NotebookLM 完整使用教學:把一疊資料丟進去,它幫你做出懂你資料的 AI 研究助理

At the end of the semester, a student preparing for the national exam had over a dozen PDFs spread out on her desk - laws, cases, and lecture notes that added up to thousands of pages. What she was most anxious about wasn't that she didn't understand, but that she "couldn't finish reading" them, and every time she wanted to go back and find a certain concept in a particular file, she had to flip through for a long time. I asked her to upload all the files to NotebookLM and ask it questions directly. Ten minutes later, she looked up and said, "It can actually tell me which page the answer is on."

This is exactly what sets NotebookLM apart from general AI chatbots: it doesn't use the entire internet's knowledge to answer you, but only responds based on "the data you give it," along with the page number of the source. It won't make things up or go off-topic because it only reads your materials. For people who need to digest a large number of documents, this design is tailor-made. This article will guide you through getting familiar with it.

What is NotebookLM

NotebookLM is an AI research and note-taking tool launched by Google. The core concept is simple: you upload your data (PDFs, Google Docs, web pages, videos, or text) as "sources," and then all of NotebookLM's answers are based only on these sources, with references to the specific file and section.

This approach differs from that of general assistants like ChatGPT or Gemini. General assistants can chat about anything, but they might mix internet information with your data, making it hard to distinguish what's from your files and what's fabricated. NotebookLM limits its scope to your sources, solving the problem of "I only want to know what my data says." For how to ask questions effectively, you can first look at the six most common pitfalls for AI beginners.

What can it be used for

• Reading and exam preparation: Upload all your lecture notes and past exams, and ask it questions about concepts, have it organize key points, or create a quick Q&A for before the exam.
• Research paper digestion: Throw in over a dozen papers and ask, "What are the stances of these papers on a certain controversy?" It will help you compare them horizontally and mark the sources.
• Project file organization: Upload requirement documents, meeting records, and specifications, and when you need to find a resolution or trace the origin of something, just ask it.
• Turning data into audio guides: This is one of its most popular functions, which will be discussed in detail later.

How to use it: from creating your first notebook

1. Log in and create a notebook: Use your Google account to go to notebooklm.google.com, and click "New Notebook." A notebook is like a theme; it's recommended to create one for each subject or project, rather than mixing everything together.
2. Upload sources: Add PDFs, Google Docs, URLs, YouTube video links, or copied text. A notebook can hold multiple sources, which is one of its strengths - it will treat these data as a whole to understand them.
3. Start asking questions: Use natural language in the dialogue box, such as "Help me organize the core arguments of the third chapter of this lecture note." After the answer comes out, each key point will have a reference mark next to it, which you can click to jump to the original text, making it very convenient to verify.
4. Generate summaries and key points: You can ask it to "create a study guide" or "list all key terms and their explanations," and it will generate them based on your sources, without including external information.
5. Save as notes: Useful answers can be pinned as notes in the notebook, gradually accumulating into your own knowledge base.

Advanced tips

Audio Overview (audio guide) - turning data into podcasts. This is one of NotebookLM's most popular functions. It will generate an audio program based on your sources, with two AI hosts discussing the content, similar to listening to a podcast. You can listen to it while commuting or exercising, which is equivalent to turning reading data into listening to data. According to Google's official information, Audio Overview supports over 80 languages, and non-English versions are "full-length" and have the same depth as the English version, providing a much better Chinese experience. You can also customize the tone and length.

Video Overview. Taking it a step further, it can generate video guides with images, allowing you to choose the format, language, and visual style (such as whiteboard or watercolor). According to official information, Video Overview has also been expanded to about 80 languages. If you want to quickly turn dry data into easy-to-understand audio or video, this is very useful.

Mind Map. It can automatically generate an interactive mind map based on your sources, drawing out the high-level concept structure so you can understand the overall context and relationships at a glance. Clicking on a node can also expand it for deeper exploration, which is especially useful for reviewing the overall structure.

Using it with other tools. NotebookLM is good at "focusing on your data," but it's not suitable for real-time internet searches. For looking up the latest news, you can combine it with AI search tools like Perplexity that provide sources. One tool handles your data, and the other handles the outside world, making for a smooth division of labor.

Precautions

It is only loyal to your sources, so the sources must be good. NotebookLM only reads your data and doesn't fabricate information, but conversely, if the data you upload is incorrect or outdated, it will also follow the mistake. It won't help you judge whether the source is correct or not; that responsibility still lies with you.

Sensitive and confidential data must be evaluated. Before uploading company secrets or others' personal information, you must ensure you have the right to do so and understand how the data will be handled, and you cannot neglect compliance.

It is not used for looking up the latest news. Because it only reads your sources, it won't know anything you haven't uploaded. If you want to ask about "what happened this week," please use another tool.

Evaluation from TheAI Academy

I use NotebookLM for research and teaching, and it's one of the few tools that really save me time. Its most intelligent decision is to limit its answer scope to your data - by doing so, it immediately solves more than half of the old problem of AI making things up, and you get an assistant that is very familiar with your data.

What's impressive about NotebookLM is not that it knows a lot, but that it only says what's in your data - in an era where AI often talks nonsense, this "having limits" is the most precious.

Data sources

• Google blog: NotebookLM Audio/Video Overviews and multi-language updates
• Google blog: NotebookLM Video Overviews and Studio upgrades

The above is compiled based on publicly available information, with functions accurate as per Google's official announcements.