How to Use Claude for Stock Analysis (Step-by-Step)

Claude doesn’t have a Bloomberg terminal. It doesn’t pull live prices. What it does have is the ability to compress three hours of fundamental research into fifteen minutes — if you know how to prompt it.

Most traders who try AI for stock analysis get garbage output because they ask garbage questions. ’What do you think about Tesla?’ is not a research process. It’s a search bar with extra steps. The traders extracting real edge from Claude are treating it like a senior analyst: feeding it structured data, asking for specific frameworks, and stress-testing their own thesis against it.

This guide gives you the exact workflow — from initial screening logic to pre-earnings positioning — with copy-paste prompts that produce actionable output. No theory. No preamble. Just the process.

Why Claude Outperforms Generic ChatGPT Prompts for Equity Research

Claude’s architecture favors long-context reasoning, which matters enormously in equity research. You can paste an entire 10-K section, a full earnings transcript, or a competitor’s press release and ask Claude to synthesize it against a specific investment question. GPT-4 handles this too, but Claude’s responses on financial documents tend to be more structured and less prone to confident fabrication on numerical specifics — a critical failure mode when you’re making capital allocation decisions.

The real advantage isn’t Claude vs. GPT. It’s structured prompting vs. conversational prompting. Analysts who treat Claude as a dialogue partner — iterating on outputs, pushing back on conclusions, asking ’what would make this thesis wrong’ — get qualitatively different results than those running one-shot queries. The workflow below is built around that iterative model.

One concrete benchmark: running a DuPont decomposition on a mid-cap industrial manually takes 45–60 minutes of spreadsheet work. Claude can produce the full framework narrative, flag the ROE driver (margin expansion vs. leverage vs. asset turnover), and identify the one number most likely to revert — in under two minutes, if you feed it the right inputs.

Step 1 — Define Your Analytical Frame Before You Open Claude

The single biggest mistake: opening Claude and typing a ticker. Before you touch the prompt window, you need to know what decision you’re making. Are you evaluating a new long idea from a screener? Sizing into an existing position ahead of earnings? Checking whether a thesis is still intact after a guidance cut? Each question requires a different analytical frame, and Claude will follow whatever frame you give it — including a vague one.

Map your question to one of four research modes: (1) Discovery — you have a ticker and need to understand the business model and competitive position from scratch. (2) Valuation — you understand the business and need to pressure-test whether the current price implies a reasonable return. (3) Catalyst — you have a view on a near-term event and need to structure the risk/reward. (4) Surveillance — you own the stock and need to monitor whether the thesis is intact.

Each mode gets a different base prompt structure. Mixing them produces muddled output. If you’re in Discovery mode on a name like $AXON or $CELH, you want Claude building context, not running DCF math on numbers it doesn’t have. Lock the mode first.

• Discovery: business model, moat, revenue drivers, key risks — no valuation yet
• Valuation: multiples context, historical ranges, implied growth rates, margin assumptions
• Catalyst: event identification, consensus expectation, upside/downside scenario framing
• Surveillance: thesis checklist, recent data vs. original assumptions, position sizing logic

Step 2 — Feed Claude the Right Raw Material

Claude’s output quality is a direct function of input quality. For fundamental analysis, the highest-value inputs are: the most recent earnings call transcript (investor.relations pages or Seeking Alpha transcripts work), the MD&A section of the latest 10-Q, and any recent sell-side initiation reports you have access to. You don’t need all three — even one well-chosen input transforms the output.

Paste the raw text directly into the prompt window. Don’t summarize it yourself before feeding it — you’ll introduce your own biases and omit the details Claude would have caught. If the document is too long, prioritize the MD&A, the risk factors section, and management’s forward guidance language. Those three sections contain 80% of the signal in any 10-Q.

For macro-sensitive names — think $XOM, $FCX, or any rate-sensitive REIT — also paste in the relevant macro context: the last Fed statement paragraph, a commodity price summary, or the most recent CPI print. Claude will integrate it if you give it to the model; it won’t invent it if you don’t.

You are a senior equity analyst. I am going to paste the MD&A section from [COMPANY]'s most recent 10-Q. After reading it, do the following:
1. Identify the three most important forward-looking statements management made.
2. Flag any language that represents a change in tone from typical corporate guidance (hedging, new risk disclosures, omissions).
3. Summarize the key revenue and margin drivers management cited.
4. List two questions you would ask management on the earnings call based on what they did NOT address.
Here is the MD&A text: [PASTE TEXT]

Step 3 — Run the Valuation Pressure-Test

Claude won’t pull live price data. Work around this by providing it yourself: current market cap, enterprise value, trailing and forward revenue, EBITDA, and the consensus EPS estimate. These are available in thirty seconds on any finance data site. Once Claude has those numbers, it can run valuation logic that would take a junior analyst an hour.

