What is this article talking about?
This article compiles the complete content of the "NotebookLM Application Encyclopedia" free lecture (100 people). The value of NotebookLM lies in turning data into a knowledge space that can be questioned, compared, and organized. This article teaches you step by step how to use it from a data storage tool to a knowledge analysis assistant that can give you action suggestions. Who is
· People who often have to read a large number of reports, papers, and teaching videos, and are unable to finish them and are afraid of missing the key points
· Those who have already put the data into the NotebookLM summary and want to take the next step
· Learners who want to organize multiple teachers and sources into their own AI consultants
· The four-stage evolutionary route of reading reports, and the specific prompt words for "let the report read me"
· Real cases of multi-source knowledge base and cross-field collaborative curriculum
· Two well-trodden data organization skills: PDF traps and Markdown content boundaries
First positioning: two directions of AI empowerment
AI applications now have two clear general directions. A study released by OpenAI in September 2025 tracked the behavior of 700 million users around the world and found that most of the current usage is concentrated in the first type, and a small part is in the second type.
Let AI do its own operations: write articles, translate, make presentations, and generate pictures. This is where most people's current usage is.
Verify ideas, assist decision-making, and find blind spots. Few people use it, but this is where knowledge workers widen the gap.
NotebookLM takes the second path. Rather than waiting for AI to become more automated, it is a more practical route to organize your own knowledge first and let AI have the ability to be your consultant.
The four-stage evolution of reading reports
In the era without AI, there was too much information and it was often impossible to read it all.
What most people do now. Frequently asked questions also follow: If we use AI summarization, have we really learned anything?
Please ask AI to summarize all the ten reports first, quickly determine which one needs to be read first, and then read that one intensively. AI is responsible for filtering, but you still have to do the intensive reading yourself.
Throw the report to AI and ask it to give me suggestions and action plans suitable for me based on my situation.
You can turn on the memory mode, but the dialogue memory will be mixed with errors and outdated information. A cleaner approach is to prepare a personal profile.
Basic information, core identity, positioning, career path, a quick introduction specifically for AI consultants.
The prompt word for the fourth stage is very simple:
From "I will read the report" to "The report will read me", the value of the same information is completely different.
From one report to an entire batch: a multi-source knowledge base
One report can give suggestions, but what about ten reports? What about a certain teacher’s twenty or thirty video series? The capacity of NotebookLM can support this kind of gameplay: according to the plan in early 2026, a notebook can hold 50 sources in the free version, and 300 in the paid version (the actual quota is subject to the official announcement). As soon as the magnitude is increased, it changes from "use a report to read my resume" to "use the knowledge base of 20 teachers to read my two or three years of work diaries."
Cross-disciplinary cooperation is another practical scenario. When I was a photography teacher, I held a parent-child photography class with a teacher in the physical and mental field. I put my course materials over the years in a knowledge base, and he put his. When we wanted to collaborate on a syllabus, I opened a new notebook, imported the sources from both sides, and told AI, "Based on these two syllabuses, help me design eight classes of one and a half hours each." A version of the collaborative syllabus that can be actually used came out.
There is a well-learned detail here: the knowledge base must be maintained separately, and there is no need to merge the entire batch. In the early days, we poured the two people's data into the same database. Later, we found that this database could only serve this collaboration. If he wanted to do other projects, my photography data would be mixed in. The same goes for me if I want to collaborate with others. Maintaining them separately, authorizing sharing when necessary, and importing them once is a clean approach.
Two surprising data organization skills
Tip 1: Use cloud files as a central knowledge base, be careful with PDFs
The source format is recommended to use Google documents or plain text. For most people in the Google ecosystem, the connection between files and NotebookLM is smoothest; if you have your own knowledge base tool (like Obsidian), the same applies, and the concepts are the same.
Be careful with PDFs. PDF looks beautiful because it hides a lot of invisible typesetting codes in addition to text. When AI reads it, it will be disturbed by these codes and colorful diagrams. It’s okay to lose three copies, but it’s easy to exceed the AI’s processing limit by losing twenty complex PDFs. For content that requires precise numbers, it is recommended to convert it to plain text, Markdown, or paste it directly into a Google document, which is more stable than losing PDF.
Tip 2: Use Markdown to draw content boundaries
Markdown is a very simple grammar for writing articles. Use hash marks to distinguish main titles, second-level titles, and third-level titles. Its value is to help draw the boundaries of the content: let the AI clearly know where this paragraph ends, where the next paragraph begins, and which three key points belong to the same level.
Take my meeting record prompts as an example. After the three-level organization structure is clearly written in Markdown, AI can accurately know where the first layer is executed and where the second layer starts. In an era when instructions are written more and more complexly, without content boundaries, the results are really very different.
How you can practice: Build a small source package
First select three documents with the same theme, create a NotebookLM notebook, and ask it in this order:
- Ask the abstract first: What is the core idea of this batch of information.
- Ask again about comparison: In what ways do these sources contradict each other?
- Then look for blind spots: My current understanding is this (write it down), please go back to the source to look for supporting evidence, counterexamples and gaps.
- Attach your personal profile at the end and ask for action suggestions: Based on my situation, what this batch of information suggests I should do.
After completing these four steps, you will realize the difference from a data organization tool to a knowledge analysis assistant. When publishing the collation results to the outside world, remember to clearly mark which ones come from the source, which ones are your compilations, and which ones are AI's inferences, so that the answers can be traced.