How to research a stock with AI

The seven steps

1. Step 1

   Start with the business, not the chart

   Before any AI output, answer in one sentence what the company sells and who pays for it. If you cannot, no amount of analysis will help — the numbers will have no context to sit in.

2. Step 2

   Pull provider-verified fundamentals

   Revenue, earnings, margins, free cash flow, and share count. Use a tool that names its data source and returns the same figures on a second run. Note anything odd — a stock split, a one-off charge, a negative earnings quarter.

3. Step 3

   Check the trend, not the snapshot

   One quarter tells you almost nothing. Look at several quarters of revenue growth and margins together, and ask whether the direction is improving, flat, or deteriorating.

4. Step 4

   Value it with more than one method

   Earnings-based multiples break down for unprofitable companies, so pair them with a sales-based or cash-flow-based view. Where the methods disagree, the disagreement itself is the insight.

5. Step 5

   Model bear, base, and bull scenarios

   Write down the assumption behind each case — growth rate, margin, exit multiple. A single price target hides the risk; three cases show you how much you are relying on things going right.

6. Step 6

   Compare against analyst consensus

   Not to follow it, but to find out where you differ and why. If your base case is far above consensus, name the specific reason. If you cannot, revisit your assumptions.

7. Step 7

   Size the position against your portfolio

   Check sector concentration and overlap with what you already hold, then decide the position size before you buy. Finish by writing a two-line thesis and the condition that would prove you wrong.

Prompts that actually help

Notice what these have in common: each one asks the model to state its inputs. A prompt that asks for a conclusion without inputs gives you something you cannot check.

Step two in practice looks like AI stock analysis: earnings breakdown — the six layers of an earnings report and how to keep an AI read honest.

This workflow runs in any tool, which makes it a fair way to test two of them side by side — see best stock research apps: how to compare them for the five categories of research app and six checks to run before you commit to one.

Five mistakes to avoid

• Accepting a number without checking which provider or filing it came from.
• Asking for a price prediction instead of a scenario range.
• Letting the AI write the thesis for you, so you have nothing to test later.
• Ignoring share count changes after buybacks or a stock split, which distorts per-share figures.
• Researching in isolation and only thinking about position size after you have already bought.

Keep reading