AI that answers from your own book notes: ChatGPT vs NotebookLM vs GleanNote

Ask a general chatbot about a book you just finished and you will usually get a confident answer. Sometimes it is right. Sometimes it describes a chapter that doesn’t exist, blends the book with the author’s next one, or offers a “quote” the author never wrote. The model isn’t being careless; it is answering from patterns in its training data, not from the pages in front of you.

The fix has a name: grounding. A grounded assistant answers from a specific set of sources you provide, and ideally shows you which ones it used. This piece compares three approaches—general chatbots such as ChatGPT, Google’s NotebookLM, and GleanNote’s note-based chat—with one question in mind: where do the answers come from?

Why grounding matters for book questions

When you ask about a book, you usually want one of three things: what the author argued, what you thought about it, or how two ideas connect. Only the first is anywhere in a model’s training data, and even then only for well-known titles, and only as a blurred average of reviews, summaries and discussions.

• Invented quotes. Language models generate plausible sentences, and a plausible sentence in an author’s style is not a quote.
• Popular-take drift. Answers pull toward what most people say about a book, not toward what you found important.
• No memory of your reading. Without your notes, the model can’t know what you underlined, where you disagreed, or which chapter changed your mind.

Grounding changes the contract. The model is told to answer from the sources it was given, so “that isn’t in your notes” becomes a legitimate answer instead of an invitation to improvise.

ChatGPT and other general chatbots

General assistants such as ChatGPT and Gemini are remarkable at explaining concepts, drafting and brainstorming. For a well-known book they can give you a decent overview, and many can now search the web or accept file uploads.

That last part matters. If you paste your notes into the conversation or upload them as a file, the chatbot can answer from them for that session. The friction is in the workflow: your notes have to be exported and pasted each time, a large library doesn’t fit neatly, and nothing stops the model from mixing your notes with what it already “knows” about the title. Without pasted sources, treat any specific quote or page reference as unverified.

NotebookLM: grounded in uploaded sources

Google’s NotebookLM is built around grounding. You create a notebook, add sources—PDFs, Google Docs, web pages, pasted text and other supported formats at the time of writing—and ask questions. Answers come with inline citations that point to the passage they came from, which makes checking them easy.

For documents you can upload, it is excellent: a research paper, a course reader, a public-domain book, a folder of meeting notes. It is less natural for two common reading situations. Paper books have no file to upload, so you would first need to photograph, transcribe and assemble your highlights into a document. And a library of highlights that grows by a few notes every day doesn’t map neatly onto notebooks of uploaded files, which you have to keep current yourself.

Highlight tools with AI features

Highlight managers such as Readwise have added AI features on top of the highlights they already store—summaries, quiz-style prompts and ways to ask questions of your highlights—with specifics that vary by product and change often. Their strength is that the source material is already yours: every highlight is something you chose.

The limit is the one every highlight tool shares: the AI sees what was imported. Ebook and article highlights flow in easily; highlights from paper books need a capture step first.

GleanNote: Talk to your Book and Ask your library

GleanNote builds its AI on top of a library you capture yourself: books, chapters and notes, stored on your device first. Notes come from photographing a page and selecting the lines you want, which on-device OCR turns into text, or from typing a highlight with a comment.

• Talk to your Book (Premium) is a chat about one book that uses only the notes you captured for that book as context. Ask what you noted about a theme, compare two chapters, or have it quiz you before a discussion.
• Ask your library (AI Max) extends the same idea across books, so you can ask how two authors you have read approach the same question.
• Free readers get 3 AI questions to try Talk to your Book before deciding whether an upgrade is worth it.

Because answers are built from your notes, they reflect what you found important, in your own words and comments, rather than the internet’s consensus on the title.

Side by side

Privacy, offline use and honest limits

Privacy is part of grounding. With a general chatbot you decide what to paste; with NotebookLM your sources live in your Google account. GleanNote keeps your library on the device first. When you ask a question, it sends only the notes needed for that question through GleanNote’s server to the AI model. The app never holds an AI key, and there is no model running on your phone. Cloud backup is an optional Premium feature, and it is the only thing that needs a sign-in.

Capture works offline, which suits reading on a train or a flight. The AI features need a connection, so you can capture now and ask questions later.

The limits matter as much as the features:

• The AI is only as good as your notes. A book with three bare quotes will produce thin answers.
• It won’t know chapters you never captured. If a question depends on a part of the book you skipped, the right answer is that it isn’t in your notes.
• It is not a substitute for the book. Grounded chat helps you think with what you kept; it doesn’t read the parts you didn’t.
• Grounding reduces invented answers but doesn’t make any model infallible. Check anything you plan to quote against the page.

Frequently asked questions

Can ChatGPT quote a book accurately?

Not reliably from memory. Give it the exact text—paste the passage or upload your notes—and check any quote against the book before you use it.

Is NotebookLM good for book notes?

It is very good for text you can upload, and its citations make answers easy to check. For paper books, you first need to turn your highlights into a document.

Does GleanNote send my whole library to the AI?

No. Your library stays on your device first, and a question sends only the notes needed to answer it.

Does GleanNote’s AI know the full text of the book?

No. It answers from the notes you captured. That keeps answers tied to your reading, and it means gaps in your notes show up as gaps in the answers.