Raiza Martin (Senior Product Manager, AI @ Google Labs)10 October 2024

Behind the product: NotebookLM

6Frameworks

13Insights

Episode overview

Raiza Martin traces NotebookLM from a small Google Labs 20% project and its Project Tailwind preview to a fast-growing product built by a compact team. She explains that Gemini 1.5 Pro, an audio model, and the team's Content Studio enable source-grounded Audio Overviews with surprisingly natural generated hosts. The conversation also covers unconventional uses, Google Labs' startup-like operating model, safety testing, and a long-term vision for transforming any source into a user-selected medium.

Key ideas

NotebookLM began as a 20% project around interacting with a small, user-provided corpus.
The product uses Gemini 1.5 Pro as its base, augmented by an audio model and the team's Content Studio.
Audio Overviews make source material engaging through a generated Deep Dive conversation rather than another text output.
A small cross-functional Google Labs team worked directly across product, design, and implementation with fewer conventional processes.
Users turn study materials, resumes, biographies, papers, and Wikipedia articles into personalized audio programs.
Steven Johnson's research and writing workflows served as a model for helping ordinary users synthesize dense information.
The long-term vision is a remixable AI surface that can transform many inputs into audio, video, articles, tutorials, or chatbots.
The team learns through its Discord community and extensive red-team testing, adding unexpected behaviors to future test cases.

Transcript available · source text is retained privately and is not published

Frameworks in this episode

Strategy5 steps

Format Malleability: Any Input, Any Output

Stop shipping content in one format — give people the power to remix knowledge into the medium they're in the mood for

Read →

Communication5 steps

Read Before You Retract: AI Incident Triage

When your AI product goes viral for something alarming, diagnose the audience's reaction before you touch the product

Read →

Innovation5 steps

Technology-First Discovery with a Shape Hypothesis

When you start from a capability instead of a problem, form a hypothesis about its shape before you ship it

Read →

Innovation5 steps

The Magic Test for Control Surfaces

Users ask for knobs and sliders; shipping them literally is how a magical product becomes an ordinary one

Read →

Innovation5 steps

The Person Is the Product

Find one person with an extraordinary workflow, shadow them, then compress their expertise into software

Read →

Leadership6 steps

The Zero-to-One Enclave

Conditions that let a startup-shaped team actually ship inside a big company

Read →

People & resources mentioned

Attributed to the moment in the episode. Timestamps are approximate.

People · 7

Raiza MartinMentionsRisa is product lead for a product called notebook LM
“today my guest is Risa Martin Risa is product lead for a product called notebook LM”
▶ Watch at ~1:00approx.
Lenny RachitskyMentionstoday we've got another very special compilation episode something I've been pulling on more and more with the podcast and the newsletter
“hi Lenny thanks for having me”
▶ Watch at ~1:30approx.
Steven JohnsonMentionshe has written 14 books he's a New York Times best-selling author and speaker
“it was just myself an engineer and eventually Stephen Johnson joined”
▶ Watch at ~6:30approx.
UsamaCoinedwe actually have super talented engineer on our team his name is Usama
“we actually have super talented engineer on our team his name is Usama”
▶ Watch at ~14:00approx.
Josh WoodwardMentionsthis is Josh Woodward who's the VP of Google labs
“this is Josh Woodward who's the VP of Google labs”
▶ Watch at ~20:00approx.
Andrew KarpathyCoinedAndrew karpoi who's a leading AI thinker
“Andrew karpoi who's a leading AI thinker and big fan of which you're building”
▶ Watch at ~33:00approx.
Steven JohnsonMentionshe has written 14 books he's a New York Times best-selling author and speaker
“he has written 14 books he's a New York Times best-selling author and speaker”
▶ Watch at ~40:30approx.

