NotebookLM (Gemini Notebook) course: an assistant that doesn’t make things up

The fourth course in MONA's free AI course series, covering the tool that is taking off across Vietnam: NotebookLM from Google, which was renamed in mid-July 2026 to Gemini Notebook (you will see both names online for a while, but it is the same product). More than 30 million people and over 600,000 organizations use it, and the reason it is hot in Vietnam is easy to see: it turns your pile of documents into a Q&A assistant that answers only from those documents, plus the ability to turn documents into a podcast in natural-sounding Vietnamese. This course differs from the previous three in one way: nothing to install, no terminal, just a browser and a Google account. That makes it the course you can hand straight to your staff without anyone complaining it is too hard. Everything about using NotebookLM, from loading sources to creating a Vietnamese podcast, fits in the seven parts below.

Part 1: What is NotebookLM, and three criteria that tell you when it beats ChatGPT

NotebookLM is a Google AI tool with a very different philosophy from ChatGPT or the chat version of Claude: instead of answering from the model's vast general knowledge, it answers only from the sources you load (documents, web pages, videos). People in the field call this grounding: every answer must be tied to a source, with clickable citations so you can verify it at the origin. If it is not in the sources, it says plainly “the sources do not mention this” instead of making something up to round out the story.

From that philosophy come three criteria for choosing tasks, worth memorizing so you never wonder “should this go to NotebookLM or ChatGPT”:

1. You cannot accept AI making things up. Contracts, tax regulations, health records, technical documentation: one wrong word causes real trouble. This is NotebookLM's home turf.
2. Your information is scattered across many places and formats. A dozen PDFs, a heap of Google Docs, a few YouTube videos, meeting minutes: NotebookLM swallows it all into one place and connects the pieces.
3. You need to turn that scattered pile into one clean output. A summary, a Q&A document, a news briefing you can listen to: it produces these fast and sticks to the sources.

Conversely, work that needs creative ideas (flowery ad copy, brainstorming new ideas) is not its territory: precisely because it is tied to your sources to prevent fabrication, it is less imaginative. Skilled users combine both: NotebookLM handles the “correct” part, Gemini or Claude handles the “good” part (Part 6 covers this rhythm in detail).

Part 2: Your first notebook: loading sources the right way

The working unit of this tool is the notebook: a “folder” for one topic, holding that topic's sources. The first and most important principle: one separate notebook per topic, do not mix tax documents with recruiting documents in one place, because every answer draws from all the sources switched on in that notebook.

The sources you can load are richer than many people think:

• Document files: PDF, Google Docs, Google Slides, text files. A quick tip for finding sources, straight from the original course: to find the manual for a device, search “device name + user manual + PDF” and you will usually find the official version.
• Trang web: paste a link and it reads the whole page. If you hit a page that blocks reading (it happens), here is a workaround: open the page, copy all the text, and paste it as a “pasted text” source. A minute slower, identical result.
• Video YouTube: paste a video link and it reads the transcript. Loading an expert's video on exactly your notebook's topic is the cheapest way to “invite an advisor”.
• Sound: audio files of meetings and lectures.

The current limits, so you can plan accordingly: the free version allows up to 50 sources per notebook, each source holding up to half a million words, which means one notebook can take the whole filing cabinet of a department; the paid version (included in Google AI Pro/Ultra plans) raises the source cap and adds more tools. And a tip for getting past the cap, from the original course: if you hit the source limit, merge several documents of the same type into one file before loading, since the cap counts sources, not thickness.

If a document exists only as a scanned PDF and you want to edit its content before loading it, PDF to Word converter free on mona.media, the tool has built-in text recognition for scans.

