How to research a stock with AI, step by step

To research a stock with AI, use the assistant as a research partner, not an oracle: ask it to explain the business, pull current fundamentals and earnings, summarise recent news with sources, and lay out the bull and bear case. Then write your own one-paragraph thesis and check the key claims against the company's filings. The AI speeds up the gathering and explaining. The judgement stays with you.

Here's a seven-step method that works in ChatGPT, Claude, Gemini or any assistant.

First, give the assistant real data

A plain chatbot answers from its training, which has a cutoff date. It may not know last quarter's results, this week's news or the current price. That's fine for explaining concepts, and risky for anything time-sensitive.

You have two ways to fix this:

• Use an assistant with web search, and ask it to cite its sources.
• Add a connector for your own data. An MCP connector gives the assistant what web search can't see: your holdings, watchlists and notes. Bywell, for example, provides get_holdings, get_journal (your thesis and sell conditions) and get_goals, so the assistant can relate the news it finds to what you actually own.

Either way, get into the habit of asking "what's the source and the date?" See What is an MCP server for investing? for how connectors work.

Step 1: Understand the business

Before any numbers, make sure you can explain what the company does.

• "Explain how this company makes money, in plain English. What are its main products or segments?"
• "Who are its customers, and who are its main competitors?"
• "What would have to go wrong for this business to struggle?"

If you can't explain the business to a friend in two sentences, the rest of the research won't mean much.

Step 2: Look at the fundamentals, in context

A handful of numbers go a long way: revenue growth, profit margins, the P/E ratio (price compared to earnings) and debt. None of them mean much alone.

• "Pull the fundamentals for this stock: market cap, P/E, EPS, dividend yield, 52-week range and beta. Explain each in one line."
• "How does its P/E compare to a close peer, and what might explain the difference?"

A high P/E isn't automatically expensive, and a low one isn't automatically cheap. It often reflects what investors expect about growth. Compare to the company's own history and to peers.

Step 3: Review recent earnings

Earnings reports show how the business is actually doing, and how the market reacts to it.

• "Show me the last four quarters of results against estimates. Any pattern?"
• "What did the company say about guidance in its most recent report?"
• "When is the next earnings date?"

Our guide on how to read an earnings report covers what to look for.

Step 4: Scan the news, and sort signal from noise

• "Summarise the news on this company from the last 30 days, with sources and dates."
• "Which of these stories could change the long-term picture, and which are noise?"

Signal tends to be earnings, guidance changes, leadership changes, big contracts, regulation, or shifts in the industry. Noise tends to be price-target tweaks, vague rumours and hot takes.

Step 5: Ask for the bull case and the bear case

This is the single most useful prompt in AI stock research:

• "Give me the strongest bull case and the strongest bear case for this company, with sources and dates."

Asking for both sides guards against the assistant simply agreeing with you. If you already like the stock, spend extra time on the bear case. You can push further: "What would a sceptical analyst say about the bull case you just gave me?"

Step 6: Check how it fits your portfolio

A good company can still be a poor fit if it doubles down on something you already own.

• "If I held this, how would it change my sector concentration?"
• "Do any of my existing holdings or ETFs already have significant exposure to this company?"

With a connector that reads your holdings, the assistant can answer these using your actual positions and portfolio weights.

Step 7: Write your own thesis

Finish by writing, in your own words, one or two sentences on why you would own this and what would change your mind. You can ask for help, but make it yours.

• "Help me write a one-paragraph thesis for this company, including what would break it."

A written thesis is what you'll come back to when news hits. Instead of reacting to the price, you check the news against the thesis.

Common mistakes when researching with AI

• Asking "should I buy this?" You'll get either a refusal or an overconfident answer. Ask for considerations and trade-offs instead.
• Trusting numbers without dates. A P/E from last year can be very different today.
• Letting it confirm what you want to hear. Always ask for the bear case.
• Skipping the primary source. For anything that matters, check the company's filings on SEC EDGAR or SEDAR+.
• Mixing up tickers. Confirm the company name, especially for Canadian listings, which some tools write with a .TO suffix.

This is general information, not investment advice.

A worked example, in outline

Suppose you're curious about a hypothetical company, "Example Corp," a mid-sized software business. A research session might look like this.

• Business: You learn it sells subscriptions to small businesses and has two main competitors.
• Fundamentals: Its P/E is higher than a peer's. The assistant explains that investors may be expecting faster growth, and notes what would have to be true for that to hold.
• Earnings: It has beaten estimates in most recent quarters, but guidance was trimmed last quarter.
• News: One story about a new product; several about analyst price targets, which the assistant labels as noise.
• Bull and bear: The bull case is steady subscription revenue. The bear case is slowing growth and pricing pressure.
• Fit: One of your ETFs already holds a small amount of it.
• Thesis: You write two sentences, including "I'd reconsider if subscription growth keeps slowing for two more quarters."

None of this tells you whether to buy. It tells you what you'd be buying, which is the point.

Make it a habit

You don't need to do all seven steps for every idea. Steps 1, 5 and 7, understanding the business, arguing both sides and writing a thesis, will already put you ahead of most snap decisions.

If you'd like an assistant that already knows your holdings and has news, fundamentals and earnings on hand, try Bywell. For ready-made questions, see AI investing prompts for your own portfolio.