This is general information, not financial advice.
The 5 prompts I’d run in Claude before buying any stock. They take the work out of reading an annual report front to back, and they’re built to catch the numbers Claude gets wrong.
Before you start
This is a research method, not financial advice. It doesn’t tell you what to buy, and nothing Claude says in these chats is a recommendation. If you need advice for your own situation, talk to a licensed adviser.
Give it the real documents. Turn web search on so Claude can pull the company’s filings. Better still, download the latest annual report from the company’s investor page (or the exchange announcements) and upload the PDF. The more it’s working from the real document, the fewer numbers it makes up.
Claude will state a wrong figure with complete confidence. That’s what prompt 3 is for. Don’t skip it.
Works for ASX or overseas stocks. Just watch the currency on every figure.
Step 1: Set up the chat
Start a fresh chat for each stock and paste this in first:
I'm researching [company name, ticker, exchange]. Before anything else, find the latest annual report and the most recent results announcement from the company's own investor relations page or the exchange's announcements. List the documents you're working from with their dates and links. For the rest of this chat, only use numbers you can trace to those documents, and state the currency for every figure. If you can't find something, tell me instead of estimating it.If you uploaded the annual report, add “Use the attached annual report as your main source” to the end.
You’ll know it worked when it gives you a list of actual documents with dates, not a summary of the company.
Prompt 1: What’s priced in
This works backwards from the share price to what the market already believes.
Using today's share price and the numbers from those documents, work out what the market is assuming about this business. What revenue growth and profit margins would it need to deliver over the next 5 to 10 years to justify the current price? Show your working step by step, state every assumption you had to make (discount rate, how many years of growth, growth after that), and tell me which assumption the result is most sensitive to. Then compare the growth the price implies to what the company has actually delivered over the last 5 years, and tell me plainly whether it looks realistic, stretched or conservative.If Claude’s using an old share price, give it today’s price yourself.
You’ll know it worked when you can say in one sentence what has to go right for the price to make sense.
Prompt 2: The premortem
It’s easy to only go looking for reasons to say yes. This one argues the other side on purpose.
It's 12 months from now and this stock has halved. Explain what happened as specifically as you can: what went wrong in the business, what the market stopped believing, and which warning signs were already visible in the filings today. Give me the 3 most plausible scenarios, ranked. For each one, tell me the specific number or announcement I should watch that would tell me it's starting to happen. Don't balance this with positives, I only want the case against.You’ll know it worked when you’ve got 3 scenarios and each one comes with something concrete to watch.
Prompt 3: The verification pass
The one nobody does. Run it after prompt 2, and again after prompt 4.
Go back through every number you've given me in this chat. Put them in a table with 4 columns: the figure, where you got it (document name, page or section, and date), whether you've now checked it against that source, and the result. Mark anything you calculated rather than read directly as calculated, and show the inputs. Mark anything you can't trace as unverified. Don't fix numbers quietly. If one was wrong, show the old figure next to the correct one.Then open the annual report yourself and check the 3 biggest numbers by hand. It takes 5 minutes.
You’ll know it worked when every row has a source or a flag, and your 3 spot checks match.
Prompt 4: The peer comparison
A stock is only cheap or expensive next to something.
Pick the 3 to 5 closest listed competitors to this company and tell me why you picked each one. Compare them on revenue growth, profit margin, debt, and valuation (price to earnings and EV to EBITDA), using each company's latest reported figures, with the currency and the date for every number. Then tell me what this company would need to be doing better than the others to justify the difference in valuation, and whether the numbers show it actually is.This prompt pulls in a lot of new numbers, so run prompt 3 again straight after.
You’ll know it worked when you can see where the company sits against its competitors and what explains the gap.
Prompt 5: The decision record
The one that pays off later. It writes down why you made the call, so you can check yourself against it.
Write me a one-page decision record for this stock using everything in this chat. Include: whether I'm buying or passing and at what price, my 3 reasons in one line each, what the price already assumes (from the earlier analysis), the premortem scenario that worries me most, the specific things that would make me sell or change my mind, and a date to review it. Plain language, no hype. Write it for me reading it in 12 months with no memory of this conversation.Fill in the buying or passing part yourself if Claude asks. That decision’s yours, not its.
You’ll know it worked when you’ve saved it somewhere you’ll actually find it on the review date.
Troubleshooting
Claude’s numbers don’t match the annual report. Trust the report. Rerun prompt 3 and paste in the correct figure yourself.
It keeps adding disclaimers or won’t give a view. Tell it you’re doing your own research and want analysis, not a recommendation. It’ll usually carry on.
It can’t find the annual report. Download it from the company’s investor page or the exchange announcements and upload the PDF.
Prompt 1 gives a different answer every time. The assumptions drive the result. Look at which one it said the answer is most sensitive to, and decide what you believe about that one.
It’s mixing up currencies. Tell it which currency the company reports in, and ask it to restate every figure in that currency before you go on.