Here is a start-to-finish example of loading sources, so you can picture the rhythm: building an “equipment manual cabinet” for a coffee shop chain. Open a new notebook and give it a proper name from the start (“Equipment manuals: bar area”), because two months from now you will have a dozen notebooks and the name is what saves you. First round of loading: PDF manuals for the espresso machine, grinder, commercial fridge and POS terminal (found with the “machine name + user manual + PDF” tip). Second round: paste links to 2-3 YouTube videos on cleaning that exact type of machine. Third round: the shop's own internal procedure file (a text file or Google Docs). Rename each source to keep things tidy (“Espresso machine: official manual”, “Video: cleaning the group head”). Fifteen minutes later, a question like “what does error E3 on the espresso machine mean” has an answer with page citations, instead of a roundabout phone call.

One source-loading habit worth building from day one, stressed because it decides half the quality: garbage in, garbage out. This tool amplifies the quality of your sources: load official documents and solid reports and the answers are trustworthy; load clickbait blog posts and it will summarize that garbage back to you very confidently. Choosing sources is the step where people must be good, and the machine cannot carry it for you.

Part 3: Source-grounded Q&A: citations, choosing sources, and the “forgot to save” trap

Once the sources are loaded comes the best part: ask in plain language and get an answer with a small numbered citation after each point. Click a citation and it jumps straight to the original passage in the source (the exact PDF page, the exact stretch of video transcript) so you can check for yourself. This is the real difference from ordinary AI chat: you do not have to “trust”, you can verify.

Four techniques to use it well, drawn from the original course and our own experience:

1. Choose sources by question. Every answer draws from the sources currently ticked in the left column. To ask about one document only, untick the other sources. A valuable example from the original course: a recruiting notebook holds many candidates' files, and during an interview you tick only that person's file, so the tool does not mix one candidate's details into another's.
2. When you see a good answer, save it as a note (save to note) RIGHT AWAY. The trap most people fall into: close the notebook and the conversation disappears, because Google deliberately does not store or train on your data (a security plus, in exchange for you saving what you want to keep). Saved notes have another nice trick: combine several notes and convert them into a new source, crystallizing your findings into an official document of the notebook.
3. Use the suggestions area if you are new. Every notebook comes with a guide section: an automatic summary of each source, ready-made templates (Q&A document, executive summary) and suggested questions. If you are new to the tool, just click the suggested questions, they are surprisingly on target.
4. Ask a question that is “not in the sources” to see how honest it is. Try asking about something the sources surely do not mention: it will say plainly that the sources do not cover it (and politely point to a few nearby topics). This reflex of “daring to say I don't know” is why people who work with legal, medical and technical documents trust it.

Here is what a real Q&A round looks like with a “project brain” notebook: you ask, “based on this week's meeting minutes, which items does the supplier still owe us and what are the deadlines?” The tool returns a list of four items, each line with a numbered citation. Click the citation on the line “mold delivery 5 days late” and it jumps to the exact stretch of the meeting transcript where both sides agreed on that. You save the answer as a note named “Supplier outstanding items, week 30”, and next Friday you simply ask, “compared with the week 30 note, which items has the supplier delivered this week”. Keep going like this and the notebook becomes a project logbook you can talk to, and every claim in it can be traced back to the original minutes.

Part 4: Audio Overview: turning documents into a Vietnamese podcast

This is the feature that sent NotebookLM viral worldwide, and good news for us: it now supports Vietnamese, among more than 80 languages, with voices so natural that many people cannot tell it is a machine. One click turns a dry pile of documents into a two-voice podcast episode, a back-and-forth conversation about exactly that content: introductions, questions and answers, analogies, sounding like two professional hosts.

Creating one takes a minute to learn: in the notebook, the right-hand area has an Audio Overview section; click create and the tool builds it automatically (allow a few minutes for thick sources). To get the flavor you want, open the customize panel BEFORE creating and give instructions in plain language. The episode plays right in the notebook, and you can download the file to share with a group. Three customization instructions that give the best results in our experience: state who the LISTENER is (“new staff who know nothing about the industry”), state the FOCUS (“only dig into the returns and exchanges process”), and state the desired LENGTH (“keep it to 10 minutes”). Without those three, the tool decides for itself, and the default is often more sprawling than you want.

