NotebookLM Audio Overviews of Your Paper: What They Do and Don't Do

You upload your paper to NotebookLM, press Audio Overview and go make a coffee. A few minutes later two friendly AI hosts are talking through your work, and they mostly get it right. It sounds like a real podcast episode about your paper.

Then comes the question every author asks: can I just post this as the audio summary of my paper? Put it on my faculty page, share it on LinkedIn, send it to the people who funded the work? For studying a paper, a NotebookLM audio overview is genuinely good. Publishing one as the author is a different job, and it helps to know where the two part ways.

What NotebookLM Does, Checked Against Google's Own Pages

A note on the name first. In July 2026 Google renamed NotebookLM to Gemini Notebook, calling it "the same standalone product" and saying more than 30 million people use it. The help pages now use the new name; this post sticks with NotebookLM, the name you probably know. Everything below is as Google's help pages describe it at the time of writing (October 2026).

Audio Overviews

Google's Audio Overview help page describes them as "deep-dive discussions between AI hosts" built from the sources you upload. You pick one of four formats:

• Deep Dive (the default): two hosts unpack and connect topics in a lively conversation.
• The Brief: a single speaker delivers the key takeaways in under two minutes.
• The Critique: two hosts give a constructive evaluation of material like an essay.
• The Debate: two hosts argue different perspectives in a formal back and forth.

You can choose from 80+ languages (Google's figure), pick a length from Shorter, Default or Longer (a choice the page labels "English Only"), and add a prompt to focus on particular topics or adjust the expertise level. An interactive mode, currently English only, lets you join the conversation and ask the hosts questions by voice. The page is plain about the limits: "Audio Overviews, including voices, are AI-generated and may contain inaccuracies or audio glitches."

Video Overviews

The Video Overview help page lists three formats: Cinematic (for users 18 and over, English narration only), Explainer (a structured overview, with 80 narration languages listed) and Short (about 60 seconds, with 73 listed). Generation can take "sometimes, more than 30 minutes," and the same warning applies to voices and visuals.

Google's plans page adds that a visible watermark is placed on generated images and videos. Eligible AI Pro and Ultra subscribers can turn it off with a toggle, and it is applied automatically if you live in India, South Korea or Vietnam.

Cost and limits

The standard plan is free: the same plans page says you can "sign up free of charge" with a Gmail account, and paid Google AI plans raise the limits. Usage is compute-based and refreshes every 5 hours until you reach a weekly limit, and Google notes that the limits may change.

Where It Is Genuinely Good

For understanding a paper, NotebookLM is hard to beat on price and speed. It is a good way to get through a stack of papers before a journal club, to hear a dense methods section explained a second way, or to listen to a colleague's preprint on a walk.

It also handles difficult material respectably. In a review of AI research paper podcast tools on Thesislink, Auckland University of Technology's blog for postgraduate researchers, its editor Dr Anaise Irvine ran her own PhD thesis of about 90,000 words through two free tools. NotebookLM produced a 23 minute deep dive in about five minutes, free to download, and in her words the hosts "nailed the central premise." Her verdict: it was "a good enough representation of my thesis to give me confidence sharing it as a point of entry." The other tool she tested, Academia.edu's podcast feature, got her thesis title wrong and garbled its key terms.

Studying a Paper vs Publishing It as the Author

When you listen to an overview of someone else's paper, a small error costs you little: you go back to the PDF. When you publish an overview of your own paper, every listener takes it as your account of your work. Six things change.

Accuracy: you are the reviewer

Irvine's review carries the warning too. The NotebookLM overview she praised also "had a tendency to overstate implications and overgeneralise," though "only slightly." Her thesis is about fictional clones and genetically engineered characters, yet the podcast applied the ideas to real people, to "your" or "our" biology and closed on a prediction that, in her view, was "extrapolating too far from the content of the thesis and creating a false impression." She would not trust AI-generated research podcasts "where the stakes are even moderately high." (We looked at the same tendency in general chatbots in what ChatGPT does and does not do for your paper.)

If you publish one, listen to the whole thing with the PDF open and check every claim. Google's help page describes customizing an overview before you generate it (format, language, length, a prompt); it does not describe editing the generated audio, so a fix means a new prompt and a fresh listen. The Brief (one speaker, under two minutes) is far easier to check than a Deep Dive that ran 23 minutes for Irvine's thesis.

Attribution and a link to the paper of record

A downloaded audio file travels without its context. When you post it, adding the title, the authors and a link to the published version is your job: in the post and on the page. You can also try asking the hosts to name the paper through your prompt, then check that they did.

Label it as AI-generated, too. Google's Generative AI Prohibited Use Policy bars "misrepresenting the provenance of generated content by claiming it was created solely by a human, in order to deceive." Two AI hosts discussing your work should be introduced as exactly that. If the audio could count as part of the published record, check your journal's AI policy as well.

