How to use ChatGPT for Stock Analysis: FULL Tutorial 2025 | NotebookLM

Invest Insight
9 followers
If I want to analyze a chart and I don't know how to do it, I can give it a prompt and put an image of the chart and get it analyzed.
This is a very interesting use case.
Nowadays, a lot of people are thinking that there is no need to analyze the company, there is no need to understand the company.
I directly give everything to AI and ask AI when to buy and when to sell.
So can it do all this?
A little while ago, there was a company called Zen Technology Ltd.
I rejected this company for around Rs. 300 and then I bought it back for Rs. 800.
I rejected it for Rs. 300 and bought it back for Rs. 800.
The numbers that you are seeing in this chart, it is a very tedious task to find these numbers.
This data is not in the screener.
This data is not available in the screener.
This data is hidden somewhere in the annual report.
There is one more thing.
Whatever sources you have given it, you can make a podcast by doing its audio overview.
And I just ask this prompt.
Let's see what it gives by analyzing this chart.
Most of us don't know how to ask questions.
So for that, I think the first step we have to do is we have to go to chatgpt and ask that
Just before starting the podcast,
for all of you who are interested in investing,
there is a small update that Shubham is doing a webinar on Sunday.
For those who don't know about Shubham,
Shubham became financially free at the age of 29.
That's why he left his Amazon US job and shifted to Bangalore, India.
And became a full-time investor.
So if you are interested in this webinar and want to learn about Shubham's experiences,
then we have given its link in the description and first comment.
Anyway, let's start with this amazing podcast.
What is the fastest way to analyze any company by using Shubham AI?
So a while ago, there was a company called Zen Technology Ltd.
Now I had analyzed its stock for about 300 rupees.
But because I had not read that industry till date, I was reading it for the first time.
So basically at that time I realized that this company is outside my circle of competence.
Because I was seeing that company for the first time.
I rejected this company for about 300 rupees.
And then slowly I understood the company.
It took me a few months to understand the industry and what was happening.
And then I bought it back for 800 rupees.
Rejected it for 300 rupees and bought it for 800 rupees.
Now there were some events in between.
This company got some more orders and all that.
But the reality is that, as Warren Buffett says, circle of competence.
Because this thing was outside my circle of competence.
Because it was outside my circle of competence, I could not gather the courage to put in the money.
And AI may give you so many hi-fi funds to buy and sell.
But the reality is that when you are going to invest money, you will not be able to trust it.
But at that time, I did not use AI much.
But if today the same problem comes with me.
That there is an industry or a company that I do not understand.
The work that used to be done in the first few months.
To understand the in and out of the industry.
Now I feel that I can do that work easily in 3-4-5 days.
I will tell you another very interesting use case.
So if a lot of people are new to investing, they do not know how to look at charts.
So for example, if I see a chart on any trading view.
So if I understand the financial part, I have also seen the chart of Kotak Mahindra.
Now there is some volume in the chart, there is a price, I can see some things.
But according to the support resistance, there are many patterns.
According to all of them, if I have to analyze this chart, I do not know it myself.
So I actually give it a prompt and download this chart.
I can put the image of this chart and get it analyzed.
This is a very interesting use case.
So for example, here I downloaded this image.
Now if I go to chat gpt.
For example, I upload the same image of Kotak here.
And I just ask this prompt.
Now see here, this prompt will seem very complicated to you.
You are a technical analyst with a lot of experience in recognizing chart pattern indicator.
Now I asked this prompt from chat gpt.
So basically I told him that this is what I am trying to do.
I want to analyze a company's chart.
Give me a prompt for that.
So he gave me a very good prompt.
And I think this is the best way.
So the maximum mistake people make is that they are increasing efficiency by asking this prompt.
And the second is that you tell him to break it in steps.
And make a prompt for every step.
So breaking a bigger problem into smaller chunks and asking a prompt for every small chunk.
I think it will double the efficiency.
And then you can, if you feel that there is something more or less, then you can edit it.
You can do whatever you want.
So let's see what it gives by analyzing this chart.
See, it identifies support resistance properly.
Where is the immediate support?
Where is the resistance level?
What chart patterns are visible in this?
He is saying head and shoulder or double top.
But it has not been confirmed yet.
Moving averages, as we see, we do stage analysis.
According to that, it is also telling that it is above 20 DMA.
But it is also testing 20 DMA.
But when it is above 200 DMA.
According to RSI MSCD, it is overbought or oversold.
After doing all this, it gave a summary.
I think the summary is most important.
This is a high probability bullish pullback setup.
If you look at its chart, it is a good chart.
