How to analyze a stock with AI

Analyzing a stock used to mean reading 10-Ks, watching earnings calls, comparing financial ratios, and cross-referencing news, work that took hours per ticker. AI tools like Claude and ChatGPT compress the same work into a focused conversation. This guide walks through a repeatable five-step framework you can use on any stock today.

The five questions AI should answer for every stock

Before you put money behind a ticker, you want clear answers to five questions. Use these as your starting prompts; refine each one as the conversation unfolds.

What does this company actually do? If you can't explain it in two sentences after the answer, keep asking.
How does it make money? Which segments drive revenue and which drive profit? Are those different from each other?
Why might it grow from here? Real growth drivers, not platitudes. New geographies, new products, pricing power, regulation.
What could break the thesis? The single best question. Competition, regulation, key-customer concentration, margin pressure.
How is it priced relative to the answer? Forward earnings, free cash flow yield, comparable peers.

The five-step analysis workflow

1. Frame the question for the AI

Don't ask “is NVDA a good stock”, too open. Ask “summarize NVDA's most recent earnings, what surprised analysts, and what guidance management gave for the next quarter.” Specific prompts produce specific answers.

2. Pull the public record

Ask the AI to summarize the latest 10-K and 10-Q in plain English. Have it list the top three risks management itself disclosed. These are the risks the company's lawyers thought big enough to print. Take them seriously.

3. Compare against peers

“Compare NVDA's gross margin and revenue growth to AMD and AVGO over the last four quarters.” AI is excellent at this kind of cross-company comparison, which is tedious by hand.

4. Stress-test the bear case

Tell the AI: “Argue the strongest bear case for NVDA in 200 words.” You're not trying to convince yourself it's bad, you're trying to find out whether the bear case is reasonable. If it is, you know the risks you're taking.

5. Connect Walnut and check portfolio fit

Sign up for Walnut, connect your broker and AI tool. Now ask: “If I add NVDA to my AI Infrastructure basket at 25%, what does my overall portfolio look like, concentration, sector exposure, theme overlap?” The AI sees your real holdings and answers concretely.

What to verify by hand

AI is fast, but it can hallucinate specific numbers. Always cross-check the data points your decision actually depends on:

Quarterly revenue and earnings figures, verify against the company's investor-relations page.
Guidance numbers, verify against the actual earnings release or 8-K.
Insider buying or selling, check on EDGAR or your broker.
Recent ratings changes, check a primary source, not an AI summary.

Try it in Walnut

With Walnut, your AI also sees how a stock fits in your real portfolio. Free to start.

Create a free Walnut account

FAQ

What AI tools are best for stock analysis?

Claude and ChatGPT are both strong for stock analysis, they can read filings, summarize earnings, and compare peers. Connect either to Walnut so the AI also sees your portfolio context (existing positions, weights, recent performance) when it answers.

Can AI replace a stock-research subscription?

Partly. AI can replace summarization and comparison work that paid services charge for. It cannot replace primary data (financials databases, real-time prices, alternative data). For most retail investors, AI plus public filings plus Walnut covers the analysis work needed.

What questions should I always ask AI when analyzing a stock?

Three: what does this company do (in two sentences), how does it make money (and is that defensible), and what's the most likely reason this thesis could be wrong. Asking the third question, the bear case, is where most investors leave money on the table.

How is this different from just searching the news?

AI summarizes across many sources, weighs recency, and answers your specific question instead of giving you a list of links to read. It also remembers your context across the conversation, so you can iterate on a thesis without re-explaining it.