AI document preparation

Gemini Notebook (NotebookLM): PDF too large or not importing?

Checked October 4, 2026. Official limits can change; follow the message shown in your account.

Short answer (checked October 4, 2026): Gemini Notebook (NotebookLM) takes up to 200MB per uploaded file and up to 500,000 words per source, with up to 50 sources in a free notebook. A PDF can fail on words or source count even when it is small. Source: Google Gemini Notebook help.

A PDF that fails to import does not automatically need stronger compression. The problem can be the file size, the amount of text, a source-count limit, or a document that your reader cannot open correctly. Diagnose the limit before changing the file.

Google's student page says NotebookLM became Gemini Notebook in July 2026. Both names still appear in searches and older guides. This page covers the consumer notebook product, not NotebookLM Enterprise.

Build your preparation checklist

The limits that matter

These are separate constraints. A small text-heavy PDF can still hit the word limit. A large scanned PDF can hit the byte limit with few selectable words. Splitting one file also uses more source slots. Compression changes bytes, not the number of sources, and image compression will not solve a word-count limit.

Choose the smallest change that solves your problem

  1. Open the original in a PDF reader. Check that every page opens and that it is a file you are allowed to use. Record its size, page count, and any error message.
  2. If the file exceeds 200MB, extract the chapter or section you need. If you need the whole book, make ordered chapter files and check each file's size. A page range is not a size guarantee.
  3. If the text exceeds 500,000 words, reduce the scope or split the source. Re-saving images at a lower quality will not remove the text limit.
  4. If the notebook is full, remove sources you no longer need, use another notebook, or merge genuinely related documents if the merged source stays within both per-source limits. Keep a source manifest so citations remain understandable.
  5. If it is within the limits, try a small non-sensitive file to separate a document-specific error from an account or service problem. Do not keep uploading the same failing file without checking the message.

Worked preparation example

Suppose a 240MB textbook has a 30MB chapter on the topic you need. The full file exceeds the published size limit; the extracted chapter does not. That does not prove it will import: it must also satisfy the word limit and your available source slots. Keep the original and name the extract biology-ch03-original-pages-41-68.pdf.

Those numbers are an illustrative scenario, not a test of Google's service. We do not claim to have uploaded your files or benchmarked import success.

Keep the evidence attached to the answer

Create a short manifest: source title, edition or date, original PDF page range, printed page range, output filename. Printed page 1 may come after a cover and a table of contents, so use the PDF thumbnail index when extracting. Include enough surrounding text to preserve the argument and its qualifications.

Check a known phrase in both the original and the output. After import, ask for that passage and follow the citation back to the source. A confident answer is not proof that a chart, footnote, or table was understood.

Private or scanned files

Use a local PDF editor for confidential extracts unless your organization allows the cloud service. Only use material you have the right to upload. If a scan has no selectable text, use our scanned PDF preparation guide. Do not convert a text PDF into page images just to make it smaller.

PageReady does not upload, compress, or split your documents. It explains the route; you use your own PDF software. Split/extract steps · Compression quality checks

Official sources and scope

We checked the documentation, not a paid plan or upload benchmark. When a source count or file-size message differs in your account, record it and follow the current official help for that account.