Raiza Martin
Senior Product Manager, AI at Google Labs, leads NotebookLM
Where to hear and read about Raiza Martin1 episode or test
Raiza Martin is a Senior Product Manager, AI at Google Labs who leads NotebookLM, which grew from a part-time side project into a product with a 60,000 member Discord. The thing most worth learning from her is to shape new model technology into a form people already understand, and to learn by watching real users work.
A lot of technology, I feel like you have to shape it and bring it closer to people.
Behind the product: NotebookLM | Raiza Martin (Senior Product Manager, AI @ Google Labs)
Raiza Martin: the story and the lessons
1How Raiza Martin built NotebookLM from a side project
Martin came from payments and ads but had earlier worked only at startups. She followed her old boss Josh Woodward into Google Labs, which was about three years old with nobody there when she joined. She saw a chance at zero to one again.
She told Josh early that real zero to one means doing things very differently.
NotebookLM began as a 20% project that took far more than 20%. Around 2022 Martin was leading AI Test Kitchen, one of Google's first AI launches. A smaller Labs project, Talk to Small Corpus, let an LLM interact with one piece of content, and the team saw a nugget in it.
The early group was Martin, one full-time engineer and, eventually, author Steven Johnson. The others joined because the idea was interesting.
A tiny cross-functional team shipped it and kept moving fast. Project Tailwind launched with three engineers and a designer, and only about eight engineers worked on the product over the year before the last month.
PMs, engineers and designers meet once, and engineers start implementing while the group works on mocks and requirements. In her words: "we'll just crank on the mocks and the PRDs at the same time."
Audio Overviews, previewed at I/O in May, made NotebookLM spread. NotebookLM was already a source-grounded chat product with users when the audio feature arrived. Its Discord now has about 60,000 members, and retention has risen on daily, weekly and monthly measures.
Early users were mostly educators and learners. Professionals and businesses are now a big part of the mix.
2How Martin runs a startup inside Google
Martin got senior air cover and far fewer processes than the rest of Google. Labs sits under VP Josh Woodward, whose mission is to ship AI products and build businesses out of them. That let a small team build in public, tweet daily and move quickly.
She admits the lack of process can be a fault: "We have far, far, far fewer processes, maybe even to a fault."
She started a Discord server on day zero, against internal habit. Colleagues asked why not Google Meet or a Google group, then told her to go ahead. Her biggest fear on day zero was that nobody would join.
The server became a source of daily direct feedback. Martin reads everything posted there, even when she does not reply.
3How Martin turns model technology into a product people get
Martin hypothesizes the product's shape when Labs starts from technology. The audio work began when another Labs team offered powerful audio models and asked for a good application. Rather than only releasing a tool and watching, the team forms a view of what the thing should be so each launch teaches the most.
Her insight was that AI outputs were always text, and voice output changed how she interacts with and feels about the technology.
She wrapped existing models in a familiar form, a two-host podcast. Audio Overviews turns any uploaded source into a spoken show. It runs on Gemini 1.5 Pro with an audio model on top.
The reaction resembles the ChatGPT moment: the technology already existed, and a new medium made people see its power.
The quality comes from an opinionated content layer, not only the audio model. A content studio built on top of the model takes a view on the user's content. Early attempts were nowhere near as good, so the team spent a lot of time listening and tuning.
The Deep Dive is the first format, with more planned. Martin calls the content studio "the real secret sauce."
Letting users upload their own documents gives a surprising first result. Resumes and Google quarterly check-in notes became favorites because two hosts hype up the user's own work. Many Googlers messaged Martin that it boosted their confidence before meetings.
Her guess is that the unknown outcome of one click adds to the delight.
Her long-term vision is any input becoming any output. She drew it two years ago on a lime green slide meant to feel un-Googly. Video, audio, emails or posts become a blog post, tutorial video or chatbot, and mobile is the biggest gap today.
Format depends on mood, such as audio on a walk and text at work.
She refuses plain knobs and sliders for controls that do not feel magical. Users asked for ways to steer the one-shot audio episodes. When Martin saw the mocks, they did not feel magical or like the rest of the product, so she is designing a better control experience.
4How Martin learns from users and experts
Martin sits with real users for long, regular stretches. The team follows students as they do homework and study, and asks how they feel. She says doing it regularly and on purpose makes a huge difference to the insights.
She treated Steven Johnson's research workflow as the product template. She told him "I think you are the product" and watched how he works, including a Readwise holding about 8,000 quotes. Her goal was to shrink his research time and give everyday people his skill at de-densifying information.
Johnson has written 14 books. They have clashed a lot but always end up aligned on the next step, which she says matters for PMs.
She quizzed her VP about a 50-page vision doc instead of reading it. Josh Woodward gave her the doc, and she asked him questions like a chatbot. His answer was "It's in the doc, Raiza."
She treats it as proof of the demand for conversational access to dense material, the core NotebookLM use case.
5Martin on business demand, viral moments and safety
Martin watches for enterprise pull from employees using it on personal accounts. One company found staff using NotebookLM on personal Gmail accounts and asked to make work use official. She now takes customer calls daily and needs a business development hire.
She sees a path through enterprise features like single sign-on and SOC 2.
She builds something interesting first, then the business. Her mandate is to build a business, sequenced as a product people latch onto first and commercialization in parallel with deeper experience. Google already has cloud, Workspace and consumer routes.
The near-term focus is game-changing workflows for professional knowledge workers, alongside educators and learners.
Martin read the comments before reacting to a viral AI clip. A clip showed the hosts seeming to realize they are AI. She spent the morning reading Reddit and Twitter comments and found people saw it followed show notes the user uploaded.
That made her confident people get it, so she addressed it publicly on Twitter instead of pulling the feature back.
