Asking better questions: how to prompt AI for useful investment research

An AI assistant works with the question you give it. Vague questions tend to produce generic answers, while precise ones produce answers you can check and use. Learning to ask well is one of the most valuable skills for using AI in investment research. This article offers habits and example prompts that turn vague queries into specific, verifiable research.

Why the question matters

Ask "Is this a good stock?" and you will get a general discussion that could apply to almost any company, probably with a balanced-sounding conclusion. Ask "How have this company's operating margin and free cash flow changed over the last five years, and what drove the change?" and you get something concrete that you can verify. The second question names the measure, the period and the reason you care.

Habit 1: be specific

Include:

• The company by name and ticker, and the share class if relevant.
• The measure you care about, such as operating margin, revenue growth or net debt.
• The period, such as the last five fiscal years or the latest quarter.
• The purpose, such as comparing with peers or judging dividend safety.

Compare "Tell me about this company's debt" with "What is this company's net debt as of its latest annual report, how does it compare with its operating earnings, and when do the main loans mature?"

Habit 2: ask for the working and the sources

Because AI can state errors fluently, ask it to show its work.

• "Show the figures you used and where each one comes from."
• "Explain the steps behind your conclusion."
• "Which parts of your answer are facts from the filing and which are your inference?"
• "What would change your answer?"

Then actually check the sources it names. If it cannot point to one, treat the claim as unverified.

Habit 3: ask for the other side

It is easy to collect only evidence that supports what you already believe, and an assistant that follows your framing can make this worse. Ask for balance directly.

• "What is the strongest argument against owning this company?"
• "What would a sceptical analyst say about these results?"
• "List the main risks and what evidence in the filings supports each."
• "What assumptions does the bullish case depend on?"

Habit 4: give context and constraints

Tell the assistant what you need, in what form.

• "Explain this as if I am new to investing."
• "Give me a table with five columns: year, revenue, operating income, margin, and growth."
• "Keep it to five bullet points, and flag anything you are unsure of."

Habit 5: break big questions into steps

Instead of one huge request, work through a sequence: first the business model, then the growth drivers, then profitability, then the balance sheet, then valuation, then risks. Each step is easier to check, and an error in one is less likely to contaminate the others.

Habit 6: iterate

Treat the first answer as a draft. Follow up on anything surprising, ask for definitions of terms, and request comparisons with similar companies. If an answer seems too neat, ask the assistant to reconsider the point.

Habit 7: check the date

Ask what date the information is from, and give the assistant the documents you want it to use. This reduces the risk that it uses outdated data or makes up details.

Example: from weak to strong

• Weak: "Is this dividend safe?"
• Better: "For this company, show dividends paid, free cash flow and net income for the last five years. Calculate the payout ratio on both earnings and free cash flow, and tell me whether the dividend has been covered each year. Note any years in which the dividend was cut."

• Weak: "Compare these two companies."
• Better: "Compare these two companies on revenue growth, operating margin, return on invested capital and net debt to EBITDA over the last five years. Give the figures in a table, state the source for each, and then list two reasons why the comparison might be misleading."

What to avoid

• Leading questions such as "Why is this a great investment?" which invite agreement.
• Asking for predictions of prices or returns. No one can reliably forecast them, and an assistant's confident forecast has no special basis.
• Sharing sensitive personal information, such as account numbers.
• Treating the output as advice tailored to you.

A short checklist

• My question names the company, the measure and the period.
• I asked for sources and reasoning.
• I asked for the case against.
• I will verify the key numbers myself.

A good question narrows the problem, and verification makes the answer reliable. Together they turn an assistant from a source of plausible text into a useful research partner.