Gemini Notebook

Gemini Notebook is Google Labs' retrieval-augmented study and research assistant, known as NotebookLM until the July 2026 rename. A notebook is a set of sources the reader chooses — PDFs, documents, slides, websites, pasted text, or the transcripts of YouTube videos — and every answer the assistant gives is drawn from that material with citations back to it. Where a general-purpose chat assistant answers from its training data, Gemini Notebook answers from the reader's own shelf of documents, and that constraint is the product.

What you can do with it

Ask questions across the whole source set, summarize long documents, and follow citations to check an answer against the passage it came from. Beyond chat, Notebook produces study and presentation artifacts: Audio Overviews, a podcast-style conversation between two synthetic hosts that can be interrupted with questions; Video Overviews, including short vertical versions added in 2026; Mind Maps, Reports, Flashcards, Quizzes, Data Tables, Infographics and Slide Decks. A July 2026 update gave each notebook a secure cloud computer for running code, which extended the tool from summarizing sources toward analyzing them.

Who it is for

Self-directed learners revising from their own course materials, instructors assembling readings into a questionable digest, and researchers working through a literature set. The free tier is open to anyone with a personal Google account and is where the audience for classroom use sits; higher limits, longer sources and sharing come with the paid Workspace and Google One AI tiers.

Notes and caveats

Google states that personal-account content is not used to train its models unless the user submits feedback, that Workspace uploads, queries and responses are not used for training even when feedback is given, and that data is not shared with third parties; deleted notebooks, however, cannot be recovered. The capability claims above are the vendor's own, and a district or institution should confirm them against its data processing agreement rather than a product page. Coverage in 2026 also described a lawsuit over the voice used in generated podcasts, which Google denied, a reminder that audio outputs raise likeness questions separate from the source-grounding safeguards. Treat generated flashcards, quizzes and summaries as drafts to check against sources, and note that the free tier's limits change without notice.

Connected Concepts

RAG (Retrieval-Augmented Generation), AI Literacy, Self-Directed Learning, Retrieval, Spacing and Interleaving, Generative AI, Multimodal AI, Video in Education, Hallucination Risk