Why? Because if you see, after going to the proper momentum, there was a pullback.
And now it is almost on support.
So kind of it is a good pullback setup.
He identified it perfectly.
The 20 DMA support is critical.
If it breaks, then this acts as a support.
If it breaks after that, then this will act as a resistance.
So I think this is another way.
But it is just for the overview.
It is not that I finalize it and invest money or do something.
So whatever things you use in your journey.
For example, if you want to make an indicator.
So you can make an indicator by saying this.
For example, you want to make an indicator of stage analysis.
We do stage analysis.
Stage 1, Stage 2, Stage 3, Stage 4.
And on that basis, we categorize companies.
It is very easy.
You just have to give a prompt.
Make an indicator for me to analyze technically.
And then I can use that indicator like Pinescript.
Like we saw in the last video.
We made an indicator.
We also made strategies.
And we also back-tested the strategies.
To see which strategy is working well and which is not.
And that was just for the testing.
Many people may not think that you have to put the money there.
It is more that if you have an idea, you can test it.
So that was the thinking model.
Recently, we had a task.
We wanted to make a tool for the community.
Where we wanted to analyze all banks and NBFCs.
For example, this was the tool that we were trying to build.
All PSU banks, private banks, NBFCs, housing finance companies.
The main data of all these things.
Cost of fund.
Cost to income.
NPAs.
Price to book.
And so on.
We wanted to analyze all these things.
The problem is that analyzing all these things.
The numbers you see in it.
This is a very tedious task.
This data is not in the screener.
This data is not available in the screener.
This data is somewhere in the annual report.
For example, this is a company-specific ratio.
Or a bank-specific ratio.
Which is somewhere in the annual report.
And I have to go to the annual report every year.
I have to bring all these data.
And there is a lot of data.
ROE.
Book value.
Gearing ratio.
EPS.
AUM.
That was the problem at hand.
Collect all this data.
Visualize it.
And then we can compare.
Which bank is good.
Which bank is bad.
Which is cheap.
Which is expensive.
In fact, the maximum code in this.
It is written by him.
The whole chat was done by GPT.
This is our shipping dashboard.
In this, we have to parse the shipping data.
Based on that, we have to make a dashboard.
How much growth is going to come in which company.
How is the shipping data.
The coding work of this was also done by chat GPT.
So, I think these are some of the ways that we are using it.
But yeah, I think sky is the limit.
I think it's more about how far you can think.
Where you can use it.
Understand this.
So, now to do this.
I started doing it.
To extract the data of a company.
I did it personally.
It took me 3 hours to extract the data of a company.
Okay.
And this is when I left all the work.
I am looking for each data point in a focused way.
I am taking it out.
Because there is no data of 1 year.
There is data of the last 10-12 years.
So, I have to go to every annual report.
There is some data that is not available.
Which I have to go separately and find on the internet.
There is some data that I have to calculate.
Because it is not a standard format.
That you will get it in the form of a table.
So, it is not like that.
It is somewhere in the middle of a sentence.
Like this.
So, now I started doing it.
I started doing 2-3 companies.
I thought that this is very difficult.
It will not be possible.
And I had to do the entire sector analysis and present it on the weekend.
So, I thought that it will take me a month to collect this data.
So, I thought that I will try to help LLM.
So, as they say that it takes intelligence to cheat.
If you say it randomly.
Make a financial report for me.
Actually, you can go to YouTube.
You will get all these videos.
People will give their prompts.
They will say, analyze this company.
And after analyzing, make a complete report for me.
Okay.
And then I will invest from that report.
The reality is that he will definitely make a report.
Okay.
He will tell you everything.
That yes, this is happening.
This is happening.
But there are nuances in every company.
A lot of things are suppressed inside.
When you actually listen to a management while talking.
We have that test to understand humans.
Who is actually right?
Who is lying?
AI cannot do that thing.
Many times it happens that there will be a lot of important things.
Which you might think are important.
But AI will miss it.
It does sentiment analysis.
It does definitely.
You are right.
But it is still difficult.
There will be no conviction.
You will have to see eventually.
You need that citation.
Where are the numbers?
Correct.
So.
And the main thing is that investing is a game of putting in your money.
So will you just trust a decision made by a machine?
Unless you have the conviction.
He has 5 track records.
Which he is saying that according to that.
So there are only two things.
Either you spend time and effort.
Understand that thing.
And then invest money.
Or you actually do its back test.
Give me the full result in 10-20 years.
Then I can be confident in one way.
Whatever you are saying is right.
Then I can trust you and put in my money.
So anyways.