Google expands red-team test cases whenever users find a new gap. Large red teams test as many risk areas as they can imagine. A situation not anticipated is added to the test cases.
The team would pull a feature back if it felt clearly unsafe, though the aim is not to need to.
Behind the product: NotebookLM | Raiza Martin (Senior Product Manager, AI @ Google Labs)
NotebookLM grew from a 20% project run by one engineer, a PM and author Steven Johnson into a product with a 60,000 member Discord and fast enterprise interest. The episode shows how a tiny team ships fast inside Google, wraps new model tech in a delightful format (Audio Overviews), and learns by watching users work.
Martin came from payments and ads but had earlier worked only at startups. She followed her old boss Josh Woodward into Google Labs, which was about three years old with nobody there when she joined. She saw a chance at zero to one again.
She told Josh early that real zero to one means doing things very differently.
NotebookLM began as a 20% project that took far more than 20%. Around 2022 Martin was leading AI Test Kitchen, one of Google's first AI launches. A smaller Labs project, Talk to Small Corpus, let an LLM interact with one piece of content, and the team saw a nugget in it.
The early group was Martin, one full-time engineer and, eventually, author Steven Johnson. The others joined because the idea was interesting.
A tiny cross-functional team shipped it and kept moving fast. Project Tailwind launched with three engineers and a designer, and only about eight engineers worked on the product over the year before the last month.
PMs, engineers and designers meet once, and engineers start implementing while the group works on mocks and requirements. In her words: "we'll just crank on the mocks and the PRDs at the same time."
Audio Overviews, previewed at I/O in May, made NotebookLM spread. NotebookLM was already a source-grounded chat product with users when the audio feature arrived. Its Discord now has about 60,000 members, and retention has risen on daily, weekly and monthly measures.
Early users were mostly educators and learners. Professionals and businesses are now a big part of the mix.
Martin got senior air cover and far fewer processes than the rest of Google. Labs sits under VP Josh Woodward, whose mission is to ship AI products and build businesses out of them. That let a small team build in public, tweet daily and move quickly.
She admits the lack of process can be a fault: "We have far, far, far fewer processes, maybe even to a fault."
She started a Discord server on day zero, against internal habit. Colleagues asked why not Google Meet or a Google group, then told her to go ahead. Her biggest fear on day zero was that nobody would join.
The server became a source of daily direct feedback. Martin reads everything posted there, even when she does not reply.
Martin hypothesizes the product's shape when Labs starts from technology. The audio work began when another Labs team offered powerful audio models and asked for a good application. Rather than only releasing a tool and watching, the team forms a view of what the thing should be so each launch teaches the most.
Her insight was that AI outputs were always text, and voice output changed how she interacts with and feels about the technology.
She wrapped existing models in a familiar form, a two-host podcast. Audio Overviews turns any uploaded source into a spoken show. It runs on Gemini 1.5 Pro with an audio model on top.
The reaction resembles the ChatGPT moment: the technology already existed, and a new medium made people see its power.
The quality comes from an opinionated content layer, not only the audio model. A content studio built on top of the model takes a view on the user's content. Early attempts were nowhere near as good, so the team spent a lot of time listening and tuning.
The Deep Dive is the first format, with more planned. Martin calls the content studio "the real secret sauce."
Letting users upload their own documents gives a surprising first result. Resumes and Google quarterly check-in notes became favorites because two hosts hype up the user's own work. Many Googlers messaged Martin that it boosted their confidence before meetings.
Her guess is that the unknown outcome of one click adds to the delight.
Her long-term vision is any input becoming any output. She drew it two years ago on a lime green slide meant to feel un-Googly. Video, audio, emails or posts become a blog post, tutorial video or chatbot, and mobile is the biggest gap today.
Format depends on mood, such as audio on a walk and text at work.
She refuses plain knobs and sliders for controls that do not feel magical. Users asked for ways to steer the one-shot audio episodes. When Martin saw the mocks, they did not feel magical or like the rest of the product, so she is designing a better control experience.
Martin sits with real users for long, regular stretches. The team follows students as they do homework and study, and asks how they feel. She says doing it regularly and on purpose makes a huge difference to the insights.
She treated Steven Johnson's research workflow as the product template. She told him "I think you are the product" and watched how he works, including a Readwise holding about 8,000 quotes. Her goal was to shrink his research time and give everyday people his skill at de-densifying information.
Johnson has written 14 books. They have clashed a lot but always end up aligned on the next step, which she says matters for PMs.
She quizzed her VP about a 50-page vision doc instead of reading it. Josh Woodward gave her the doc, and she asked him questions like a chatbot. His answer was "It's in the doc, Raiza."
She treats it as proof of the demand for conversational access to dense material, the core NotebookLM use case.
Martin watches for enterprise pull from employees using it on personal accounts. One company found staff using NotebookLM on personal Gmail accounts and asked to make work use official. She now takes customer calls daily and needs a business development hire.
She sees a path through enterprise features like single sign-on and SOC 2.
She builds something interesting first, then the business. Her mandate is to build a business, sequenced as a product people latch onto first and commercialization in parallel with deeper experience. Google already has cloud, Workspace and consumer routes.
The near-term focus is game-changing workflows for professional knowledge workers, alongside educators and learners.
Martin read the comments before reacting to a viral AI clip. A clip showed the hosts seeming to realize they are AI. She spent the morning reading Reddit and Twitter comments and found people saw it followed show notes the user uploaded.
That made her confident people get it, so she addressed it publicly on Twitter instead of pulling the feature back.
Google expands red-team test cases whenever users find a new gap. Large red teams test as many risk areas as they can imagine. A situation not anticipated is added to the test cases.
The team would pull a feature back if it felt clearly unsafe, though the aim is not to need to.