A year of interviews, videos, or voice memos contains a book. A long, unstructured transcript contains a polished podcast script. NotebookLM reads across your entire corpus and extracts the structure, voice, and ideas that are already there — so you're not writing from scratch, you're curating what you've already said.
Prompts6 free + 24 premium
Output formatsBook · Podcast · Article · Course
Input typesTranscripts · Interviews · Notes
Best forCreators, coaches, speakers, experts
Premium access$19.99 lifetime
Most content repurposing tools treat this as a summarization problem: take the long thing and make it shorter. That misses the actual challenge. The hard part isn't condensing — it's restructuring. An interview rambles. A video transcript reads nothing like prose. A year of weekly videos has no arc that's visible to the person who made them.
NotebookLM solves the restructuring problem because it can read 25 transcripts simultaneously, find the ideas that recur across all of them, and propose a structure based on what you've actually said — not what you think you said. The result is a book or podcast that sounds like you, because it is you. Every quote, insight, and framework came from your own mouth. NotebookLM just found the architecture.
Book Architecture
The book that's already in your content
NotebookLM reads across all transcripts and proposes a chapter structure based on recurring ideas — organized as a coherent argument, not a content dump.
Voice Extraction
Your specific style and vocabulary
Before any drafting, extract your voice patterns: recurring phrases, argument structures, analogies, and the specific way you explain your core ideas.
Podcast Scripts
From raw transcript to clean episode
Turn a single rambling interview into a tightly structured solo or co-hosted episode — with intro hook, segment breaks, and a clear close.
Perspective Shaping
Reframed for a specific audience or voice
The same transcript, restructured from a founder's perspective vs. an analyst's perspective vs. a practitioner's perspective. Same facts, different frame.
Idea Mining
Frameworks you didn't know you had
NotebookLM identifies implicit frameworks — ideas you reference repeatedly but have never formalized — and names them so they become teachable concepts.
Chapter Drafts
First drafts from your own words
Each chapter built from direct quotes, paraphrased in your voice. No hallucination risk — every claim traces back to something you said in a transcript.
Setup Tutorial
One-time setup · 45–90 min
The quality of your output is almost entirely determined by the quality of your transcript corpus. Here's what to include and how to prepare it for the best possible results.
Source type
What to include
Preparation tip
Video transcripts
Your best 15–25 videos, prioritizing depth over breadth. Choose videos where you explain your thinking at length — tutorials, Q&As, and teaching content work better than highlight reels.
Add the video title and rough category (e.g. "TOPIC: Leadership / DATE: 2024-03") to the top of each transcript file. Clean auto-captions if they're badly broken — obvious word errors confuse the AI more than filler words do.
Interview transcripts
Interviews where you (or the subject) speak at length about their ideas, experiences, or methodology. Include both sides of the conversation — interviewer questions provide useful context for the answers.
Label speakers clearly throughout: "INTERVIEWER:" and "GUEST:" or the person's name. If the interview is with a specific individual whose voice you're capturing, their name should be consistent throughout.
Voice memos or notes
Rough thinking-out-loud recordings, unfinished drafts, or notes to self. These are often the richest source of authentic voice and original ideas — use them even if they feel too raw.
Transcribe with Whisper, Otter.ai, or similar. Label with date and context at the top. Don't clean too heavily — the rambling is often where the best frameworks are hiding.
The subject's bio & context
A document describing who the person is, their expertise, their audience, their main body of work, and the transformation their content creates for readers or viewers. Essential for perspective-shaping prompts.
Write this in 3–5 paragraphs from the first person. Include their specific vocabulary — the terms, phrases, and frameworks they already use. This is what makes output sound like them specifically.
Existing published work
Articles, book chapters, essays, or course materials already published by the subject. Provides a baseline for their public voice — different from their unpolished transcript voice.
Include these to let NotebookLM see the gap between "thinking out loud" and "published voice." This gap informs the editing instructions you'll give in later prompts.
