You add a PDF to a NotebookLM notebook, and it comes in empty - no text, no summary, nothing to ask about, sometimes with a flat "source is empty". Or it imports but every question returns "I don't have information about that." If NotebookLM can't read your scanned PDF, the cause is almost always the same one thing, and it's fixable.
NotebookLM reads the text layer of a document. A scanned or photographed PDF is a picture of text - there's no text layer underneath - so there's nothing for it to index. This guide explains what's happening and gives you two clean ways to fix it.
Why NotebookLM chokes on scanned PDFs
A PDF can contain either real, selectable text or just an image of a page. Documents that started digital - exported from Word, Google Docs, or a design tool - have a proper text layer. Documents that were scanned, photographed, faxed, or signed with an image-based tool often don't. They look identical on screen, but only one of them is machine-readable.
NotebookLM's source ingestion expects that text layer. When it isn't there, the source imports as blank or near-blank, and the model has nothing to ground its answers in.
How to tell in five seconds: open the PDF and try to drag-select a sentence. If the text highlights, NotebookLM should read it fine. If nothing highlights, it's an image-only file - that's your culprit.
Fix 1 - OCR the PDF, then re-upload
OCR (optical character recognition) reads the picture of text and reconstructs a real text layer. Run your scanned PDF through an OCR step, save the output, and upload that copy to your notebook. NotebookLM will then have actual text to summarise and cite.
The catch is that this is an extra tool and an extra step every time, OCR quality on messy scans varies, and you end up with a second copy of every file to manage. Our scanned-PDF OCR guide covers how to do this well.
Fix 2 - Read the scanned PDF in the browser instead
If you keep running into image-only files, it's often faster to skip the upload-and-convert loop entirely and read the PDF where it already is. Sidenote is a browser reading assistant (Chrome, Edge and Firefox) that OCRs scanned and photographed PDFs automatically, then lets you summarise and question them - no separate OCR tool, no re-upload.
Two things make it a natural fit for the exact problem NotebookLM has here:
- OCR is built in. Open an image-only PDF and Sidenote runs it through OCR before answering, so a document with no text layer becomes summarisable and citable.
- Every answer is verifiable. Each claim carries a scroll-to-source citation: click it and the PDF scrolls to the exact sentence and highlights it. Claims that can't be matched to a real passage are dropped server-side, so a bad OCR read shows up as "not found" rather than a confident guess.
NotebookLM vs an in-browser reader for scanned files
| NotebookLM | In-browser reader (Sidenote) | |
|---|---|---|
| Scanned / image-only PDFs | Imports blank - OCR yourself first | OCRs automatically on open |
| Workflow | Upload to a notebook | Read the file in place |
| Verify a claim | Source-level citation | Click to the exact highlighted sentence |
| Audio overview | Yes | No |
| Best for | Synthesising a set of uploaded sources | Reading and citing whatever you have open |
NotebookLM is genuinely good at what it's for - pulling a set of sources into one place and synthesising across them, with its standout audio overviews. It just needs readable text to start with. For scanned files specifically, a reader that OCRs on open removes the friction. If you're weighing the two more broadly, see the full Sidenote vs NotebookLM comparison and our roundup of NotebookLM alternatives.