NotebookLM Limits 2026: Sources, Notebooks, File Size and Plans

NotebookLM limits by plan

You can add 50 sources to a notebook on the free tier, and you can create 100 notebooks — so 5,000 source slots in total, at no cost. Paid Google AI plans raise both numbers: 100 sources on AI Plus, 300 on AI Pro, 500 or 600 on the two AI Ultra tiers. Every limit below is per user, and Google labels the whole table “subject to change”.

	Free	AI Plus	AI Pro	AI Ultra (20TB)	AI Ultra (30TB)
Sources per notebook	50	100	300	500	600
Notebooks	100	200	500	500	500
Total source slots	5,000	20,000	150,000	250,000	300,000
Words per source	500,000	500,000	500,000	500,000	500,000
Upload size per source	200MB	200MB	200MB	200MB	200MB
Price per month	$0	$4.99	$19.99	$99.99	$199.99
Early access to features	Standard	Early	Priority	Priority	Priority

Limits verified 20 October 2026 against Google’s Gemini Notebook plans page and its sources documentation. Prices verified 18 October 2026 against Google’s subscriptions page and tracked on the AI pricing tracker.

The arithmetic that decides the tier. Per extra source over the free tier’s 50, AI Pro is the cheapest capacity Google sells: AI Plus costs $0.10 per extra source per month, AI Pro $0.08, AI Ultra (20TB) $0.22 and AI Ultra (30TB) $0.36. Per 1,000 total source slots the gap widens — $0.25 on Plus, $0.13 on Pro, $0.40 and $0.67 on Ultra. If sources are your constraint, Pro is the answer: the Ultra tiers are priced for Flow credits, storage and model access, not notebook capacity.

One caveat worth knowing before you buy: Google’s own two limit tables disagree. The consumer plans page lists 500 sources per notebook for AI Ultra (20TB), while the Workspace and Education limits breakdown lists 400 for the equivalent “Expanded Level” access. Budget for 400 if you are buying through Workspace.

Daily limits moved — and mostly disappeared

Every guide to NotebookLM quotes “3 Audio Overviews per day” for the free tier. That number is no longer on any consumer plan page. Google moved personal plans to compute-based usage limits that “refresh every 5 hours until you reach your weekly limit”, and now describes capacity only in relative terms: AI Plus is twice the standard allowance, AI Pro four times, AI Ultra five or twenty times AI Pro. No absolute daily counts are published for personal plans.

The absolute per-day numbers survive in exactly one place — the Workspace and Education breakdown — and they are worth reading even if you are a personal subscriber, because they are the only published figures for what an access level actually yields:

Per day	Standard access	More access	Higher level	Expanded	Highest
Chats	50	200	500	1,000	5,000
Audio Overviews	3	6	20	40	200
Video Overviews	3	6	20	40	200
Reports, quizzes, flashcards, mind maps	10 each	20 each	100 each	200 each	1,000 each
Deep Research	10/month	3	20	30	200

Two readings matter. The free tier’s notebook and source numbers match “Standard access” exactly, which is the best evidence that 3 Audio Overviews and 50 chats a day is still roughly what a free account gets — but that is an inference from a Workspace table, not a published consumer limit. And if your organisation is on Workspace Business Standard, Business Plus, Enterprise Standard or Enterprise Plus you already sit at “Higher level” — 20 Audio Overviews and 500 chats a day, 300 sources per notebook — so a personal AI Pro subscription on top adds nothing. Education Plus with the Teaching and Learning add-on lands one rung lower.

How big a single source can be

The per-source ceiling is 500,000 words or 200MB for an uploaded file, and it does not vary by plan — a free account and a $199.99 Ultra account get the same allowance. Since 500,000 words is roughly five long non-fiction books, the limit you hit is almost always the source count, not the source size.

Supported source types: Google Docs, Google Slides (up to 100 slides), Google Sheets (capped at 100,000 tokens), .docx, .pptx, PDF, plain text, Markdown, CSV, ePub, Play Books ebooks (text only), web URLs, YouTube URLs for public videos, pasted text, images, audio files, and your own Gemini chats.

Two gaps to plan around. Excel .xlsx is not on Google’s list — native Google Sheets is supported and CSV is supported, so spreadsheet work means converting first. And YouTube has no video-length limit as such; the constraint is the caption file, which must come in under the same 500,000 words.

The pricing landscape

NotebookLM is not sold on its own. The free tier needs only a Google account, and every paid tier arrives as part of a Google AI subscription:

• Free — $0. 100 notebooks, 50 sources each. Genuinely useful rather than a crippled trial, and enough for most students and most one-off research projects.
• Google AI Plus — $4.99/month. 200 notebooks, 100 sources each, roughly twice the free compute allowance, 400GB of storage. The tier most NotebookLM write-ups omit entirely, and the cheapest way past the 50-source wall.
• Google AI Pro — $19.99/month. 500 notebooks, 300 sources each, four times the free allowance, 5TB of storage, higher model access and watermark removal on generated output. The value pick if you are hitting limits at all.
• Google AI Ultra — $99.99 (5×) or $199.99 (20×)/month. 500 notebooks, 500 or 600 sources, the highest compute multiples, 20–30TB of storage.
• Workspace and Education. Access level follows your edition rather than a separate purchase — see the table above.
• Enterprise. Via Google Cloud, described only as “5× or more” notebook artefacts. Custom pricing, SSO, admin controls.

Prices verified 18 October 2026 against Google’s subscriptions page. A visible watermark is applied automatically to generated output for residents of India, South Korea and Vietnam regardless of tier.