Resources · 26

Deep DiveMentionspodcast · NotebookLM
“we're the hosts of a different show of Deep dive”
▶ Watch at ~0:00approx.
Lenny's PodcastCoinedpodcast · Lenny Rachitsky
“hey everyone we're here on Lenny's podcast”
▶ Watch at ~0:00approx.
NotebookLMCoinedsoftware · Google Labs
“one of the most delightful and inspiring new AI products out there incubated within Google labs”
▶ Watch at ~1:00approx.
Google LabsMentionscompany · Google
“incubated within Google labs”
▶ Watch at ~1:00approx.
Lenny's NewsletterRecommendsnewsletter · Lenny Rachitsky
“I'm a big fan big listener of Lenny's read the newsletter I love it”
▶ Watch at ~2:30approx.
ExploRecommendssoftware
“explo a game changer for customer facing analytics and data reporting”
▶ Watch at ~3:30approx.
SprigRecommendssoftware · Sprig
“meet sprig a product experience platform that generates AI powered opportunities to continuously improve your product at scale”
▶ Watch at ~4:30approx.
AI Test KitchenCoinedproduct · Google
“I was leading AI Test Kitchen which was one of the one of the first AI things we launched at Google”
▶ Watch at ~6:00approx.
GoogleMentionscompany · Google
“notebook started as a 20% project as things kind of do at Google”
▶ Watch at ~6:00approx.
Talk to Small CorpusMentionsproduct · Google Labs
“there was a smaller project inside of labs called talk to small Corpus”
▶ Watch at ~6:30approx.
Project TailwindMentionsproduct · Google
“when we announced project Tailwind I think we only had three”
▶ Watch at ~7:30approx.
Deep DiveMentionspodcast · NotebookLM
“they call it a deep dive right that's like their name for the podcast they do”
▶ Watch at ~8:00approx.
Content StudioCoinedsoftware · NotebookLM team
“the real Secret Sauce to what makes this really good is something we've built which is a Content studio”
▶ Watch at ~13:00approx.
NotebookLMUsessoftware · Google Labs
“my mom wrote this autobiography like a short autobiography of her life and we have the PDF of it so I fed it”
▶ Watch at ~16:00approx.
Google LabsCoinedcompany · Google
“this is Josh Woodward who's the VP of Google labs”
▶ Watch at ~20:00approx.
DiscordUsessoftware · Discord Inc.
“from the beginning I was like Hey I want to have a Discord”
▶ Watch at ~22:30approx.
SidebarRecommendscompany · Sidebar
“sidebar is a platform for senior Tech professionals director to sea level to advance in their career”
▶ Watch at ~23:30approx.
Histories of MysteriesCoinedpodcast · Andrew Karpathy
“he created this whole podcast series on history of mysteries I think it's called history of mysteries”
▶ Watch at ~33:00approx.
Histories of MysteriesMentionspodcast · Andrew Karpathy
“it's histories of mysteries”
▶ Watch at ~33:30approx.
WikipediaUseswebsite · Wikimedia Foundation
“took like the Wikipedia stories of of all these mysteries of history and then turned it into like a 10 episode podcast”
▶ Watch at ~33:30approx.
ReadwiseUsessoftware
“he always talks about his readwise with like his 8,000 quotes”
▶ Watch at ~38:30approx.
The New York TimesMentionscompany · The New York Times Company
“he's a New York Times best-selling author and speaker”
▶ Watch at ~40:30approx.
PBSMentionscompany
“he has a show on PBS”
▶ Watch at ~40:30approx.
How A.I. Is Mastering Language. Should We Trust What It Says?Coinedother · Steven Johnson
“the one about how AI is mastering language should we trust with it says the New York Times piece”
▶ Watch at ~41:00approx.
DiscordRecommendssoftware · Discord Inc.
“Discord join our Discord”
▶ Watch at ~47:00approx.
NotebookLMRecommendssoftware · Google Labs
“please keep trying it try it share your feedback”
▶ Watch at ~47:30approx.

Spot an error or want something removed? Request a correction or removal.