Sharing: what a public link involves

The Audio Overview help page offers three ways to share: a link, the whole notebook, or a downloaded file. Two details matter for authors. First, public notebook sharing "is only enabled for consumer accounts" and is currently disabled for Workspace Enterprise and Education accounts, so if your notebook lives in your university Google account, a public link may not be an option at all.

Second, a link to the audio requires the notebook itself to be shared or made public, with viewers given access to the full notebook. Google's notebook help page notes that viewers "may still be able to access sources and artifacts" even in the focused Chat View. The good news: once the notebook is public, anyone with the link to a shared artifact can view it without signing in to Google. The same page asks you not to "share copyrighted content without the necessary rights to do so," so if your source is the publisher's PDF, check what your publishing agreement allows before you make that notebook public.

Putting it on your faculty or lab page

The download is the most portable option: an audio file you can upload anywhere. But then it is a file on a page, and whether your faculty profile can host and play audio depends on your university's web system, which can mean a ticket to whoever runs the site. For the page itself, see how to make your lab website's publications page worth reading.

Multilingual versions

Here NotebookLM is strong: 80+ languages for Audio Overviews, by Google's count. The catch for an author is review, not reach. Each language is a separate generation, and you can probably check only one or two yourself. Before publishing a Hindi or Portuguese version, find a colleague who reads the language and the field. Note too that the length choice is labeled English only, and interactive mode is English only.

Knowing who listened

Per the notebook help page, notebook analytics show "users per day" (people who made at least one query) and "queries per day" for the past seven days, and only once the notebook is shared with at least four other users and has chat activity within that week. That measures questions asked in the notebook, not plays of the audio. A downloaded file posted elsewhere reports nothing back. For a grant report or a broader-impact section, you need numbers from wherever you host the audio.

NotebookLM vs a Scripted, Reviewed Audio Brief

The NotebookLM column reflects Google's help pages at the time of writing (October 2026). The right column describes the publishing route: a script written from the paper, checked against it, then voiced, whether you do it yourself or use a service.

	NotebookLM Audio Overview	Scripted, reviewed audio brief
Best for	Studying a paper, yours or anyone's	Publishing your own paper to people who may never open the PDF
Cost	Free standard plan; compute-based limits that refresh every 5 hours up to a weekly cap	Your time, or a paid service
Format	Deep Dive, Brief, Critique or Debate; two AI hosts by default	One fixed script you control, voiced by you or a service
Accuracy check	After generation, by you; Google warns it "may contain inaccuracies"	Before release, with the script checked against the paper
Fixing one line	No editing described on the help page: new prompt, regenerate, listen again	Edit the sentence in the script
Languages	80+; length choice labeled English only, interactive mode English only	Each language is a translated script, reviewed before release
Credit and paper link	Yours to add when you post it, or to ask for in your prompt	Written into the script and the page around it
Public sharing	Link needs a shared or public notebook (public sharing is off for Workspace Enterprise and Education accounts), or download the file	Hosted on your own page
Listening data	Notebook analytics count users and queries, not plays	Whatever your page or player records

When NotebookLM Is the Right Choice

Use it, without guilt, when:

• You are studying: a literature review, a journal club, a paper outside your field.
• You want to hear how your own paper lands with an outsider: if the hosts misread your main finding, a newcomer may too, and that shows which sentences need work.
• You are sharing privately with your lab or collaborators (personal Gmail accounts can share a notebook with up to 50 users).
• You want a first explanation to learn from before you write your own.

That matches Irvine's own conclusion: a reputable tool can be useful for sources of "background" level significance, and for anything you would cite, read the paper. Reach for something else when the audio will carry your name to people who will never open the PDF.

Where Amplify Fits

Amplify is built for the publishing side of this. Its audio brief is generated from the PDF you upload and nothing else (no open-web filler), and it carries the title, the authors and a link back to the paper of record.

• Review: each format arrives as three variants you can edit line by line or regenerate. A person reviews every format before release, and nothing publishes until you approve it.
• Languages: English, plus 21 more on request (22 in total), including Spanish, Mandarin, Hindi and Portuguese.
• Placement: an embeddable widget, one script tag per page and one marker per paper, puts the audio on your faculty profile or lab site, where readers can read, listen and ask questions about the paper without leaving the page.
• Evidence: analytics show reads, listens, questions asked, languages, and which format moved people, which is useful for grant reports.

The same order also makes a video abstract, a vertical short and an infographic from the paper. You keep ownership, we do not train on your work, and you can delete a paper and its artifacts at any time. And if all you need is to understand a paper, NotebookLM is free and good at it: keep using it.

AI overviews will keep getting better, and the gap between a study aid and a published summary may narrow. It will not close on its own. When your name is on the audio, someone still has to check it against the paper, credit it, put it where people look and find out who listened.

See what Amplify makes from one paper