How I use AI to research a stock from scratch, in one sitting

Everybody starts to reserach a stock with good intentions

You get through the first part, start forming a view and then something else takes priority .

Work, a call, life in general..

You come back to it two days later and you’ve lost the thread. So you skim the rest instead of reading it properly, glance at a couple of headlines,

and make a decision on maybe sitxy percent of the picture.

Two things can happen from here and both are versions of the same regret.

Either you buy anyway, on the incomplete picture,

and the stock does something you didn’t see coming because you never got to the part of the filing that would have told you.

Or you close the tab and tell yourself you’l come back to it properly this weekend, and you never do,

and six months later the stock is up sixty percent and you’re the person at dinner saying “yeah, I looked at that one.”

Neither outcome is really about being wrong.

It’s about never finishing the research in the first place.

That risk never fully goes away, a stock can move against a perfect research process just as easily as against a rushed one.

But the specific regret of not having had the time is the fixable part and that’s the part AI can help solve.

Not by predicting the stock, but by compresing the hour you don’t have into the twenty minutes you do.

Here’s the AI workflow I run instead, the one that gets me from ticker to opinion in one sitting, without needing three uninterrupted hours to do it.

The order matters more than the depth

Most of the value in research comes from asking the right question at the right stage, not from reading more.

If you start with valuation, you’ll rationalize whatever multiple you’re looking at.

If you start with the bear case, you’ll talk yourself out of good businesses before you understand them.

The sequence below moves from understanding to numbers to risk to price, in that order, because that’s the order that keeps each step honest.

Five steps. About thirty minutes with the filing open next to your AI tool.

It turns a ticker into a position you can actually defend.

Step 1: How does this business actually make money

Before anything else, I want the plainest possible answer to the plainest possible question.

Not the mission statement. Not the investor deck language.

Where does the revenue physically come from, how concentrated is it, and what would have to keep happening for that to continue.

Prompt: the plain-English business model

Read the business description and revenue segment data below.
Explain in plain language how this company actually makes money.
Break down revenue by segment or product line, and tell me how concentrated it is,
whether in customers, geographies, or product categories.
Then tell me what has to keep being true, structurally, for this revenue to continue.
Not risks in general, the specific mechanism that generates the money.
If the business model is more complicated or fragile than the company’s own description suggests,
say so directly.

Step 2: Where are the numbers actually trending

Once I understand the business, I want the trend, not the snapshot.

A single quarter tells you almost nothing.

Margins, revenue growth and returns on capital over the last several years tell you whether the business is getting stronger or whether the story is doing more work than the numbers.

Prompt: the trend behind the numbers

Using the financial data below, show me the trend over the last three to
five years for revenue growth, gross and operating margins and return on invested capital.
Tell me whether the trend is improving, flat, or deteriorating, and identify the point,
if there is one, where the trajectory changed.
Separate organic performance from anything driven by acquisitions, buybacks,
or one-time items. I want to know how the core business is actually doing,
not how the reported numbers look.

Step 3: What’s the one thing that breaks this

Every filing has a risk factors section, and none of it is specific enough to act on.

I want the AI to find the single most likely point of failure, not a list. This forces precision, and precision is what turns a risk factor into something you can actually watch for in the next few quarters.

Prompt: single point of failure

Based on the business description and financials below,
identify the single most likely failure point for this thesis over the next two to three years.
It should be specific and falsifiable,
something that either happens or doesn’t, not a vague concern like “competition increases.”
Then explain the earliest signal that would show this failure point starting to happen.
What would show up first, in the numbers or in company commentary, before the thesis is
obviously broken.

Step 4: Who else is exposed to the same thing

A company rarely fails or succeeds in isolation. Suppliers, competitors, and customers are usually further along the same curve, and their numbers often show the turn before this company’s do.

I ask the AI to map the compnay against its closest comparisons, on fundamentals rather than on how the market currently prices them.

Prompt: comparison against peers

Compare this company to its two or three closest competitors on revenue growth, margins,
and returns on capital, using the data below.
Tell me where this company is genuinely ahead, where it’s behind,
and whether any premium or discount in its current valuation relative to peers looks
justified by the fundamentals or looks like it’s coming from sentiment.
If the comparison isn’t clean, for example different business mixes or reporting periods,
tell me that instead of forcing a comparison that doesn’t hold up.

Step 5: What does the price already assume

Last question, and the one that ties everything together.

A stock isn’t chaep or expensive in the abstract.

It’s priced for a specific outcome .

I want the AI to work backward from the current price and tell me what growth, margins, and timeline would actually need to happen to justify it,

then compare that to what the company has delivered so far.

Prompt: reverse engineering the price

Based on the financials and current valuation below, work backward and tell me what growth rate,
margin trajectory, and time horizon the market is effectively pricing in for this stock to justify
its current multiple.
Compare that to what the company has actually delivered over the last four to six quarters.
Don’t tell me whether the stock is cheap or expensive in the abstract.
Tell me specifically what has to be true for the current price to make sense,
and how big the gap is between that and reality.

By the end of these five steps, I’m holding the specific numbers and mechanisms behind the stock.

Sometimes the picture holds together and I move to sizing the position. Sometimes step three or step five quietly kills the thesis before I’ve committed any capital.

Either way, I know exactly why and I know what to watch for next quarter.

That’s the actual point of this process.

Not to make research faster for its own sake,

but to make sure the opinion you end up with is one you could explain to someone else, line by line,

instead of one that just felt right after an hour of tabs.

These prompts work best with the actual real filing data in front of the AI, not a summary you’ve already written yourself.

The less pre-digested the input, the more honest the AI’s output.

Here you can find my compilation of best to use Claude prompts for stock research,

Interesting thing is that you can be completely right about a business and still lose money on the stock.

Being right about the thesis and making money from it are two different skills and the five steps above only cover the frst one.

What follows below is the part that gives you even more to protect the money with, the difference between a good call on paper and a good call that pays out.

Below you can find three more steps I run after the first five, before any money moves.

Read more