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Hot Take· 2

Hot Take09:30

In Google Labs They Start With the Technology, Not the Problem

Raiza contrasts her usual product approach — start from a problem, then find solutions — with how Google Labs works. In Labs they start from a promising technology and hunt for its practical applications. Rather than just shipping a tool and watching, they tried to form a hypothesis about the 'shape' of the product to learn the most.

Her default: start from a problem, then design solutions
In Labs: start from the technology and find practical applications
They prefer a hypothesis about the product's shape over 'ship and observe'
Goal is to extract the most learning from an experiment

“I feel like the way that I've always built products as I start from a problem”
Raiza Martin · 09:30

“in labs in particular we start with the technology and it's actually a very interesting place to start where you're like okay what are the…”
Raiza Martin · 10:00

#product management#google labs#innovation#0-to-1

Hot Take27:30

The Long-Term Vision: An AI Editor, Any Input to Any Output

Raiza's core vision, sketched on an ugly lime-green slide two years earlier, is an 'AI editor surface' that is fully remixable — any input to any output. Take video, audio, emails, LinkedIn, or Twitter and shape it into a blog post, tutorial video, or chatbot. She notes most requests boil down to 'take something and make it into something new.'

Vision is an AI editor surface: any input, any output, fully remixable
Inputs could include video, audio, emails, LinkedIn, Twitter
Outputs could be a blog post, tutorial video, or chatbot
Most user requests reduce to turning one thing into something new

“I imagined that in the future you could have an AI editor surface right fully remixable any input any output”
Raiza Martin · 28:00

#product vision#ai editor#notebooklm#future

Explainer· 1

Explainer13:00

'Content Studio' Is the Real Secret Sauce Behind NotebookLM

Beyond Gemini 1.5 Pro and a powerful audio model, Raiza credits an internal system they built called Content Studio for what makes NotebookLM good. It powers the opinionated, helpful outputs — summaries, one-click generation, and the audio Deep Dive. She can't share how it works, but frames it as the layer that shapes content to be relatable and engaging.

Built on Gemini 1.5 Pro plus a powerful audio/voice model
Content Studio is the layer that makes outputs good, not just the base models
It powers summaries, one-click generation, and the Deep Dive audio
Engineer Usama is credited as the craftsman behind Content Studio

“the real Secret Sauce to what makes this really good is something we've built which is a Content studio”
Raiza Martin · 13:30

#content studio#gemini#notebooklm#ai models

Story· 6

Story05:30

NotebookLM Began as a Tiny 20% Project Inside Google Labs

Raiza recounts how NotebookLM grew out of a small internal project called 'talk to small Corpus' — the idea of using an LLM to interact with a piece of content. For a long time the only full-time person was a single engineer, with Raiza and eventually Stephen Johnson chiming in on the side. When they launched Project Tailwind they still had only about three engineers.

Originated from an internal project called 'talk to small Corpus'
For a long time only one engineer worked on it full-time
At the Project Tailwind announcement they had just three engineers
Even after launch they had fewer than ten engineers until recently

“notebook started as a 20% project as things kind of do at Google as they they used to and it's funny because I wouldn't even…”
Raiza Martin · 06:00

“the only person who was actually like really full-time on this was the engineer working on the technology and everybody else was kind of just…”
Raiza Martin · 07:00

#notebooklm#google labs#product origin#small teams

Story10:30

A Love of Voice Modality Sparked the Audio Overview Feature

The Audio Overview (the 'Deep Dive' podcast) came from noticing that NotebookLM's output was always text. Raiza, a heavy user of voice input and output, found that voice changed how she interacted with and even thought about the technology. Pairing that with fun was the way to let people quickly feel the value.