The real value for busy people in Vietnam: anything you “must read but never get around to reading” can be turned into something to listen to. An 80-page industry report on the drive. A new procedures document at the gym. And do not stop at the default button: there is a customize option before creating, where you can say things like: “focus on the impact for retail businesses, assume the listener is not technical, go into detail on the numbers.” The episode comes out exactly the way your target listener likes it.

Its sibling is Video Overview (also available in Vietnamese): it turns sources into a presentation video with illustrations, good for training material to hand out to your team. This pair is the application we find most valuable for Vietnamese businesses: turning internal documents into Vietnamese audio-visual content for staff, without spending a cent on outside production. Part 5 goes deeper.

Part 5: Five sample notebooks for Vietnamese businesses: copy and run

The most valuable part of the course, combining use cases from the original course with scenarios we find fit Vietnamese businesses. Each notebook below is a “machine” you can build in one session:

1. Equipment and software manual cabinet. Load all the manuals for the equipment, machines and software your company uses (find official PDFs with the Part 2 tip). From then on the whole company asks in one place: “what does the blinking red light on the barcode printer mean”, “how do I set permissions in the accounting software”. No more asking around or calling the supplier and waiting half a day every time something breaks.

2. Tax and paperwork notebook. Load the tax regulations relevant to your industry and your accountant's reports from previous periods. When filing time comes, ask: “what are our tax obligations this quarter”, “can expense X be counted as a deductible expense, and which source says so”. The answer comes with citations pointing to the exact clause, something you would never dare trust from ordinary AI chat. The accountant still makes the final call, but the lookup step drops from an hour to ten minutes.

3. Recruiting notebook. Load the role's competency framework, the company's interview question set, candidate files and notes from earlier rounds. Before each interview: “what are this candidate's strengths and areas to probe further based on their file”, “give 10 interview questions based on the shift supervisor competency framework”. Remember the source-ticking rule: switch on only the file of the person you are about to interview.

4. Project brain. One notebook per major project: the plan, meeting minutes (export the transcript from Google Meet/Zoom and load it directly), documents from similar completed projects. Then ask: “my open items after yesterday's meeting”, “draft an email summarizing this meeting”, “what lessons from the last project apply to this one”. Project management is almost tailor-made for this tool: the essence of the job is gathering scattered information and synthesizing it into something easy to digest.

5. Market watch notebook. Load industry reports, solid analysis articles, and competitors' pricing pages and public documents (combined with the price-page screenshots you already know how to collect automatically from the advanced course). Ask: “how do competitors' pricing strategies differ”, “what are the industry trends for the second half of the year, and which source says so”. Then create a custom Audio Overview, “for a business owner to listen to in the car, 15 minutes, focused on action implications”: you get your own market intelligence briefing in Vietnamese, one episode a week.

A general operating tip for all five notebooks: for sources that are Google Docs/Slides there is a re-sync button after the original file changes, one click and the notebook updates; and from the paid version on, Drive sources can even sync automatically. That means you feed the notebook once and it lives alongside your company's documents, with no need to reload from scratch.

Part 6: What's new in 2026, real limits, and how to pair it with other AIs

The July 2026 rename is more than a new coat of paint. A few new things worth knowing: every notebook now has a secure cloud computer that can run code, which means it does not stop at answering: it can export finished files: Excel spreadsheets, PDFs, PowerPoint slides built from the content of your sources (rolling out gradually to Pro users). The notebook has also been connected to the Gemini app (create and use notebooks right in the app, with two-way sync) and is about to arrive in Google Search's AI Mode as well. The direction is clear: from a standalone tool to a layer of “sourced knowledge” running across the Google ecosystem.

