5 minutes of estimated reading time
AI can be a useful research assistant when you are looking into a stock, but it is a poor oracle. Use it to understand a company and to challenge your own thinking, and do not use it to tell you what to buy.
AI chatbots can explain a balance sheet in seconds, summarize a long report, and answer follow-up questions at midnight. They can also state a wrong number in the same confident tone as a right one. In this article, we will look at what AI does well, where it goes wrong, and a simple workflow that keeps you in control.
What AI Does Well
Take Sarah, a 32-year-old renter who invests $400 a month, with $40 of it going into individual stocks. She is curious about one company and wants to understand it before she puts money in. AI can help her in several ways:
Explaining. It can turn jargon such as "free cash flow" or "operating margin" into plain English, and it will answer follow-up questions without getting impatient.
Summarizing. If she gives it a long section of a company's annual report, it can condense the key points.
Organizing. It can turn a pile of notes into a checklist of what to look into next.
Playing devil's advocate. It can lay out the strongest case against a stock, which is the case most of us are least keen to hear.
Comparing. It can describe how two companies in the same industry make their money differently.
Notice that every item on this list is about understanding and questioning. None of them is about predicting.
Where AI Gets Things Wrong
AI is useful, but it has predictable weak spots:
Invented details. A chatbot can produce a specific revenue figure, date, or quote that sounds right and is not. It can even invent the source it claims to be citing.
Outdated information. Unless it is connected to live data, it knows only what it learned up to a certain date, so it may miss the latest quarter, a new lawsuit, or a recent acquisition.
Slips in arithmetic and units. It can confuse millions with billions, quote the wrong year, or miscalculate a ratio.
Agreeing with you. If your question signals what you hope to hear, an AI may lean toward saying it.
No view of the future. It cannot know where a share price is headed, and neither can anyone else.
There is evidence that financial documents are a particular challenge. In late 2023, the company Patronus AI released FinanceBench, a set of 10,000 question-and-answer pairs based on public filings. In its tests, GPT-4 Turbo with a retrieval system failed 81% of the questions in a sample, and models that could read long documents directly failed about 21% to 24%. The company's chief executive said the margin of error was too big for financial applications.
Those results are a snapshot from 2023, and AI tools have improved since. But the lesson has not changed. In investing, a small error in one number can change a conclusion, so the numbers need to be checked.
A Five-Step Workflow
Bring your own data. Paste in the text or figures from the company's actual filing instead of asking the AI to recall them. This removes the biggest source of invented numbers.
Ask it to explain and to point to the source. Request a plain-English summary, and ask it to quote the sentence or table each point comes from.
Ask for the opposite case. After it explains why a company looks good, ask for the strongest argument against it, and what evidence would change its view.
Check the math yourself. Any ratio that matters should be recalculated with a calculator or spreadsheet. Better still, ask the AI to show its steps so you can follow them.
Verify the key facts at the source. Before you act on anything important, confirm it in the filing itself. Companies' annual and quarterly reports are available to the public on the SEC's EDGAR website.
If you take only one habit from this list, make it the last one. The AI helps you think, and the filing tells you what is true.
Example Prompts You Can Adapt
Good prompts supply the data, ask for evidence, and invite disagreement. Here are five you can copy and adjust:
Here is the Risk Factors section from this company's latest annual report. Summarize the five most significant risks in plain English, and quote the sentence each one comes from.------------------------------------------------------------------------------------------------------
Explain how this company makes money, as if I am new to investing. If you are unsure of anything, say so instead of guessing.------------------------------------------------------------------------------------------------------
Here are the income statement and balance sheet figures. Calculate revenue growth, gross margin, current ratio, and debt to equity, and show each calculation step by step.------------------------------------------------------------------------------------------------------
Give me the strongest case against owning this stock, and tell me what evidence would change that view.------------------------------------------------------------------------------------------------------
List the questions I should research before making a decision, ranked by importance.Each prompt does at least one of three things: it gives the AI the real documents to work from, it asks to see the reasoning or the source, or it asks for the other side of the story. Prompts that ask "Is this a good stock to buy?" do none of them, and they invite a confident answer that nothing supports.
A Worked Example
Let us say Sarah is looking at Brightleaf Coffee Co., a fictional coffee chain invented for this example. She pastes in these figures from its latest annual report, in millions of dollars: operating cash flow of $55, capital spending of $45, current assets of $140, and current liabilities of $90. Then she asks the AI to calculate free cash flow and the current ratio, showing its steps.
Because she supplied the numbers, she can check the answer in seconds:
Question | Correct answer | What to do if the AI disagrees |
|---|---|---|
Free cash flow (operating cash flow minus capital spending) | $55 minus $45, or $10 million | Ask it to redo the steps and point out where they differ |
Current ratio (current assets divided by current liabilities) | $140 divided by $90, or about 1.56 | Recalculate on a calculator and trust your own result |
If the AI answers $100 million or a ratio of 1.2, Sarah has caught an error before it influenced a decision. That is the whole point of the workflow: the AI does the first pass, and she keeps the final check.
A Warning About AI and Investment Scams
AI has also become a marketing hook for fraud. In January 2024, the SEC, FINRA, and NASAA issued a joint investor alert on AI and investment fraud. It warns about:
Promises of guaranteed returns. Claims of high guaranteed returns with little or no risk are, in the alert's words, classic warning signs of fraud.
Unregistered platforms and promoters pushing "AI trading systems."
AI-themed stock hype, sometimes through pump-and-dump schemes, especially with tiny companies about which little public information exists.
Deepfakes and impersonation, where AI is used to clone voices and videos of real people.
Overreliance on AI output, which the alert notes can be inaccurate, outdated, or manipulated.
The alert's practical advice is worth following: check that a professional or platform is registered, review a company's disclosures on EDGAR, compare promotional claims with the company's real business, and confirm information against more than one source.
Common Mistakes to Avoid
Asking the AI to recall numbers from memory. Supply the figures, and check them against the filing.
Asking "should I buy?" The question has no reliable answer, and it hands your judgment to a tool that cannot see the future.
Accepting the first answer. Ask follow-ups, ask for sources, and ask for the opposite view.
Feeding it your conclusion. If you tell it you love a stock, it may politely agree. Ask neutral questions.
Trusting "AI-powered" products with big promises. A label does not make a strategy safe.
Skipping the source. If you would not bet your own money on a number, do not trust it until you have checked it.
The Short Version
Use AI to explain, summarize, organize, and challenge, and do not use it to predict or to decide.
Its weak spots are invented details, outdated information, arithmetic slips, and agreeing with you.
Bring your own data, ask for sources and steps, and ask for the opposite case.
Recalculate the numbers that matter, and verify key facts in the filing.
Be wary of any product that promises guaranteed returns with "AI."
Used this way, AI can make you a faster and more curious researcher. The final judgment, and the responsibility for it, remain with you.
Disclaimer: This article is for educational purposes only and does not constitute financial, legal, or tax advice, and it is not a recommendation to buy or sell any security. AI-generated information can be inaccurate, incomplete, or out of date, and it should always be verified against original sources. Brightleaf Coffee Co. is a fictional company, and its figures are invented. All investing involves risk, including the possible loss of principal. Consider speaking with a qualified professional before making financial decisions.