Recommendation: free for almost everyone; AI Plus at $4.99 if 50 sources is the only pinch; AI Pro at $19.99 for large notebooks or daily artefact generation. Buy Ultra for Gemini, Flow credits or storage — notebook capacity is the worst-value part of that bundle.

What NotebookLM is in 2026

NotebookLM is a source-grounded AI research assistant built by Google and powered by Gemini models. You create a notebook, add sources, then chat with them or generate artefacts from them. The differentiator is that every answer cites the specific passage in your source material that supports it: click the citation, land on the exact sentence.

That framing is the whole product. ChatGPT and Claude answer from general knowledge plus whatever context you paste; NotebookLM answers only from your sources and declines to invent. For research work, this is categorically better.

A naming note: Google renamed the product Gemini Notebook on 16 July 2026, calling it “the same standalone product”. notebooklm.google now redirects to notebook.google and the help centre has moved to support.google.com/gemininotebook/; Google says automatic redirects keep existing shared notebooks and user links working. Neither announcement mentioned pricing, and the limits above are the ones published since.

From your sources the Studio panel generates Audio Overviews (two AI hosts discussing your material in a 10–20 minute podcast, interruptible mid-episode), Video Overviews, Mind Maps, Slide Decks exportable to PPTX, Reports, Infographics and Data Tables, and Quizzes and Flashcards. Google never states which Gemini model it runs on — the plan pages say only “access”, “higher access” and “highest access”.

What NotebookLM genuinely does best

Source-grounded answers with citations. The foundational value, still done better here than anywhere else — for research where accuracy matters, verifiable citations change the nature of the tool.

Audio Overviews. Not a gimmick. Upload a paper, generate the overview, listen on a walk — the conversational format makes dense material digestible in a way the abstract does not.

Synthesis across many sources. Fifty PDFs, one question, one combined answer citing specific sections of specific papers. No general chatbot handles this much context this cleanly.

Study material generation. Flashcards and quizzes from your own reading turn passive material into active recall in one click — the reason it dominates recommendations for students and for teachers.

Privacy. Google states that uploaded content is not used to train models.

Where NotebookLM falls short

Not a writing tool. It resists going beyond your sources. For drafting or original creative work you want Claude or ChatGPT.

The source wall is real on free. 50 per notebook sounds generous until a literature review hits 60 papers, and splitting across notebooks loses the cross-source synthesis that is the point.

Consumer limits are now unpublished. Compute-based quotas refreshing every five hours are harder to plan around than “3 per day”, and Google gives no absolute figure for personal plans.

No Excel support. Google Sheets and CSV only, with Sheets capped at 100,000 tokens.

English first, and confusing in education. Other languages work but quality varies, and which access level an institution holds depends on an edition mapping most administrators have never read.

NotebookLM vs. the alternatives

For source-grounded research with citations: NotebookLM > Perplexity > Claude with Projects > ChatGPT. NotebookLM is purpose-built; the others are general tools doing a decent impression.

For academic literature search: Elicit and Consensus search the literature; NotebookLM works only on sources you supply. Complements, not competitors.

For audio from documents: NotebookLM, with no real competition.

For many sources at once: NotebookLM wins outright — nothing else holds 300 cited sources in one workspace.

For category context, see best AI productivity tools, best AI education tools and best AI data analysis tools.

Who should use NotebookLM

• Students — the free tier covers entire courses
• Researchers and academics — multi-source synthesis is the core need
• Analysts and consultants synthesising reports, papers and transcripts
• Journalists working through large document sets
• Teachers building study material from course content
• Anyone needing cited answers about specific documents rather than the internet

Who shouldn’t use NotebookLM

• Creative writers — wrong tool for generating new work
• Casual users without research needs — ChatGPT fits better
• Spreadsheet-heavy workflows — no Excel, and Sheets is token-capped
• Non-English-heavy research — quality varies

The verdict

NotebookLM is worth it, and for most people it is worth exactly $0. The free tier’s 100 notebooks and 50 sources each cover entire courses and most research projects, and the per-source ceiling of 500,000 words is the same one Ultra subscribers get.

Upgrade only against a specific limit. If that limit is sources, Google AI Pro at $19.99 is the cheapest capacity Google sells — $0.08 per extra source per month against $0.22 on Ultra — with AI Plus at $4.99 a sensible half-step. If the limit is generation volume, the Workspace tables suggest Pro-level access yields about 20 Audio Overviews a day against roughly 3 on free. And if your employer pays for Workspace Business Standard or better, you sit at that level already and a personal subscription buys nothing.

Before paying anything, open your largest notebook and count the sources. Most people who think they need a paid tier are at eleven.

What changed

• 20 Oct 2026: Limits re-verified against Google’s help centre — free tier confirmed at 100 notebooks and 50 sources, AI Pro at 500 and 300; per-source ceiling 500,000 words or 200MB on every tier.
• 20 Oct 2026: Consumer plans dropped published per-day caps for compute-based quotas that refresh every 5 hours against a weekly limit; absolute daily figures now survive only in the Workspace and Education breakdown.
• 20 Oct 2026: Google AI Ultra corrected to $99.99 (5×) and $199.99 (20×) from a stale $249.99; Google AI Plus at $4.99 added as the cheapest route past 50 sources.
• 20 Oct 2026: Google Sheets confirmed supported with a 100,000-token cap and CSV supported, correcting an earlier claim that spreadsheets were unsupported outright.
• 16 Jul 2026: Google renamed NotebookLM to Gemini Notebook; notebooklm.google redirects to notebook.google and automatic redirects keep existing shared notebook links working.