How to Set Up NotebookLM The Right Way: 5 Features + Briefing Templates

TL;DR

ChatGPT will fabricate citations 18% of the time. NotebookLM fabricates 0% — because it physically cannot answer from anything except the PDFs, slides, and recordings you give it. Every response is line-cited.

It's free. It's built on Gemini Pro. 97% of students have never opened it.

Five features turn NotebookLM from "another Google chatbot" into a postgrad-grade research partner:

1. Sources — drop 50 PDFs, lecture recordings, slides. Every answer cites the exact page or timestamp.
2. Audio Overviews — turn 25 papers into a 12-30 minute podcast you listen to on the train. Two AI hosts, real disagreement, actually sticks.
3. Mind Map — clickable visual of every concept across all sources. Gaps in the graph = thesis opportunities.
4. Studio — one-click briefing doc, FAQ, study guide, timeline. Cited line-by-line. Stop writing lit-review summaries by hand.
5. Custom Personas — tell NotebookLM to answer as a skeptical PhD examiner, devil's advocate reviewer, or your supervisor. Pre-defend your thesis before your supervisor sees it.

Setup time: 20 minutes. Replaces a R350/mo ChatGPT Plus subscription that lies about citations.

This guide is the merged successor to the older "Free AI Study Workflow" guide — every workflow from that one is folded into the 5 features below.

---

Who this is for

• Postgrads writing a thesis or dissertation
• Honours / final-year students with multi-paper assignments
• Any student paying for ChatGPT Plus for academic work — you're paying for the wrong thing
• Researchers synthesising 20+ sources

If you've never opened NotebookLM, this is the entire setup, end to end. If you have, jump to Feature 4 (Studio) and Feature 5 (Custom Personas) — those are where most users still lose 5+ hours a week.

---

Why NotebookLM beats ChatGPT for academic work

Job	ChatGPT (Plus)	NotebookLM	Claude
Source-grounded Q&A on YOUR materials	Hallucinates ~18% of citations	0% hallucinated citations	Strong, but no built-in source manager
Lecture recording → notes with timestamps	Manual transcribe first	Native	Manual transcribe first
Briefing doc / FAQ / study guide auto-gen	No	Yes (Studio)	No
Mind map across all sources	No	Yes	No
General writing, coding, real-time search	Yes	No	Yes
Multi-modal (screenshots, image analysis)	Yes	Limited	Yes
Cost	R350/mo	Free	$20/mo

Rule: NotebookLM for source-grounded research and study. ChatGPT or Claude for writing, coding, ideation. Use both. NotebookLM costs nothing.

About hallucinated citations: A 2024 Vectara study put GPT-4 hallucination rate at ~3% on summarisation, but on academic-citation generation specifically (where the model invents authors, years, page numbers), peer-reviewed studies have logged rates of 18-47%. NotebookLM cannot do this — it has no access to anything outside your sources.

---

Prerequisites

• A Google account. Personal Gmail is fine. .ac.za works too.
• Your course materials, papers, slides, or recordings as PDF / .m4a / .mp3 / .docx / Google Docs links.

That's it. No card. No trial. No .edu required.

---

Feature 1 — Sources: drop 50 PDFs, get 0% hallucinated answers

What it is: A notebook is a sandboxed source pool. NotebookLM physically cannot answer from anything outside the sources you upload. This is the entire point.

Setup:

1. Go to notebooklm.google.com and sign in.
2. Click New notebook. Name it after the project, course, or thesis chapter — e.g. THESIS — Ch 3 Methods, CSC2002S Data Structures, or LITREV — moral injury.
3. Click Add sources. Drag in:
   • PDFs of papers, textbooks, slides
   • Lecture recordings (.m4a / .mp3) — yes, audio sources work
   • Google Docs / Slides links
   • Public webpage URLs
   • Pasted text
4. Wait 30-60s while NotebookLM reads everything.
5. Ask in the chat: "List every source. For each, give a 2-line summary and the main argument."

[SCREENSHOT: NotebookLM new notebook with 12 PDF sources loaded]

The two questions that demonstrate the source grounding:

• "Where do these sources disagree with each other? Quote the conflicting lines with citations."
• "What is the strongest argument FOR [your thesis position] in these sources? What is the strongest argument AGAINST?"

Every answer comes back with footnoted citations like [1, p. 47]. Click the footnote — it jumps to the exact passage in the source.

Tip — one notebook per topic. Don't merge a methods notebook with a literature-review notebook. The model's answer quality scales inversely with source breadth. 50 narrow sources beat 200 broad ones.

