AI for Investment Research - NotebookLM

Key Takeaways

• NotebookLM is a powerful AI tool for investment research.
• It searches within uploaded documents, unlike ChatGPT.
• It understands concepts, making searches more effective than Ctrl+F.
• If an SEC filing URL contains ix?doc=/, remove it before uploading to NotebookLM.
• Keep separate notebooks for each company to improve search accuracy.

NotebookLM - Better than Keyword Searches

Investment research is no longer just about scanning documents manually—AI is
changing the game. In this series, we’ll explore how AI tools like Google NotebookLM can help analysts find critical insights faster and more efficiently.

Unlike ChatGPT, which generates responses based on its pre-trained knowledge and web searches (if browsing is enabled), NotebookLM strictly relies on the sources you upload, ensuring that every response comes directly from your selected documents. Even better, it cites the exact source of the information, allowing you to verify it yourself.

How to feed NotebookLM?

There are four ways to add sources: (a) upload a file (PDF, TXT, MP3), (b) provide a website link (such as a 10-K filing), (c) connect Google Drive, and (d) simply copy and paste content, such as an earnings call transcript.

In my experience, it's better to upload files as PDFs or copy and paste the content.

Getting Started with NotebookLM

1. Visit NotebookLM
2. Click “Try NotebookLM” and log in with a Gmail or Google Workspace email.
3. Click “Create new” to start a new notebook.
4. Upload a PDF, Google Doc, website link, or paste text.
5. Start asking questions.

Alternative to Keyword Searches

One of the most frequently used shortcut keys in an analyst’s life is Ctrl + F. The Ctrl key is often the first to wear out on an analyst’s keyboard. In English, the same concept can be expressed using different words. Searching for "divest" may not always be effective if the document uses terms like "sell," "dispose of," or "spin-off." For example, if an announcement states, "In January 2025, the company
entered a definitive agreement to sell the Europe-North segment for $625 million," a keyword search for "divest" would miss it. Searching for "sale" or "sold" may not yield precise results either, as "sale" is a common word found in 10-K filings, often referring to revenue, product sales, or various transactions unrelated to divestitures.

NotebookLM solves this issue by searching for the entire concept, not just exact
keywords.

NotebookLM solves this issue by understanding the concept behind your query. For example, if you ask about divestments, it will search for related terms and phrases, helping you find exactly what you need without manual keyword searches.

Sample task: Finding international revenue for Xponential Fitness

Let's take Xponential Fitness (XPOF) as an example.

Question/task: Find the percentage of revenue generated from international markets in the company's 2023 10-K.

(a) Regular method

If you search for "international" using Ctrl+F in the 10-K, you will get 59 hits, but even after reviewing them, the answer isn't easily identifiable.

(b) Using NotebookLM (Efficient Method)

Upload the 10-K PDF to NotebookLM and ask, “What percentage of revenue comes from international markets?” eliminates the need for manual searching.

NotebookLM extracts the relevant data, shows the computation, and cites the exact source. Instead of manually scrolling through 59 instances of the word "international," you get a precise answer instantly.

See the below interaction:

Best practices

1. SEC weblink error

If you upload an SEC filing link, NotebookLM may not read it correctly if it contains “ix?doc=/” in the URL.

Example (incorrect link format): https://www.sec.gov/ix?doc=/Archives/edgar/data/90168/000009016824000060/sif-20240930.htm

How to fix?: Simply remove ix?doc=/ from the URL.

Corrected link:
 https://www.sec.gov/Archives/edgar/data/90168/000009016824000060/sif-20240930.htm

Why?
SEC filings use Inline XBRL (iXBRL), embedding machine-readable data in HTML. NotebookLM cannot process iXBRL links, so you must use the standard HTML link.

2. Attachments

As I mentioned above, copy-pasting the content or attaching the PDF documents are better than feeding URLs. Many times, URLs don't work.

3. Organizing notebooks for each company

In NotebookLM, each folder is called a “notebook.” When you ask a question, NotebookLM searches only within that notebook. As such, to improve search accuracy, each company should have its own notebook. Rename your notebook based on the company or research topic for better organization.

Limits:

• Free Plan: 100 notebooks, 50 sources per notebook
• Paid Plan: 500 notebooks, 300 sources per notebook

4. Computation/financial model

Based on my experience, it’s best to avoid questions that require computation. Instead of asking, 'What percentage of revenue does the company generate from China?' ask, "How much revenue is derived from China?" While you can ask computational questions, don’t rely on the answers—they may be incorrect. Most importantly, don’t attempt to build a financial model using NotebookLM.

Practical use beyond investing

NotebookLM isn’t just for investment research. You can also upload appliance manuals for devices like cars, HVAC systems, ovens, and lawnmowers, ask questions about error codes—such as “What does E5 on my washing machine mean?”—and quickly search for troubleshooting steps without flipping through pages wasting time—it’s about enhancing efficiency across all aspects of life.