Google's Gemini Notebook Now Lets You Query Your Purchased Google Play Books
Google's Gemini Notebook can now pull information from books users have purchased on Google Play Books through a new feature called Expert Intelligence, the company announced on August 27, 2026. Google Labs editorial director Steven Johnson demonstrated the feature to The Verge during a briefing. The Verge
Expert Intelligence lets users bring Google Play Books titles into Gemini Notebook and ask questions, generate plans, infographics, and AI podcasts based on the book contents. Compatible titles display a "Gemini Notebook" listing under the "Tools" badge on Google Play Books, giving users a clear signal that a given book supports the feature.
More than 100,000 books from publishers including Penguin Random House, Johns Hopkins University Press, Macmillan Publishers, and O'Reilly Media will support Expert Intelligence at launch. The O'Reilly inclusion is notable for practitioners in software engineering and data science, where the publisher's catalog is deeply embedded in day-to-day reference and learning workflows.
Google also worked with 15 authors, including Michael Pollan, Kim Scott, Steven Pinker, and Jennifer Wallace, to create what it calls "Featured Notebooks." These are customized by the authors themselves, who add supplementary source material and write intro letters for readers navigating their books through Gemini Notebook.
The feature arrives with access controls designed to address what has been a persistent concern with AI-assisted reading and research tools: the tension between useful extraction and copyright protection. Gemini Notebook includes protections preventing users who do not own a purchased book from accessing its full text or related information when a Notebook is shared. Non-owners see the source listed but are directed to Google Play Books to purchase it. This mirrors the source-grounding architecture that has been part of the product since its original 2023 launch as NotebookLM. Source-grounding means the AI is constrained to reason over material the user has provided rather than drawing on its own training data, and responses come with citations showing relevant original quotes to make fact-checking straightforward. Google noted at the time that while source-grounding appeared to reduce hallucination risk — the tendency of language models to produce confident but incorrect information — fact-checking against original material remained essential. Google Blog, 2023
The product now called Gemini Notebook has accumulated a substantial feature set since its introduction. Originally launched in July 2023 as NotebookLM, a virtual research assistant that grounded language models in a user's own content, the product could initially work with Google Docs, with Google promising additional formats to come. By September 2024, the Audio Overview feature could turn documents, slides, and charts into generated podcast-style discussions. Video Overviews were upgraded in October 2025 with visuals powered by the Nano Banana model and a new "Brief" format for quick summaries. In December 2025, slide deck generation arrived via the Nano Banana Pro model, converting research notes into presentations. In June 2026, Google announced new Gemini models and code execution capabilities. The rebrand to Gemini Notebook came in July 2026, with all existing notebooks remaining fully accessible. Google Blog, 2026
Expert Intelligence fits into a familiar trajectory for this product: start with a constrained set of source types, prove the grounding mechanism, then expand. The original 2023 launch supported Google Docs. Three years later, the source surface includes purchased books from major publishers, and Google has stated plans to expand Expert Intelligence to scholarly articles, magazine articles, and newspapers. Google also plans to bring Expert Intelligence to AI Mode in Search and the Gemini app, which would take the feature well beyond the standalone Notebook interface.
The architecture here is worth attention. Source-grounding, as Google has defined it from the outset, constrains the model to reason over user-provided material rather than generating from parametric knowledge — the information baked into a model during training — alone. Expert Intelligence extends that principle to licensed, purchased book content. The access-control layer, which gates full-text retrieval behind verified ownership, is the mechanism that makes this palatable to publishers who have otherwise been hostile to AI training on their catalogs. Whether the 100,000-title launch catalog and the publisher list that includes Penguin Random House and Macmillan signals broader industry acceptance of this model, or selective participation by a subset of willing partners, is a question the launch itself does not answer. The expansion to newspapers and scholarly articles, if it arrives, will test whether the same ownership-gated approach transfers to content types where purchase and ownership models differ fundamentally from trade books.
For technology professionals who already use Gemini Notebook as a research tool, Expert Intelligence adds a source category — purchased nonfiction and technical titles — that was previously inaccessible to the grounding pipeline. The ability to query across multiple books, cross-reference their claims, and generate structured outputs like infographics and podcasts from their combined content is a straightforward extension of what the product already does with uploaded documents and notes. The publisher participation and the access controls are the parts that required work.