The real limits, stated fully so you do not expect the wrong things:

• Less imaginative. Its output is correct but often dry. The standard rhythm of skilled users: NotebookLM handles understanding and sourced synthesis, then carry that synthesis over to Gemini or Claude for writing it well (ads, scripts, proposals). This matches the division-of-roles principle taught throughout the series: each tool does what it does best.
• It cannot replace the person who signs off. It cites sources accurately, but the sources may be outdated, incomplete, or the policy may have changed. For tax, legal and medical processes: the machine looks things up, the person signs off, no reversing that.
• Sensitive data needs to be weighed against your company's standards. Google commits not to train on user-uploaded data, and the organization edition comes with Google Workspace's enterprise commitments. Even so, our principle stays the same: top-secret documents do not go onto any cloud service without approval from the person in charge of security.

Part 7: Rolling it out in your company on a 2-week plan, and MONA's take

NotebookLM needs nothing installed, so the rollout plan is much leaner than in the earlier courses. Two weeks is enough for a company to get started:

Week 1: build 2 notebooks, assign 2 owners. Pick the two most painful problems from the five samples in Part 5 (in our experience, the equipment manual cabinet and the project brain deliver value fastest). Give each notebook a “librarian”: gathers the right sources, names them properly, and prepares 5-7 sample questions pasted into a note for colleagues to copy.

A set of 5 sample questions for the “librarian” to paste at the top of the notebook for colleagues to copy (using the equipment notebook as the example, swap in the contents of your own): “1. Machine X shows error [error code], how do I handle it? 2. What steps are in the end-of-day cleaning procedure for area Y, and which source sets them? 3. Until when is device Z under warranty, and what voids the warranty? 4. Compare the manufacturer's instructions with our internal procedure: is there anywhere we do things differently? 5. Write a one-page summary of [topic] for newcomers.” Question 4 is the most valuable: it makes the tool cross-check sources, exactly what it does best, and it often turns up mismatches nobody in the company noticed.

Week 2: roll out to the team + one training audio. Open access for the team to try it, with exactly one usage rule: when you see a good answer, save it as a note, and cite the source when you take it into a report. Then serve the showpiece to the team: take the company's latest procedure document, create a custom Vietnamese Audio Overview “for new staff, friendly tone, 10 minutes”, and share it in the group chat. A document nobody read suddenly becomes a podcast episode everyone listens to on the commute: that is usually the moment the whole company says “aha” and goes off to build notebooks for their own departments.

And the familiar boundary of the series, applied here: NotebookLM is a knowledge cabinet for Q&A, not an operating system. When your needs go beyond that to “the machine works on that data by itself” (answers customers by itself, processes orders by itself, runs processes by itself), that is the territory of custom-built automation software with permissions and someone accountable, in line with MONA's automation articles. If you want someone to sit down and map out the route from a knowledge cabinet to a self-running system for your business, call 1900 636 648 or send it to info@themona.global, and the MONA team will advise you straight, as always.

Quick answers

Does learning to use NotebookLM cost money? Yes and no. The free version is enough for individuals and small teams: 50 sources per notebook, the full Q&A toolkit, Audio Overview. The paid version (within the Google AI Pro/Ultra plans) raises the source cap and adds file export and automatic Drive sync.

Are NotebookLM and Gemini Notebook two different products? One product, renamed in mid-July 2026. The address is unchanged, and your data and existing notebooks stay as they were.

Does it work well in Vietnamese? Yes. Q&A in Vietnamese is smooth, and Audio Overview and Video Overview speak natural-sounding Vietnamese (among more than 80 supported languages). This is why it took off in Vietnam.

How is it different from ChatGPT or Claude? A different philosophy: NotebookLM answers only from the sources you load, with verifiable citations, so it almost never makes things up but is less imaginative. Regular AI chat answers from general knowledge, is more creative, but you have to check it yourself. Combining both is the smartest approach.

Is it safe to upload company data? Google commits not to use uploaded data to train models, and conversations are not stored unless you save a note. Ordinary internal documents are fine; for top-secret documents, follow your company's own security standards, with a person in charge approving.

The next course in the series: Vibe coding with Gemini and Antigravity, the missing piece of the Google toolkit trio in your hands. See you there.