To do this exercise.
Like I said.
It was taking a lot of time.
So I thought.
I give this prompt.
I give the annual report of this company.
I try to extract this data from chat gpt.
Will you be able to extract?
So I did exactly the same thing.
I went to chat gpt.
And I put a basic prompt.
So as an example.
I will show you here.
Now if I put an annual report here.
And ask for a basic prompt.
Give me these 5 numbers.
So when I did this.
So actually I am getting these numbers.
But what is he doing a lot of times.
He was extracting some data from the internet.
Along with.
Like they call hallucination.
So because.
Most of us.
Don't know how to ask questions.
So for that.
The first step we have to do.
We have to go to chat gpt and ask.
I am trying to.
Find out the key ratios.
Of a bank.
From an annual report.
Please help me.
With the prompt.
To get those numbers.
Okay.
I told him to extract the prompt.
And that is because.
This prompt.
Which we will make ourselves.
We will make something basic.
Give me these 5 numbers.
So a lot of times.
He was extracting some data from the internet.
So I had to make it strict.
That you have to get the data from here.
After that.
If he extracts 10% wrong data.
Then it is of no use to me.
So in the prompt.
I have to make him understand.
That whatever you are doing.
The context behind it.
Bring 2-3 paragraphs together.
Give me the output in the form of a table.
So that I can verify that number.
Whether he has extracted that number correctly or not.
Because on this.
I will write the algorithm.
I am actually taking the decision of investing on it.
So this prompt engineering.
Is very important.
So you are a banking analyst.
Only the document I provide.
Do not use external sources.
Extract the key financial ratios.
And he told me all the main ratios.
After the table.
Show the last 2 paragraphs of the document.
For source verification.
Do not ask any clarifying questions.
Or repeat the prompt.
Just output the table.
And add the 2 paragraphs.
So he made a very good prompt.
If I would have done it myself.
I would not have been able to make such a good prompt.
So now basically I can just give this prompt.
Add an annual report.
And actually get the data for.
Any company.
So for example.
Let's see what he gives me.
So in this.
There are many different models.
Like if you.
Like in this very good.
Proper tabular form.
He gave me all the numbers.
The best thing he did.
The data that he did not get.
Or did not know.
He said not provided.
Explicitly if he is not getting in that annual report.
He did a good job.
If I do not say this in this prompt.
Use only the document that has been provided.
Then what will he do.
He will go to the internet cell randomly.
He will get some data.
And then as much time actually.
I am taking manually.
It will take more time than that.
Because then I will have to go and verify every number.
Because I do not know if he got it right or not.
So this is just one use case.
But this.
You asked that where else do you use.
So like a lot of people.
Notebook LLM.
Notebook LLM is used.
For company analysis.
So if anyone does not know.
What is Notebook LLM.
Notebook LLM is basically a Google tool.
This is a very good tool.
What can you do in this.
In this you.
Like in chat gpt.
Upload one or two documents.
In this.
You can upload 10-15 documents.
At a time.
And then you can keep asking.
It is a conversational model.
You can continuously ask him questions.
You can ask him to do this.
For example.
There is very little elucidation in Notebook LLM.
Correct.
You have given the data.
It will give the exact output.
Yes.
There is one thing in this.
In this.
Many people do not know this feature.
Most people upload documents in this.
They are asking for their summary.
But in this link feature.
Website and YouTube.
I personally like this a lot.
Like many times.
There is a company.
For example.
When there is a conference call.
It does not come live all of a sudden.
It takes time for their transcript to come.
So the transcript is coming after a week.
Now in the meantime.
YouTube's video has come.
Now the result season is going on.
Every day.
Hundreds of companies are putting their conference calls on YouTube.
Okay.
Now it will take me 1-1.5 hours to watch that video.
Okay.
If I sit like this.
It will take 2 hours.
In Notebook LLM.
I can put its URL and take its summary.
It will generate its transcript.
And then I can keep cross-questioning.
What is the guidance given by the company.
For example.
Let's take an example.
I will show you.
For example.
If you look at Kotak Bank.
The result has come.
But their conference call.
You are getting audio recording here.
Audio recording for the call.
But the conference call transcript has not come yet.
Which I can read.
So what I will do.
I will go to YouTube and search.
Kotak Con Call.
And this is a channel.
For example, I found this video.
Which is 2 days old.
Latest Con Call.
Now either I sit for an hour and listen.
Or I go to Notebook LLM and put this link.
It will take 1 second.
Like here.
I have this doubt.
There were 2-3 things in Kotak Mahindra Bank.
For example.
As the interest rate environment is decreasing.