Step 01
Create a corpus notebook — don't mix projects
Create one NotebookLM notebook for this specific project: one person's content, one output format. If you're building a book from someone's YouTube videos, that's one notebook. If you're building a podcast from their interviews, that's a separate notebook. Cross-contaminating sources produces incoherent output.
Exception: Include published writing alongside transcripts in the same notebook only if the same person wrote both. The contrast between their polished and unpolished voice is useful data.
Step 02
Run the voice extraction prompt before any content creation
Before generating any drafts, run a voice extraction prompt (Prompt 3 below). This gives you a style reference document you can paste into every subsequent prompt — ensuring all output sounds like the specific individual, not generic AI prose.
Save the output: Copy the voice style guide into a plain text file. Paste it into every drafting prompt as a prefix: "Write in this voice: [paste guide]."
Step 03
For books: structure before drafting
Always generate chapter structure before asking for any chapter drafts. Run the book architecture prompt, review the proposed outline, revise it to match your vision, and only then ask NotebookLM to draft one chapter at a time — providing the chapter structure in each prompt.
Chapter prompting format: "Using only content from the transcripts, draft Chapter 3: [title]. The chapter should cover [3–4 bullet points from your revised outline]. Write in the voice described in this style guide: [paste guide]. Maximum 1,200 words."
Step 04
For podcasts: transcript first, script second
Don't ask NotebookLM to write a podcast script from scratch. Instead: first ask it to extract the key ideas and quotes from the source transcript, then ask it to structure those into a podcast outline, then ask for a polished script from the outline. Three steps produces far better results than one.
Step 05
Always request citations for quotes
In every drafting prompt, add: "For every direct or paraphrased quote you use, cite the source transcript by name in brackets after the sentence." This makes fact-checking trivial and ensures you can verify that no content was fabricated.
Why this matters: NotebookLM is grounded in your sources, but adding an explicit citation requirement catches the rare case where it paraphrases something loosely. It also makes it easy to go back to the original for fuller context.
Repurposing Prompts
6 Free · 24 Premium
Foundation Prompts
6 prompts — free
Run these before generating any content. They extract the structure, voice, and ideas already present in your corpus.
Recurring idea finder
"Read all transcripts in this notebook and identify the 8–12 ideas, themes, or beliefs that recur most frequently across the content. For each recurring idea: name it in 3–5 words, quote the most articulate version of it from any transcript (cite the source), and note in how many separate transcripts it appears. These recurring ideas are the core concepts around which any book, course, or podcast series should be structured."
Book architecture generator
"Based on all the ideas in this notebook, propose a complete book structure with 6–10 chapters. For each chapter: (1) a working title, (2) the central argument in one sentence, (3) 3–4 sub-points covered, and (4) which transcripts contain the most relevant content. Organize the chapters as a coherent argument that builds — not a collection of separate topics. The book should have a beginning (the problem), middle (the framework or methodology), and end (the transformation or outcome)."
Voice style extractor
"Analyze all transcripts in this notebook and write a voice style guide for the speaker. Include: (1) 10 specific words, phrases, or expressions they use repeatedly that are distinctive to them, (2) how they typically structure an argument (do they lead with the conclusion or build to it?), (3) how they use analogies — what domains do they draw from?, (4) their typical sentence length and rhythm, (5) how they handle uncertainty or qualification, and (6) 3 example sentences written in their voice that could appear in a book. This style guide will be used to ensure all drafted content sounds like them."
Single transcript podcast outline
"Take the transcript titled [source name] and restructure its content as a 20–25 minute podcast episode. Produce: (1) a 3-sentence hook that opens the episode and earns the listener's attention, (2) a clear episode structure with 4–5 segments, each with a title and 2-sentence summary of what's covered, (3) 3 specific quotes from the transcript that could be spoken verbatim in the episode, and (4) a closing summary that gives the listener one clear takeaway. Do not fabricate any content — everything must come from the transcript."
Replace [source name] with the specific transcript file name before using.
