Independent editorial review.::Pricing verified directly from each vendor, April 2026.::Affiliate disclosure
Memo · Reference shelfNotebookLM, an honest reviewFiled April 2026

Tool review · The free option that wins most of the time

NotebookLM, an honest review

What it does brilliantly. What it misses. Who should use it.

Free tier limits and pricing checked directly from notebooklm.google.com, April 2026.

In short →

NotebookLM is the best free document-summarization experience available to individual consumers in 2026. Its free tier (100 notebooks, 50 sources per notebook, 50 chat queries per day, 3 audio and 3 video overviews per day) is genuinely generous. For anyone with a Google account who needs to research, synthesise, or explore documents, start here before paying for anything else.

Where it falls short: no offline use, no real-time meeting capture, no BAA for HIPAA-sensitive content, and per-source length limits on very long single documents.

§ I.Free tier, exact limits

FeatureFree tierGoogle One AI Premium ($19.99/mo)
Notebooks100Higher limits
Sources per notebook50Higher limits
Chat queries per day50Higher limits
Audio overviews (Deep Dives)3 per dayHigher limits
Video overviews3 per dayHigher limits
PPTX slides export (2026 feature)YesYes
Data tables (2026 feature)YesYes
Interactive audio JoinYes (limited)Yes (more sessions)
HIPAA BAANoNo (enterprise Workspace only)
API accessNoNo

§ II.What's new in 2026

February 2026

PPTX slides export

NotebookLM can now generate a structured slide presentation from your notebook sources and export it as a PowerPoint file. Particularly useful for converting research synthesis into presentation-ready content.

February 2026

Data tables

Structured comparison tables generated from sources. Ask NotebookLM to compare N items across specified dimensions and it generates a formatted table rather than a prose list.

January 2026

Interactive audio Join

During an Audio Overview (Deep Dive), you can interrupt mid-sentence with follow-up questions and the AI voices respond in character.

§ III.What it does brilliantly

Source attribution in every response

Every response links back to the exact source passage that supports it. You can click through to verify the claim in the original document. This is what makes it research-grade rather than just AI-assisted reading.

Multi-document synthesis

Add 10 to 50 papers, reports, or articles as sources and ask questions across all of them simultaneously. 'What do these papers disagree on?' produces genuinely useful cross-document synthesis that you would otherwise need hours to produce manually.

Audio overview quality

The Deep Dive audio overviews (two AI voices discussing your documents conversationally for 10 to 20 minutes) are genuinely excellent for absorbing material while doing other tasks.

Maintaining context across sessions

Notebooks persist. Unlike ChatGPT or Claude conversations that reset, your NotebookLM notebook retains all sources and prior Q&A. Return to a research project weeks later and the context is still there.

§ IV.Where it falls short

!

No offline access

Requires internet and an active Google session. Not usable on flights, in areas with unreliable connectivity, or on devices without Google account access.

!

No real-time meeting capture

NotebookLM is designed for uploaded documents, not live audio streams. It cannot join a Zoom call, listen to a podcast in real time, or process a live lecture.

!

Consumer BAA unavailability for HIPAA

The consumer and Google One AI Premium tiers do not come with a Business Associate Agreement. Do not upload protected health information without a verified enterprise arrangement.

!

Very long single-document limits

Per-source length limits mean a single 500-page document may be truncated or chunked automatically. For very long single documents, Claude.ai Pro with its 1M-token context window is a better choice.

§ V.Excerpt cards

47-page paper, NotebookLM summary

"The paper argues that scaling laws for LLMs are not smooth: capability appears to jump at certain parameter thresholds. The authors note this is contested; jumps may reflect measurement choices, not real discontinuities."

[Source: Section 3.2, page 14]

5-source literature review, NotebookLM summary

"Papers 1, 3, and 5 agree that few-shot prompting improves accuracy. Papers 2 and 4 note the improvement is model-size dependent: small models see little benefit. No paper addresses non-English languages systematically."

[Sources: Paper 1 p.4, Paper 3 p.8, Paper 5 Abstract]

§ VI.Common questions

Q.01Is NotebookLM free?

Yes. NotebookLM's base tier as of April 2026 includes 100 notebooks per account, 50 sources per notebook, 50 chat queries per day, 3 audio overviews (Deep Dives) per day, and 3 video overviews per day. This is genuinely generous and covers most individual use cases without any payment. Google One AI Premium and Google Workspace add-ons provide higher limits and additional features.

Q.02Is NotebookLM HIPAA compliant?

NotebookLM's consumer tier (free and Google One AI Premium) does not include a Business Associate Agreement (BAA) and is not appropriate for documents containing protected health information (PHI). Google Workspace's NotebookLM Plus for Enterprise may include BAA options depending on your Workspace agreement; verify directly with your Google Workspace admin. Do not upload patient data, clinical notes, or any PHI to the consumer version of NotebookLM.

Q.03Is NotebookLM better than ChatGPT for summarizing documents?

For multi-document research synthesis, NotebookLM is better than ChatGPT in several important ways: it maintains source attribution (every response links back to the specific source passage), it handles multiple documents simultaneously without you needing to paste them all in each conversation, and its audio overview feature produces genuinely useful podcast-style summaries for auditory learners. ChatGPT is better for documents requiring complex reasoning, non-English primary source material, and tasks where you want to combine document analysis with general knowledge.

Q.04Are my documents used to train Google's AI models?

Per Google's stated terms as of April 2026: content you upload to NotebookLM is not used to train Google's AI models. Your notebooks are private by default and not visible to other users. Google processes your content to provide the NotebookLM service, including the AI responses, but states that this processing is not for model training. For enterprise use, Google Workspace's data processing addendum provides contractual protections. Always verify current terms at notebooklm.google.com before uploading sensitive documents.

Q.05What are NotebookLM's limitations?

NotebookLM's main limitations as of April 2026: no offline access (requires internet and Google account); no real-time meeting capture; the 50-source limit per notebook (sufficient for most research projects but constraining for large repositories); per-source length limits (very long single documents may be chunked or truncated); consumer tier has no BAA for HIPAA compliance; no API access for programmatic use; non-English language support less tested than English.

Cross-references