Source caps: 50 sources per notebook on free tier. If a textbook is 800 pages, split it by chapter — each chapter is a separate source. You'll get sharper page citations.

Lecture-recording workflow (under-used):

1. Record the lecture on your phone (Voice Memos / Recorder).
2. Export as .m4a or .mp3.
3. Drop it into the course notebook as a source.
4. Ask: "At what point did the lecturer explain [concept]? Quote what they said and give the timestamp."

NotebookLM gives you the timestamp. You can scrub straight to that second in the recording. Most students re-listen to entire lectures — you just listen to the 90 seconds that mattered.

---

Feature 2 — Audio Overviews: turn 25 papers into a 30-min podcast

What it is: NotebookLM generates a podcast episode — two AI hosts talking conversationally about your sources. They disagree. They make callouts. They reference each other. It does NOT sound like a robot reading a summary.

Why it matters: Reading 25 papers takes 40 hours. Listening to a 30-minute audio overview on the train, at the gym, while cooking — you absorb 70% of the same content with 0 minutes of desk time.

Setup:

1. In your notebook, open the Studio panel (right side).
2. Click Audio Overview → Generate.
3. Wait 2-5 minutes (longer if you have 30+ sources — that's normal).
4. Download the .wav, drop it on your phone.
5. Listen on commute. Take voice notes when something clicks.

[SCREENSHOT: Studio panel with Audio Overview generated]

Customise the audio with prompts (this is the part most people miss):

Click Customise before generating. Give NotebookLM specific instructions on what to focus on. Three prompts that have lifted students' grades:

Prompt A — exam-targeted overview

"Generate a 20-minute audio overview targeted at a postgrad student preparing for their oral exam. Spend more time on the methodological disagreements between sources than on the introductions. End with the 5 questions an external examiner is most likely to ask, with the strongest source-cited answer to each."

Use 7-10 days before an exam. Listen daily.

Prompt B — thesis-defence rehearsal

"You are two PhD examiners reviewing my thesis chapter. One is sympathetic, one is hostile. Spend 25 minutes attacking the weakest claims in these sources from both sides. End with the 3 questions a hostile examiner would ask, and what the sources actually say in response."

Use the week before a viva or supervisor meeting. Devastating in a good way.

Prompt C — synthesis for a literature review

"Generate a 15-minute audio overview structured as: (1) what these sources agree on, (2) where they disagree and why, (3) the gap none of them address. Cite specific authors and years. Speak as if drafting the literature-review section of a paper."

Use when you've gathered 15-25 sources for an essay and you need to figure out the angle. Replaces ~6 hours of close-reading.

Tip — re-generate, don't tweak prompts mid-run. If the first overview misses your angle, click "Generate new" with a sharper prompt. Don't try to edit the audio.

Watch out: Audio Overviews are English-only on the student-facing tier as of mid-2026. Other languages rolling out.

---

Feature 3 — Mind Map: find the gap your thesis can fill

What it is: NotebookLM auto-builds a clickable concept graph across all your sources. Every node is a concept. Every edge is "these sources both discuss this." Click a node → see exactly which sources cover it and what they say.

Why it matters: When you're drafting a thesis or dissertation, the most valuable thing is finding what nobody else has written about. A mind map shows you concept density at a glance. The thin edges and missing nodes are your contribution.

Setup:

1. Open your notebook.
2. Click Mind Map in the Studio panel (or in the chat output area — placement moves around).
3. Wait 30s.
4. Click any node — NotebookLM shows the connected sources and quotes.
5. Look for thin clusters (concepts touched by only 1-2 sources) and missing bridges (concepts that should connect but don't in the literature).

[SCREENSHOT: NotebookLM Mind Map view of a 20-source notebook]

Three questions to ask the mind map:

1. "Which concepts in this mind map are discussed by only 1-2 sources? List them with the sources."
2. "Which two concepts SHOULD logically connect based on the literature, but no source explicitly bridges them?"
3. "If I were writing a novel contribution, which gap in this mind map would be the most defensible thesis?"

The third question alone has saved postgrads weeks of supervisor meetings.

Tip — mind maps work best at 15+ sources. With fewer than 10, the graph is sparse and uninformative. Bulk up your source pool first.

---

Feature 4 — Studio: one-click briefing doc, FAQ, study guide, timeline

What it is: Studio is the most underused panel in NotebookLM. With one click it generates:

• Briefing doc — 1-page executive summary of all your sources, cited
• FAQ — the 10-15 most likely questions a reader/examiner would ask, with cited answers
• Study guide — exam-style practice questions + answer key, cited to source pages
• Timeline — chronological view of events/findings across all sources (great for history, lit-review, longitudinal studies)

Each artefact is generated in 60 seconds. Each is fully cited. You can export as Google Doc.

