AI Prompts That Actually Work for Investment Research

The Prompt Problem

Most people get disappointing results from AI because they give vague prompts and get vague responses.

"Summarize this manager letter" produces a generic summary. "Tell me about private equity" produces a Wikipedia-style overview. "Write a memo about this investment" produces bland corporate prose.

Better prompts produce better results. Here's how to write them.

The Framework

Good prompts have four components:

1. Context Tell the AI who you are and what you're doing. This shapes the response style and assumed knowledge level.

"I'm a portfolio analyst at a foundation with a $500M endowment. I'm preparing for our investment committee meeting."

2. Task Be specific about what you want. Not "analyze" or "summarize"—what specifically should the output contain?

"Extract the following from this manager letter: (1) key portfolio changes since last quarter, (2) the manager's current market outlook, (3) any mentioned risks or concerns."

3. Format Specify how you want the output structured. Bullet points? A table? Specific sections?

"Present this as a bullet-point summary, with no more than 3 bullets per section."

4. Constraints Limitations or requirements. Tone, length, things to include or exclude.

"Keep the total summary under 300 words. Focus on information relevant to our decision whether to maintain or reduce our allocation."

Prompts for Common Tasks

Summarizing Manager Letters

A basic prompt like "Summarize this manager letter" produces generic results. Here's a better approach:

I'm a portfolio analyst reviewing quarterly manager letters for our investment committee. From the attached letter, extract: (1) Performance summary with return vs benchmark and attribution, (2) Significant portfolio changes including new positions, exits, and sizing changes, (3) Manager's forward outlook and positioning, (4) Any red flags or concerns mentioned. Format as bullet points, with the most important items first. If something isn't clearly stated in the letter, note that rather than inferring.

Comparing Multiple Documents

Instead of "Compare these two manager reports," try this:

I have quarterly reports from two emerging market equity managers in our portfolio. Compare them across: market outlook (where do they agree/disagree?), top holdings overlap, risk positioning (defensive vs aggressive), and performance explanation. Present as a two-column comparison table, followed by a brief note on the most significant differences.

Drafting Meeting Notes

Rather than "Write meeting notes from this transcript," be specific:

This is a transcript of our meeting with [Manager Name] on [date]. Produce meeting notes that include: (1) Attendees, (2) Key discussion points in bullet format, (3) Manager's answers to our key questions, (4) Action items or follow-ups mentioned, (5) Our team's impressions if expressed. Use professional tone suitable for our internal files. Attribute specific statements to speakers where relevant.

Preparing for Manager Meetings

Don't just ask "Help me prepare for a meeting with a hedge fund manager." Instead:

I'm meeting with [Manager Name], a long/short equity manager we've been invested in for 3 years. Based on their recent performance (+5% vs benchmark -2%) and the attached quarterly letter, help me prepare: (1) Five specific questions about their positioning, (2) Three potential concerns to raise, (3) Key data points I should review before the meeting. Frame questions as open-ended and non-confrontational.

Investment Memo First Drafts

"Write an investment memo" is too vague. Try:

I need to draft an investment committee memo for a potential new manager allocation. Details: [Manager name, strategy, proposed allocation size, basic rationale]. Structure the memo with these sections: Executive Summary (one paragraph), Manager Overview (background, team, AUM, strategy), Investment Rationale (why this manager, why now), Risk Considerations (what could go wrong), and Recommendation. Use formal tone consistent with our IC process. I'll add specific data and refine—this is a first draft framework.

Due Diligence Questions

Instead of "What due diligence questions should I ask?":

I'm conducting operational due diligence on a small private equity fund ($300M AUM, Fund III, based in Chicago). Generate 10 operational due diligence questions focusing on: fund administration and NAV calculation, valuation procedures for illiquid holdings, key person provisions and succession, and compliance infrastructure for a firm this size. These should be questions that reveal potential weaknesses, not softball questions. Assume this is a first institutional-quality meeting after initial interest.

Tips for Better Results

Be Specific About What You Know If you give the AI more context, it gives better responses. Instead of "analyze this manager," try "this manager has underperformed for two quarters despite claiming to be defensive—help me understand why."

Ask for Reasoning Add "Explain your reasoning" or "Show your work" to prompts. This helps you verify the output and often produces more useful analysis.

Iterate The first response is rarely perfect. Follow up with refinements: "Focus more on the risk section" or "Make the language more concise" or "What am I missing here?"

Verify Everything AI will confidently state things that are wrong. Cross-check numbers, dates, and factual claims. The AI is a research assistant, not a source of truth.

Don't Input Sensitive Data Unless you're using an enterprise AI system with appropriate protections, don't paste confidential client information, MNPI, or proprietary data into consumer AI tools.

Building a Prompt Library

Create a document with your best prompts for recurring tasks. When something works well, save it. When you improve a prompt, update the library.

Over time, you'll have a collection of prompts optimized for your specific work. This is valuable IP—and it makes onboarding new team members much easier.

Bottom Line

AI output is only as good as AI input. Invest time in crafting better prompts, and you'll get dramatically better results. Start with context, be specific about the task and format, and iterate until you get what you need.