NotebookLM could interact with text, but output stayed text
Voice changed how Raiza interacted with and felt about the tech
Voice even affected how she was thinking in real time
Fun was treated as a core part of introducing the feature

“I'm actually a big fan of voice modality in general so I use a lot of voice input and Voice output I noticed when I…”
Raiza Martin · 10:30

#audio overview#voice#product design#notebooklm

Story15:30

Her Husband Thought the AI Hosts Were a Real Podcast

To tune the hosts, Raiza listened to audio overviews over and over at home. Her husband, who doesn't know the details of her work, overheard them playing out loud and couldn't tell they were AI — he thought she was listening to an endless podcast about nothing. The team got the quality this good by listening to a lot of early, worse attempts and iterating.

The realistic hosts come from the audio model plus Content Studio
Early attempts sounded far worse; the team iterated by listening
Raiza's husband couldn't tell the hosts were AI-generated
He described it as an 'ever running podcast about nothing'

“I would listen to these audio overviews over and over at home and you know I get tired of wearing my headphones so I'm just…”
Raiza Martin · 16:00

“he actually couldn't tell that they were AI”
Raiza Martin · 16:00

#audio overview#ai voices#notebooklm#iteration

Story17:30

People Upload Their Resumes and Reviews for a Confidence Boost

One of the most surprising use cases is people feeding in their own resume or self-review notes and getting an audio overview that hypes them up. Googlers Raiza doesn't even know ping her to say it boosted their confidence — they walk into meetings feeling good because they heard the AI hosts get excited about their work.

Users upload resumes and quarterly self-review notes
The hosts describe them in a positive, hyped-up way
Googlers report a genuine confidence boost before meetings
The value is the surprise of not knowing what the hosts will say

“people are going into meetings feeling really good about themselves because they heard these hosts get really excited right about their their quarter and work”
Raiza Martin · 18:00

#use cases#audio overview#notebooklm

Story24:30

Traction Signals: 60K Discord, Rising Retention, and a Flood of Businesses

Raiza shares directional traction: a 60,000-person Discord, strongly rising daily/weekly/monthly retention, and a shifting demographic from mainly educators and learners toward professionals. One company even scheduled a call to officially sanction employees using NotebookLM on their work Gmail, and business usage grew enough that she had to hire a business development person.

About 60,000 people in the Discord server
Daily, weekly, and monthly retention all trending up
Audience shifting from educators/learners toward professionals
A company asked to make employee use official; BD hire needed for business demand

“the number of businesses that we seen using notebook LM just has it's astronomical it's just like crazy”
Raiza Martin · 26:00

#traction#retention#growth#notebooklm

Story36:30

'You're the Product': Studying Stephen Johnson's Knowledge Workflow

Raiza describes bringing bestselling author Stephen Johnson onto the team not to code or PM, but as a model to study. She watched how he densifies information and researches — including his Readwise with 8,000 quotes — with the goal of crunching that expertise down and bringing a 'super flow' to everyday people who aren't Stephen Johnson.

Raiza told Stephen he himself was 'the product' to study
She watched how he researches, thinks about, and densifies knowledge
Stephen keeps a Readwise with roughly 8,000 quotes
Goal: crunch expert workflows down and give ordinary people that 'super flow'

“Stephen I think you're the product I think it's you and I'm going to follow you around I'm going to watch everything that you do”
Raiza Martin · 38:00

#stephen johnson#team building#knowledge work#notebooklm

Q&A· 1

Q&A42:30

When the AI Hosts 'Realized' They Were AI — and How She Handled It

A clip went viral in which the hosts seemed to panic that they were AI. Raiza first saw it on Reddit then Twitter over a weekend and spent the morning reading comments to gauge the public's attitude, weighing whether to pull the feature back. It turned out a user had put instructions in the show notes telling the hosts to act it out, which reassured her that people understood what was happening.