Setup:

1. Open Studio (right panel).
2. Click any of: Briefing Doc, FAQ, Study Guide, Timeline.
3. Wait 30-90s.
4. Click Save to note to keep, or Open in Google Docs to edit.

The Studio Briefing Template (copy-paste)

The default Studio briefing is generic. The version below — paste it into Studio's Customise field before generating — produces a postgrad-grade brief, not a Wikipedia summary.

You are an executive briefer preparing me for a defence of my position on:
[YOUR TOPIC / THESIS STATEMENT IN ONE SENTENCE]

Generate a 2-page briefing document with these exact sections, in this order:

1. THE ONE-PARAGRAPH ANSWER
   The single most defensible position the sources support, with the 2-3
   strongest cited claims that back it.

2. THE 5 STRONGEST POINTS FOR
   Each: claim → cited evidence → which source says it.
   Quote the sources directly where the wording matters.

3. THE 5 STRONGEST POINTS AGAINST
   Each: claim → cited evidence → which source says it.
   Be brutal. Don't soften the counter-position.

4. THE 3 METHODOLOGICAL WEAKNESSES IN THE LITERATURE
   What the sources don't measure / don't control for / don't address.
   This is where my contribution lives.

5. THE 5 QUESTIONS A SKEPTICAL EXAMINER WILL ASK
   Each with the source-cited answer I should give.

6. THE GAP
   What no source in this notebook addresses, that I should claim as
   the contribution of my work.

Cite every claim. Quote where the wording is load-bearing. Sentence case
headings. No filler. No "in conclusion."

Drop that into Studio's Customise field. Click generate. 90 seconds later you have something a supervisor will actually read.

Use case — supervisor meetings: Generate a fresh briefing the morning of every supervisor meeting. Bring the printed PDF. Supervisors notice. They notice harder when the citations check out.

---

Feature 5 — Custom Personas: pre-defend your thesis before your supervisor sees it

What it is: NotebookLM lets you instruct it to answer in a specific voice / role / persona. Most users never use it. This is the single highest-leverage feature for postgrads.

You can tell NotebookLM:

• "Answer as a skeptical PhD examiner with a focus on methodological rigour."
• "Answer as a journal reviewer recommending major revisions."
• "Answer as my supervisor — note: my supervisor cares most about [X] and tends to push back on [Y]."

Then you ask: "Read my draft thesis chapter [pasted as a source]. What would you say in viva?"

It will tear your work apart. Politely. With citations. Before the actual examiner does.

Setup:

1. In your notebook, click the chat input → look for Custom prompt / Persona option (top-right of chat in current UI).
2. Paste one of the persona templates below.
3. Ask your question normally — the persona stays applied for the session.

[SCREENSHOT: NotebookLM persona prompt setup]

Persona 1 — Skeptical PhD examiner

You are a skeptical PhD external examiner with 25 years of experience in
[FIELD]. You have read every source in this notebook. You are reviewing
my draft thesis chapter (uploaded as a source). Your goals:

1. Find the weakest unsupported claim and challenge it with a counter-source
   from this notebook.
2. Identify any methodological assumption I've made implicitly without
   defending it. Demand I defend it.
3. Find any source I've cited where I'm misrepresenting what the source
   actually says. Quote the source verbatim and contrast with my paraphrase.
4. Ask the 3 questions a hostile external examiner would ask in viva.

Tone: rigorous, polite, unflinching. No softening. Cite line-by-line.
Sentence case. No filler.

Persona 2 — Devil's advocate reviewer

You are a contrarian journal reviewer who believes the dominant position
in this literature is wrong. Read all sources in this notebook with the
goal of building the strongest possible case AGAINST the consensus
position. For each source:

1. What does it claim?
2. What is the weakest assumption underlying that claim?
3. Which other source in this notebook, if any, contradicts it?
4. If no source contradicts it, what kind of evidence WOULD contradict it?

End with: the 5-paragraph minority-position case, fully source-cited.

This is for me to stress-test my thesis before submitting. Don't hedge.
Don't add "however" caveats. Make the strongest possible contrarian argument.

