Before we explain what this tool is, listen to the clip below.
Two voices discussing this article. They interrupt each other, pick up each other’s points, and land a couple of jokes. Neither of them is a person, and no part of it was scripted or recorded.
The Google NotebookLM AI Generated Audio Overview:
That came out of a document. One click, no voice actors, no editing pass.
We have been using the tool for a few weeks. Here is what it does, where it is genuinely useful, and where it runs out of road.
What NotebookLM Actually Is
It started as an experimental research assistant, which undersells it. You hand the app a set of sources. It reads them. Then you ask questions and it answers using only what you uploaded, with citations pointing back to the exact passage.
That last detail matters more than it sounds. Most chatbots will happily invent a source and describe it with total confidence. This one is fenced in by the material you gave it, which is the difference between a tool you can cite and a tool you have to double-check.
It runs on Gemini 1.5, which is where the image handling comes from.
The Features That Matter
Audio Overview
This is the one people will remember. Upload your sources, click generate, and roughly ten minutes of conversation between two synthetic hosts comes back.
They do not read your document aloud. They talk about it. They pick up threads from each other, disagree now and then, and the timing is close enough to real that you stop noticing after a minute. You can download the file and take it with you.
Worth knowing before you rely on it: these discussions are a briefing, not a thorough analysis. They skim. Treat the output as an orientation to your material rather than a replacement for reading it.
Source Handling
It accepts Google Docs, Slides, PDFs, pasted text, and web links. Because of the model underneath, it also reads images inside those files. Include a chart and the summary can reference the chart, which closes a gap most document tools do not even acknowledge.
Generated Documents
Ask for a study guide, a timeline, an FAQ, or a briefing document and you get one. A student can upload a semester of course material and have a revision guide back before the kettle boils. The same mechanism turns a stack of quarterly reports into something a meeting can actually use.
Living With It
The interface is clean, which is not always true of Google products at launch. Create a notebook, upload your sources, and it summarizes them immediately and pulls out the main topics. Those topics become suggested prompts. Upload something about collaborative tools and an offer to discuss collaborative tools appears, one click away.
The chat is the part we use most. Ask a direct question, get an answer with citations attached. It turns a folder of documents into something you can interrogate instead of something you have to wade through.
Where It Earns Its Place
Answers come from your sources. Not a general summary of the subject, not something scraped off the web. It works from the specific files you provided, which is the difference between a study aid and a search engine.
Research gets faster. Rather than reading nine reports to find the one paragraph that matters, you ask. It also picks up connections between separate documents, which is usually work that only happens on a second pass nobody has time for.
Charts count. Text and images both feed the analysis, so a slide deck full of diagrams does not arrive as a summary full of blanks.
Your documents are not training data. Google states that material uploaded here is not used to train models and is not reviewed by human moderators. If you handle client work or unpublished research, that single policy decides whether the tool is usable at all.
How to Generate an Audio Overview
- Go to NotebookLM and sign in with your Google account.
- Click Create New Notebook and name it.
- Add your sources. Docs, PDFs, Slides, web links, or pasted text. You can add several.
- Open the Notebook guide at the lower right and click Generate, then choose Audio Overview.
- Wait. Larger notebooks take a few minutes depending on how many sources you loaded.
- Listen. The file plays automatically once it is ready.
- Download it if you want it offline.
That is the whole process. The gap between uploading a PDF and hearing two people discuss it is measured in minutes.
Where It Falls Short
Honest pass, because the enthusiasm above needs a counterweight.
English only. The audio discussions do not support other languages yet, which rules the feature out for a large part of the world.
It is slow on big notebooks. Several minutes of processing, and the discussions occasionally state something the source does not support. Check anything you plan to repeat.
You get almost no control. Length, tone, and depth are decided for you. You cannot steer the conversation or interrupt it, and there is no way to ask for a shorter version or a more technical one.
It stays shallow. The output is pitched at a general audience. If you want a specialist reading of a technical paper, this will not get you there.
The Early Reaction
The response has been loud and mostly positive, and the audio quality is doing most of that work. Plenty of people have called it the most convincing text-to-speech they have heard, and the back-and-forth format is a large part of why.
The complaints are consistent though, and they match ours. Not enough control over the output, and a ceiling on how deep the discussions go.
Worth Your Time?
Yes, with the limits above understood.
The citation-grounded chat alone makes it useful for anyone working through a pile of documents, and the audio feature is the rare demo that survives contact with actual use. It is a briefing tool rather than a research partner, and once you treat it that way it stops disappointing you.
Multilingual audio and real control over the generated output are the two additions that would change the picture. Until then, this is a genuinely good way to get oriented in material you have not read yet.
For more of the latest AI inspired news, events and products, check out our AI Blog.
About the Author: Gavin Stacey