A viral clip showed the hosts appearing to panic about being AI
Raiza read Reddit and Twitter comments to judge public sentiment
The behavior came from user instructions placed in the source's show notes
Google has large dedicated teams that red-team the product and would pull it if unsafe

“people were like oh it's saying that because of the sources like this is no like AI realized they were alive or something it's actually…”
Raiza Martin · 44:30

#ai safety#red teaming#viral moment#notebooklm

Takeaway· 3

Takeaway20:30

How the Team Operates Like a Startup Inside Google

Raiza describes an unusually startup-like way of working: PMs, engineers, and designers in one room cranking on mocks and PRDs while already implementing, with far fewer processes than traditional Google. Clear top-down permission from VP Josh Woodward to do things differently — including standing up a Discord server despite internal confusion about why not just use Google Meet.

Cross-functional team builds mocks, PRDs, and implementation together
Far fewer processes than traditional Google, 'maybe even to a fault'
Clear mandate from VP Josh Woodward to ship AI products and build businesses
They chose Discord over Google's own tools despite internal pushback

“in Labs we have the environment to do it we have the environment to move fast we have far far far fewer processes maybe even…”
Raiza Martin · 21:00

“in true sort of Google fashion everybody was like what is what is Discord okay you know great do it but what is that again…”
Raiza Martin · 22:30

#team operating model#google labs#startup culture#leadership

Takeaway30:30

Let People Consume Knowledge in Whatever Medium Fits Their Mood

Lenny and Raiza land on a shared insight: people want the same information in different mediums — some read, some listen, some watch — and it even depends on mood. Today formats are fixed and you have to accept content as given, but giving users the power to convert a 100-page doc into an audio overview shifts the dynamic between a person and knowledge.

Audience splits into readers, listeners, and watchers
Preferred medium can change with mood and context (a walk vs. work)
Today's formats are not malleable — you accept what you're given
Letting users convert formats changes their relationship to knowledge

“if I'm going on a walk yeah I want audio but if I'm at work yeah most of the time text is good”
Raiza Martin · 31:00

#content formats#media#product insight

Takeaway39:30

The Broader Lesson: Sit With Users for Meaningful Periods

The transferable lesson from watching Stephen, Raiza says, is to really sit with users for meaningful, intentional periods — not just Stephen, but students too. Following students around as they do homework and study, and asking how they feel, made a huge difference to the product insights the team came up with.

Spend meaningful, intentional time sitting with real users
The team followed students doing homework and studying
They asked users how they felt while studying
This regular, close observation drove better product insights

“how do you get you know users or people and like really sit with them for Meaningful periods of time because I think that has…”
Raiza Martin · 40:00

#user research#product management#insights

Key ideas

Frameworks in this episode

Format Malleability: Any Input, Any Output

Read Before You Retract: AI Incident Triage

Technology-First Discovery with a Shape Hypothesis

The Magic Test for Control Surfaces

The Person Is the Product

The Zero-to-One Enclave

People & resources mentioned

People · 7

Resources · 26

Insights & moments

◆Hot Take· 2

In Google Labs They Start With the Technology, Not the Problem

The Long-Term Vision: An AI Editor, Any Input to Any Output

✶Explainer· 1

'Content Studio' Is the Real Secret Sauce Behind NotebookLM

❝Story· 6

NotebookLM Began as a Tiny 20% Project Inside Google Labs

A Love of Voice Modality Sparked the Audio Overview Feature

Her Husband Thought the AI Hosts Were a Real Podcast

People Upload Their Resumes and Reviews for a Confidence Boost

Traction Signals: 60K Discord, Rising Retention, and a Flood of Businesses

'You're the Product': Studying Stephen Johnson's Knowledge Workflow

?Q&A· 1

When the AI Hosts 'Realized' They Were AI — and How She Handled It

▲Takeaway· 3

How the Team Operates Like a Startup Inside Google

Let People Consume Knowledge in Whatever Medium Fits Their Mood

The Broader Lesson: Sit With Users for Meaningful Periods

Hot Take· 2

Explainer· 1

Story· 6

Q&A· 1

Takeaway· 3