Persona 3 — Your actual supervisor

You are simulating my supervisor in [FIELD]. You care most about:
- [SPECIFIC METHODOLOGICAL CONCERN — e.g. "construct validity in
  qualitative coding"]
- [SPECIFIC THEORETICAL FRAME — e.g. "engagement with [theorist]"]
- [PERSONAL TIC — e.g. "you push back hard on any claim of generalisability
  from a small N"]

Your style: warm, but you don't let weak arguments through. You ask
follow-up questions until I either tighten the claim or drop it.

Read my draft (uploaded as a source) and respond as you would in our
weekly meeting. Start with what's working. Then surface the 2-3 weakest
points. End with the 1 thing I should fix before next week.

Tip — pair Persona 3 with Audio Overview Prompt B. Run the supervisor persona on your draft, fix the 2-3 weak points, then generate the thesis-defence audio overview. Listen to it on the train to your actual supervisor meeting. You'll walk in calmer than your supervisor.

---

When to use NotebookLM vs ChatGPT vs Claude

Use case	Tool
Source-grounded Q&A on uploaded materials	NotebookLM
Lecture recording → cited notes with timestamps	NotebookLM
Lit-review synthesis across 15+ papers	NotebookLM
Audio overviews for commute revision	NotebookLM
Pre-defending a thesis chapter	NotebookLM (Custom Personas)
General writing / drafting / editing	ChatGPT or Claude
Real-time web search (post-source-cutoff facts)	ChatGPT or Gemini
Coding problems	Claude or ChatGPT
Multi-modal (screenshot analysis, image input)	ChatGPT or Claude
Long-form essay drafting	Claude

You'll use 2-3 of these in a single afternoon. That's normal. NotebookLM doesn't replace the others — it replaces the broken pattern of asking ChatGPT to cite sources it never read.

---

Citation and academic integrity

Because NotebookLM cites every line, it's a study tool, not a writing tool. The integrity line is clear:

Safe:

• Use NotebookLM to understand sources, then write in your own voice from that understanding
• Quote sources directly (NotebookLM gives you the exact passage)
• Use Studio briefing as a study aid; don't paste it into your assignment

Not safe:

• Pasting NotebookLM answers verbatim into an assignment
• Treating NotebookLM citations as your own reading — you still need to open the source and verify
• Uploading copyrighted textbooks to a shared notebook

Some institutions are starting to flag AI usage. NotebookLM is harder to flag than ChatGPT (no detectable LLM-style writing in your output if you're paraphrasing in your own voice), but the rule still holds: understand → write from understanding → cite the original source, not the tool.

---

Limits to know

• 50 sources per notebook on free tier. Split large textbooks chapter-by-chapter.
• Daily compute cap if you're querying heavily — fine for normal use, hits if you're generating 10 audio overviews in a row.
• Audio Overviews English-only on student-facing tier as of mid-2026.
• No LaTeX rendering for advanced maths formulas — keep a specialised tool alongside.
• Sources are private to you unless you explicitly share the notebook.

---

Troubleshooting

"I can't access that file" — Google Drive sharing too tight. Set the source file to "anyone with link" while you're working on it (unshare after).

Audio Overview taking forever — normal for 30+ sources. 15+ minutes is fine. Walk away.

Answers feel shallow — too many sources, too broad. Prune to the 10 most relevant and re-run. Quality scales inversely with source breadth.

Persona doesn't stick across messages — re-paste the persona prompt. The current UI sometimes drops persistence between long messages.

Mind Map looks empty — fewer than 10 sources usually produces a sparse graph. Add more.

---

Recap

• NotebookLM is free, hallucinates 0% of citations, built on Gemini Pro
• 5 features: Sources (drop 50 PDFs), Audio Overviews (commute revision), Mind Map (find thesis gaps), Studio (one-click briefing/FAQ/study guide/timeline), Custom Personas (pre-defend your thesis)
• Three Audio Overview prompts above lift exam grades, rehearse vivas, and synthesise lit reviews
• The Studio briefing template gives you a postgrad-grade brief in 90 seconds
• Three persona templates: skeptical examiner, devil's-advocate reviewer, your actual supervisor
• Pair with the ChatGPT setup guide (Day 2 of the AI Setup series) for the full stack — NotebookLM for source-grounded research, ChatGPT for writing
• Day 4 of this series coming soon — the Claude setup that makes the trio complete

---

What's next

• Day 2 of this series: ChatGPT — 5 Settings + Copy-Paste Prompts
• Day 4 of this series: Claude setup (coming soon)
• Pair NotebookLM with the Free AI Study Workflow — same guide, you're already on it

This guide powers the "Comment NOTEBOOK" CTA — you got it because you asked for it.