The most valuable valuation prompt isn’t ’is this stock cheap?’ — it’s ’what does the current price imply about future growth, and is that assumption reasonable given the company’s history and competitive position?’ This reverses the usual DCF direction and surfaces the embedded assumptions the market is pricing in. For a stock like $CRWD trading at 15x forward revenue, the question becomes: what five-year revenue CAGR and terminal EBITDA margin does this price require, and has any software company of this scale actually sustained that trajectory?

Ask Claude to run three scenarios — bear, base, bull — with explicit numerical assumptions for each. Force it to state the single variable that most distinguishes the bull case from the bear case. That variable is your thesis risk. Track it every quarter.

I am analyzing [COMPANY TICKER] for a potential long position. Here are the current financials:
- Market Cap: $[X]B
- Enterprise Value: $[X]B
- TTM Revenue: $[X]B | Forward Revenue Estimate: $[X]B
- TTM EBITDA: $[X]M | EBITDA Margin: [X]%
- Forward P/E: [X]x | EV/Sales: [X]x

Please do the following:
1. Back into the implied 5-year revenue CAGR and terminal EBITDA margin this price requires, assuming a 10% discount rate and 20x terminal EBITDA exit.
2. Compare those implied assumptions to the company's historical 3-year performance.
3. State the single most important assumption separating a bull outcome from a bear outcome.
4. Give me a bear, base, and bull price target in 18 months with the key variable assumption for each.

Step 4 — Build the Pre-Earnings Checklist

Earnings events are where retail traders bleed money and prepared traders print it. The edge isn’t predicting the number — it’s understanding what the market is positioned for and identifying the specific line items that will determine the reaction. Claude is exceptionally useful for building this checklist if you feed it the right context.

Two inputs matter most here: the consensus estimate table (revenue, EPS, and any key operating metric like DAUs, same-store sales, or bookings) and the stock’s recent implied volatility. An ATM straddle pricing a 9% move on $META earnings tells you the market expects significant movement — your job is to determine whether the actual catalyst distribution is skewed toward the upside or downside relative to that implied move.

Have Claude map out the three to four metrics that most moved the stock on the last four earnings reports. This creates an empirical reaction function — you’re not guessing what matters, you’re using historical data to define the market’s attention vector. Then build your scenario matrix around those specific variables.

• Identify the consensus estimate for revenue, EPS, and one operating KPI specific to the business model
• Pull the last four earnings reactions (day-of % move) and the metric that drove each
• Note current implied volatility vs. historical average IV into earnings — elevated IV changes the options strategy entirely
• Map management credibility: have they beat/missed on guidance vs. reported metrics over the past 8 quarters
• Identify any sector or macro read-throughs that arrived in the two weeks before the print

Step 5 — Use Claude to Steelman the Other Side

Confirmation bias is the most expensive bug in any investment process. Claude is uniquely useful as a devil’s advocate because it has no emotional attachment to your position. Once you’ve built your bull thesis, feed it back to Claude with an explicit instruction: destroy it.

The steelman prompt is not ’what are the risks to my thesis’ — that produces a generic risk factors list. The steelman prompt is: ’Given the bull thesis below, construct the strongest possible bear case using only data points and competitive dynamics that are publicly available and already partially visible in recent filings.’ That framing forces Claude to produce a coherent counter-narrative rather than a disclaimer section.

Run this before every new position and before adding to an existing one. The output won’t always change your decision — but it will sharpen your conviction, clarify your exit conditions, and prevent you from being surprised by arguments you should have already stress-tested. The $NFLX long that survives a rigorous steelman is a fundamentally different position than one that was never challenged.

Here is my bull thesis on [COMPANY]:
[PASTE YOUR FULL THESIS — 3-10 sentences]

Your job is to construct the strongest possible bear case against this thesis. Requirements:
- Use only publicly available data and dynamics already visible in recent filings or news
- Do not give me generic risk factors — build a coherent counter-narrative
- Identify the one assumption in my thesis most likely to be wrong
- Tell me what a short seller who knows this company deeply would say about my thesis
- End with: the specific data point or event that would confirm the bear case is playing out