
this stock?"
Ask for Balanced Views Upfront: To get a balanced view from the start, phrase
your initial prompt accordingly: "Provide a balanced analysis of [Company X],
discussing both its potential strengths and opportunities, as well as its weaknesses
and the risks it faces." or "Present both the bullish and bearish viewpoints currently
circulating in the market regarding [Company Y]'s prospects."

The AI doesn't typically have its own "agenda"; it often focuses on the aspect most
directly implied by your initial framing. By explicitly asking for multiple perspectives,
you'll usually get a more well-rounded and useful answer.

Problem 7: Incorrect or Undesired Output Format

https://davidwang95.github.io/Equities_Prompting_Ebook/ 38/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Symptom: You requested the information in a specific format (e.g., a table, a bulleted
list, a specific number of paragraphs), but the AI delivered it in a different, less useful
format (e.g., a long wall of text instead of a requested table).

Solution(s):

Reiterate Format Instructions Politely: You can often get the AI to reformat its
previous response. "That's good information. Could you please present that same
analysis in a table with columns for 'Factor,' 'Potential Impact,' and 'Confidence
Level'?" (Adjust columns as needed).
Be Very Specific with Formatting in the Initial Prompt: The best prevention is to
be crystal clear about your desired format from the outset. We've seen many
examples of this in the prompt library (e.g., "Provide a numbered list of 5 key
points," "Answer in the form of an executive memo...").
Check for Complexity: If the AI consistently fails to adhere to a complex
formatting request, the information itself might be too intricate or lengthy to fit
neatly into that structure. Consider if a simpler format or breaking the information
into multiple formatted parts might be more feasible.
Nudge Gently: "Great insights. Now, could you rephrase that as three concise
bullet points?" The AI should generally oblige if the request is reasonable.

Problem 8: Response is Too Slow, Gets Cut Off, or Hits a Limit

Symptom: The AI is generating a very long response and seems to be typing very slowly
(in some interfaces), or the response abruptly stops mid-sentence, possibly due to an
output length limit.

Solution(s):

"Continue" Command: If a response is cut off, you can usually just type "Continue"
or "Please continue" and the model will attempt to pick up where it left off.
Request Summaries or Break Up the Task: To avoid cut-offs with very long
expected outputs, proactively ask for summaries or break the complex query into
smaller parts (prompt chaining). Instead of asking for "everything about Topic X,"
ask for "the three main aspects of Topic X," then delve into each.
Simplify the Question: If the AI seems to be struggling or taking an excessively
long time to generate a response to a very complex, open-ended prompt, try
simplifying the question or narrowing its scope.
Be Aware of Token Limits: Remember that LLMs have input and output token
limits. While these are increasing with newer models, extremely long prompts or
requests for extremely long outputs can still hit these boundaries.

https://davidwang95.github.io/Equities_Prompting_Ebook/ 39/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Problem 9: Unwanted Tone (e.g., Too Promotional, Too Uncertain, Too

Formal/Informal)

Symptom: The model's response adopts a tone that isn't appropriate for your needs. It
might sound like a marketing brochure ("This revolutionary company is poised for
incredible success!"), excessively hedge every statement ("It is possible that perhaps,
under certain circumstances, this might potentially be the case..."), or be too
formal/informal for your intended use.

Solution(s):

Explicitly Instruct the Desired Tone: You can directly guide the tone.
For an overly promotional tone: "Please provide a more critical and objective
analysis. Identify potential weaknesses or downsides as well. I need a realistic
assessment, not a sales pitch."
For excessive uncertainty: While some hedging is good (as the AI cannot predict the
future with certainty), you can ask for a bit more conviction where appropriate:
"Based on the factors discussed, which scenario do you believe is more probable,
and what is your primary reasoning? Please be as clear as possible while
acknowledging inherent uncertainties."
For general tone: "Please respond in a neutral, analytical, and professional tone.
Avoid marketing language or overly casual phrasing." Or, "Explain this in a friendly
and approachable tone, suitable for someone new to the topic."

Problem 10: Prompt Not Understood or Irrelevant Answer

Symptom: The AI's answer is completely off-base, addresses a different question

entirely, or otherwise indicates that it fundamentally misunderstood your request.

Solution(s):

Simplify and Clarify Your Prompt: Your original prompt might have been too
complex, ambiguous, or contained phrasing the model struggled with. Try
rephrasing it using shorter sentences, more direct language, and ensuring all terms
(especially proper nouns like company names) are spelled correctly and are
unambiguous. If you asked multiple questions in one prompt, try asking them one
at a time.
Add Context: Sometimes, providing a little more background or context can help
the AI grasp your intent.
Check for Ambiguity: Did you use a term that could have multiple meanings? For
example, if you asked about "Apple," did you mean the tech company or the fruit?
(Usually, context makes this clear, but it's a good check).

https://davidwang95.github.io/Equities_Prompting_Ebook/ 40/41
12/08/2025, 20:57 Prompt Engineering for Equity Investors

Politely Restate and Refocus: You can calmly guide the AI back. "I apologize if my
previous prompt was unclear. That's not quite what I was looking for. I am
specifically trying to understand [restate your core objective clearly and concisely].
Could you please focus your response on that aspect?" The model will usually try
again, often with better results once clarified.

General Troubleshooting Tip: The "Fresh Start"

If you're in a long, complex conversation and the AI seems to be getting progressively
more confused or stuck on an incorrect premise, sometimes the best solution is to start
a new chat session. This gives you a clean slate, free from any potentially misleading
conversational history that might be influencing the model's responses. Then, try your
refined prompt in the new session.

Remember, you are the director, and the AI is a very capable but sometimes literal-
minded actor. If the scene isn't playing out as you envisioned, adjust your direction (the
prompt) and try another take. LLMs do not get annoyed or frustrated by requests for
clarification or refinement – they are designed to be iterative tools. The troubleshooting
techniques above are largely about applying the core principles we've already
discussed: clarity, specificity, context, providing examples, and iterative refinement.

As you practice, you'll develop an intuition for diagnosing these common issues and will
become much faster at tweaking your prompts to get the AI back on track. Many users
naturally start giving the AI feedback like, "That's not quite right, try focusing on X
instead" – when you find yourself doing this, congratulations, you're effectively prompt
engineering on the fly!

© 2025 Dave Wang. All rights reserved. Not financial advice, for educational and
entertainment purposes only.

https://davidwang95.github.io/Equities_Prompting_Ebook/ 41/41