How to use AI for stock research without getting fooled

What AI is good at, and what it isn't

Good at: summarizing long documents (10-Ks, earnings transcripts), explaining unfamiliar business models, organizing bull and bear arguments, comparing companies along fixed criteria, and spotting what changed between two quarterly reports.

Bad at: current prices and fresh numbers (unless browsing is on), predicting stock moves, and saying “I don't know.” By default, models tend to agree with whatever thesis you hint at.

Step 1: Feed it primary sources

The biggest upgrade is giving the model the actual documents instead of relying on its memory. Download the latest 10-K or 10-Q from SEC EDGAR, or the company's investor-relations page, plus the most recent earnings call transcript. Upload them and tell the model to answer only from those documents and to quote the section it used.

Step 2: Use a structured prompt

Give it a role, your context, explicit steps, and rules about sources. Our free AI stock research prompt generator builds this for you. A minimal version:

You are a skeptical equity analyst. Using ONLY the attached 10-K and earnings transcript:
1. Explain how the company makes money, by segment.
2. Summarize 3-year trends in revenue, operating margin and free cash flow.
3. Give the 3 strongest bear arguments.
4. List what management said about guidance, with direct quotes.
Cite the page or section for every number. If something isn't in the documents, say so.

Step 3: Make it argue against you

After the first answer, follow up with: “Now argue as a short seller. What is the market missing, and what would make this stock fall 40%?” This counters the model's tendency to agree with you and often surfaces risks you hadn't considered.

Step 4: Verify every number that matters

Check any figure that drives your decision (revenue, margins, debt, share count) against the filing itself or a data provider such as Seeking Alpha. Models confuse fiscal and calendar years and mix up segments more often than you'd expect.

Step 5: Run the numbers yourself

Use the AI for the qualitative story and do the math with deterministic tools. Try our CAGR calculator for historical returns and the position size calculator before you place a trade. TradingView is useful for checking the chart.

Common mistakes

• Asking “should I buy X?” You get a hedged non-answer, or worse, a confident one.
• Trusting price targets or numbers from a model with no web access.
• Leading the witness: “Why is X undervalued?” assumes the conclusion.
• Skipping the bear case.
• Treating AI output as advice. It's a research assistant, not a fiduciary.