NotebookBloom: Why NotebookLM can find the passage but not the PDF page number
I spent more time than I expected trying to answer a simple question: if NotebookLM can point to the exact supporting passage, why can’t it give me a reliable page number?
The short answer is that a passage location and a PDF page are not the same thing.
NotebookLM appears to flatten a source into a text stream. Its citation can jump to a span of that text, which is enough to highlight the supporting words. But the data available to the page doesn’t consistently include the original PDF page boundary.
That matters because it would be easy for an extension to make the output look more academic by guessing. It could search the PDF, find a similar sentence, and print “p. 12.” The problem is that PDF extraction is messy: headers repeat, columns get reordered, OCR changes words, and NotebookLM may normalize the passage. A confident-looking page number can still be wrong.
For NotebookBloom, I chose the less impressive but safer behavior. It keeps the source title and the quoted passage so you can trace the claim back to the source. It does not invent a page number the underlying data cannot support.
If you are citing a PDF in a paper, the responsible workflow is still:
- Use the NotebookLM citation to find the supporting passage.
- Open the original PDF.
- Verify the wording and page number there.
- Build the bibliography entry from checked metadata.
It adds a step, but it is better than turning uncertainty into a precise-looking mistake.
NotebookBloom: https://notebookbloom.com
Top comments (0)