Using AI to research stocks: what it does well and where to be careful

Financial research involves a great deal of reading: annual reports, quarterly filings, call transcripts, news and analyst commentary. AI tools are good at condensing that material and at answering questions about it in plain language, which can save hours. They also make mistakes, sometimes confidently. This article explains what AI does well, where to be careful, and a sensible workflow that gets the benefits while keeping the risks under control.

What AI is good at

• Summarising long documents. A 200-page annual report can be reduced to its main points, so you can decide what deserves a closer read.
• Explaining unfamiliar terms. If you meet "goodwill impairment" or "convertible notes", an assistant can explain them in context and give examples.
• Comparing companies on the same set of measures. You can ask it to line up two or three companies by growth, margins, debt and valuation.
• Spotting changes. Asking what has changed between this quarter's filing and the last can reveal shifts in language or risk factors.
• Generating questions. Good research starts with good questions. An assistant can suggest the ones you may not have thought of.
• Organising your notes into a consistent format.

Where to be careful

• Accuracy of numbers. AI can state a wrong figure as confidently as a right one. Always check the important numbers against the original filing or a trusted data source.
• Currency of information. A model's knowledge may be out of date, and without access to current data it may describe an earlier situation as if it were today.
• Confident tone. Fluent writing is a matter of style, not evidence. A well-written paragraph can still be wrong.
• Confusing similar companies or periods. Companies with similar names, different share classes, and different fiscal years can be mixed up.
• Bias toward agreement. If you phrase a question as if you already believe an idea, an assistant may follow your lead instead of challenging it.
• It does not know your situation. It cannot weigh your goals, tax position or risk tolerance the way a qualified adviser can.

A sensible workflow

1. Define the question. For example, "Is this company's growth supported by cash flow?" is better than "Is this a good stock?"
2. Gather primary sources. Annual and quarterly filings, earnings releases and transcripts. Give these to the assistant or ask it to refer to them.
3. Ask for a first-pass summary of the business, the financials and the key risks.
4. Ask follow-up questions on the areas that matter: margins, debt, competition, management's comments.
5. Verify the facts that drive your view. Check the numbers in the filing, and recompute the key ratios yourself.
6. Ask for the opposing view. Request the strongest case against, and the risks.
7. Write down your conclusion and your reasoning in your own words, so that you can review it later.

Examples of useful prompts

• "Summarise this company's business model in plain language, including how it makes money and who its main customers are."
• "Compare revenue growth, operating margin and free cash flow for these three companies over the last five years, and show where each number comes from."
• "What changed in the risk factors section compared with last year's filing?"
• "List the three biggest risks to this company's earnings and what evidence in the filing supports each."
• "What questions should I ask that this summary does not answer?"

What AI should not decide for you

An AI tool can inform your research, but the decision about what to buy, sell or hold is yours. It should not replace thinking about your goals, your time horizon, how much of your portfolio an idea deserves, or how you would react to a loss.

Common mistakes

• Treating a summary as if it were the source.
• Acting on a number without checking it.
• Asking leading questions and accepting the answer.
• Using it to find a story that confirms what you already think.
• Skipping the step of writing your own conclusion.

A short checklist

• I started with a specific question.
• I used primary sources wherever possible.
• I verified the numbers that matter.
• I asked for the counter-argument.
• I made the decision myself.

AI speeds up the research and widens what you can cover. The judgement stays with you.