What is the guidance given by NIM.
I want to know.
What is the growth guidance given by NIM.
So let me ask the growth guidance first.
What is the growth guidance given by NIM.
According to which the company will grow.
At the next time.
So based on the sources provided.
The growth guidance for the company is as follows.
So.
Basically they have not given absolute growth.
1.5 times to 2 times GDP growth.
So let's say if GDP growth is 6%.
Then they are saying 12%.
Now the best thing is.
That was a video.
Now the source of that video.
They have given it to me in the form of a text.
I can read here.
What was the actual conversation.
So the answer given by this.
Here the source of everything is also associated.
So I understood the growth.
For example.
What is the expectation for NIMS.
I have a doubt.
So I can ask this.
And the best part is.
There are citations in this.
Like there was no citation in 4GPT.
And we were relying on.
Showing the last two documents.
In this exactly where is it.
What is it.
Correct.
So if you look at NIMS here.
So these people.
They are saying that this time NIMS.
Has been impacted.
Because the repo rate is being cut.
So there will be more impact.
In fact, if you have seen the reaction after the result.
So this stock was 5% down.
Soon after the result.
Because of what was said in the conference call.
So what did I do in this.
I had already put the link of YouTube.
I had also given an annual report.
A latest investor presentation.
Whatever was the latest data.
Now what I want to do here is.
In the latest.
There is no big red flag.
How is the company growing.
How are the company numbers changing.
I have to analyze all these things.
So there is a best tool in it.
Called Mindmap.
Explore this tool.
So I have made one here.
So it has automatically.
Fi24, Fi25.
For the current quarter.
It has made a mindmap.
What it has done.
It has understood everything.
What are the main things for it.
Asset quality, investment, liquidity.
Related party transaction and so on.
For example.
I click on CASA ratio.
It can give me different answers.
Like.
Average.
Balance.
Whatever was said in the conference call.
In the PPT.
The important thing related to it.
It has made it here.
If you talk about asset quality.
In the quarter 4 Fi25.
It has already made a mindmap.
In absolute terms.
There was a lot of GNPA.
There was a lot of fresh slippage.
So this is just one example.
For this.
It doesn't seem that valuable.
But this mindmap thing.
Many times.
It is so useful.
It gives you the complete structure.
According to your problem.
In this.
There is one thing.
Many times.
We are going for a walk.
And.
We have so many documents.
We don't have time to read them.
So I listened while walking.
So what did you do before.
Whatever sources you have to give.
After that.
You can make a podcast.
So it has made a podcast.
You can listen to this podcast while walking.
Very amazing.
It was a 1 hour conference call.
I made a summary of it.
It made a podcast of 18 minutes.
You walked for 20 minutes.
In that time.
You understood what is happening.
So things are very amazing.
I will just summarize.
First of all.
Both people are suffering.
Who are thinking in binary.
One will write one prompt.
And this magical machine is coming.
That too in almost loss.
Which is completely zero.
This is nonsense.
This is not what I thought.
But the people.
Who are able to properly prompt.
And are able to solve their problems.
If they are not able to solve it.
Ask them how to solve it.
In that too 50-60% will be automated.
Rest 50-60% you will have to do.
But that is also amazing.
You are working on 2x.
Which was already your winner.
In short.
We saw the chat gpt prompt.
How can we use it.
We saw notebook lm.
If hallucination is less.
Give data from that.
Including citations.
For that we can use notebook lm.
Give website, copy paste text.
Make mind map.
Give youtube video.
All sources.
I think 5 million words.
Then there was technical chart.
We can do analysis.
By giving chart image and csv file.
And then dashboard.
If you know coding.
If you have played with google sheets.
You can make it.
Amazing.
Earlier it was a learning curve.
Now it is not.
Like you said.
We need brain for imitating.
We should know.
Which model should I use.
If you don't know that.
You will not get good answers.
Try all.
If you want to do coding.
Then don't use chat gpt.
Or notebook lm.
Go to Claude.
Use Claude's best model.
If your problem is.
I have a lot of data.
I want to converse with it.
Go to notebook lm.
If you have general purpose problem.
Like I want to extract this data.
I want to break problem into multiple.
I want to generate prompt.
The general purpose.
I think.
Chat gpt.
Is jack of all trades.
It will do everything.
With reasonable quality.
It is like that.
But it is not specialized.
In coding and reasoning.
Claude is much better.
But in natural language.
I think chat gpt.
Is much better.
Amazing.
Anyways I hope.
You got the value.
If you have any specific question.
You can comment below.
We will try to answer.
Thank you Shubham.