Framework namer
"Read all transcripts and identify 3–5 ideas that the speaker explains consistently but has never given a formal name or label to — implicit frameworks, mental models, or methodologies they reference repeatedly. For each unnamed framework: (1) describe what it is in 2–3 sentences, (2) quote the clearest explanation from any transcript (cite source), (3) propose 2–3 possible names for it that match the speaker's vocabulary and style. Named frameworks are far more memorable and teachable than unnamed ones."
Perspective reframe prompt
"Take the transcript titled [source name] and rewrite its core ideas as if they were being explained from the perspective of [role: e.g. a first-year practitioner / a senior executive / a skeptic who was convinced / a teacher explaining to students]. Keep all the facts, examples, and frameworks from the original transcript — only change the framing, tone, and emphasis to match the specified perspective. Show the first 300 words of the reframed version, then explain what you changed and why."
Replace [source name] and [role] before using. This prompt produces different versions of the same content for different audiences.
Premium — 24 More Prompts
From raw transcripts to polished, publishable long-form.
The remaining 24 prompts cover full chapter drafting, podcast script generation, course module creation, ghost-writing workflow, interview-to-article conversion, and a complete book-production pipeline from corpus to manuscript.
All 30 prompts + 5 other premium guides — one-time access
🔒 Chapter drafter — full first draft of any chapter using only transcript content, with citations
🔒 Introduction writer — craft an opening chapter that establishes stakes, voice, and the book's central argument
🔒 Chapter transition writer — bridge sections so the book reads as a continuous argument, not separate essays
🔒 Conclusion synthesizer — write a closing chapter that lands the transformation promised in the introduction
🔒 Anecdote extractor — pull every specific story, example, or case study from all transcripts with source citations
🔒 Podcast solo script generator — 15–25 minute monologue script from a single transcript, production-ready
🔒 Co-hosted episode formatter — restructure an interview transcript as a co-hosted podcast with natural dialogue flow
🔒 Episode series planner — design a 6–10 episode podcast series from a set of thematically related transcripts
🔒 Show notes writer — complete episode show notes with timestamps, key quotes, and resource links
🔒 Article converter — turn any single transcript into a 1,000–1,500 word bylined article with a clear argument
🔒 LinkedIn long-form adapter — reshape transcript insights into LinkedIn article format for professional audiences
🔒 Email course creator — convert a topic cluster into a 5-day email course with daily lessons and exercises
🔒 Course module planner — structure transcript content into a 4–6 module online course with learning objectives
🔒 Workbook exercise generator — create reflection questions and exercises from framework content in the transcripts
🔒 Ghostwriting voice calibration — create a detailed ghostwriting brief a writer can use to match the subject's voice
🔒 Quote bank builder — extract 50 standalone quotable sentences, formatted for social media attribution
🔒 Contradiction finder — identify places where the speaker's position has evolved or contradicts itself across transcripts
🔒 Gap and hole identifier — what topics does the audience need that the existing transcripts don't yet cover?
🔒 Testimonial and case study extractor — pull every client result or transformation story mentioned across transcripts
🔒 FAQ generator — build a comprehensive FAQ from the questions answered most frequently across the corpus
🔒 Indexed glossary builder — extract and define every domain-specific term the speaker uses consistently
🔒 Book proposal drafter — generate a complete non-fiction book proposal outline: overview, market, chapter summaries
🔒 Audiobook adaptation notes — identify which sections need rewriting for audio listening vs. visual reading
🔒 Multi-format repurposing map — for any single transcript, map all 8+ formats it could be repurposed into
Important — On Accuracy and Citation
NotebookLM is grounded in your uploaded sources, which makes it significantly more reliable than open-ended AI tools for this use case. But it can still paraphrase loosely or merge similar ideas from different transcripts. Always add a citation request to every drafting prompt ("cite the source transcript in brackets after every quote or paraphrase") and spot-check quotes against originals before publishing.
The goal is to use AI for structure and first drafts, not for final copy. Everything that goes into a published book or podcast should be reviewed by the person whose voice it represents.