NotebookLM (Google) is free, slick, and even produces podcast-style audio overviews. It's an excellent tool for casually exploring a handful of documents. But once it comes to a real scientific library, when confidentiality matters, or when you have more than a few dozen PDFs, its limits show up fast. RefChat is built for exactly those cases.
Side-by-side
| Criterion | RefChat | Google NotebookLM |
|---|---|---|
| Confidentiality | 100% local option No data ever leaves your machine |
Google cloud only All PDFs transit through Google's servers |
| Hosting (cloud mode) | OVHcloud, Roubaix (France) — GDPR | Google datacenters (mainly US) |
| Source volume | Libraries of 5,000 to 20,000+ PDFs handled without difficulty | Limited (around 50 sources per notebook) |
| Sourced citations | Explicit citation of the exact passage + direct link to the PDF | Citations but coarser chunk granularity |
| Hybrid search | BM25 + dense (E5-large) + cross-encoder reranking | Dense RAG (Gemini), no explicit BM25 or reranker |
| OCR for scanned PDFs | Built-in EasyOCR | Partial — variable quality on old scans |
| Scientific parsing (GROBID) | Yes — extracts title, authors, references, sections | No — generic PDF parsing |
| Zotero / Mendeley integration | Native — point at your storage/ folder |
Manual — upload PDFs one by one (or limited batch drag-drop) |
| Multilingual | Native EN/FR/DE/ES…, multilingual E5 embeddings | Decent (Gemini), but answers often biased toward English |
| Query modes | 5 auto-detected modes (question, summary, references, author, fact-check) | Free chat + summary + outline + generated FAQ + audio overview |
| Audio overview / podcast | No | Yes — NotebookLM's flagship feature |
| Pricing | €10/month (Solo Local) to quote (Team) | Free (with quotas) / paid via Google One AI Premium |
| GDPR compliance | Documented compliance | Depends on account and region — unclear for EU pro use |
| Training on your content | Never — contractual clause | Variable Google policy depending on plan and region |
| Account required | RefChat licence (no tracking) | Google account mandatory |
When NotebookLM is the right choice
Let's be honest: NotebookLM is an excellent tool for certain use cases. It is probably the right call if:
- You want to quickly try out a dozen articles without installing anything.
- Your documents are public and confidentiality is not a concern.
- You love the podcast/audio overview format to digest a topic.
- You mostly work in English on small ad-hoc volumes.
- You don't need Zotero integration or to manage tens of thousands of PDFs.
When RefChat is the right choice
RefChat is built for advanced users and professional organisations. It stands out if:
- Your PDFs are confidential (internal reports, patents, legal files, medical data).
- You manage a real scientific library (1,000 to 20,000+ articles).
- You use Zotero or Mendeley and want it to "just work".
- You work with old scanned PDFs (expert reports, industrial archives).
- You need documented GDPR compliance for an EU professional context.
- You want citations precise at the passage level, not just per article.
- You work across multiple European languages and want an EU-native tool.
In short
NotebookLM = excellent for ad-hoc, public, exploratory use. A well-crafted consumer product.
RefChat = a professional tool for researchers, engineers and experts who own a real bibliography and have confidentiality or volume constraints. The European sovereign alternative when NotebookLM is no longer enough.
Typical users migrating from NotebookLM to RefChat
- The PhD student who hit NotebookLM's source cap halfway through their thesis and needs to index 3,000 articles at once.
- The consultancy whose client asked for a written guarantee that reports will not end up on Google's servers.
- The R&D team that wants to give access to 30 years of internal archives without exposing a single PDF to a third-party service.
- The Zotero researcher tired of manually re-uploading their PDFs to every new notebook.
Try RefChat on your own bibliography
30-minute live demo + trial on a sample of your PDFs, under NDA if needed. No strings attached.
Request a demo