Alright, let's cut through the marketing speak. You're looking at NotebookLM and you see a chat interface. You also see this "ask about your sources" button. It feels like two ways to ask questions, so what's the real, practical difference? It's not just a UI choice—it's a fundamental shift in how the tool grounds its answers, and getting this wrong is how you end up with confident, useless hallucinations.
Think of it like this:
* **Regular Chat** is an unbounded, general-purpose AI conversation. It's like opening a fresh ChatGPT window inside NotebookLM. It will use its vast, general knowledge (up to its training cut-off) to answer you. It might *infer* context from your currently open notes, but it's not strictly bound to them. This is where you go for brainstorming, for general questions, or for tasks that need knowledge outside your uploaded sources.
* **"Ask about sources"** is a constrained, citation-locked interrogation of *only the documents you have provided*. When you use this, the AI is explicitly instructed to base its answer *solely* on the content you've uploaded to that notebook. It will then cite the specific source(s) it used for every part of its answer.
Here's a concrete example from a migration plan I was analyzing last week.
I uploaded a 50-page legacy system architecture PDF and a 2-page new cloud platform spec sheet.
* If I go to **Regular Chat** and ask: "What's the best way to migrate a monolithic database to the cloud?"
* I'll get a generic, reasonably good answer about strategies (lift-and-shift, re-platform, refactor) based on the AI's general training. It's decent, but it didn't *need* my documents.
* If I click **"Ask about sources"** and ask: "What are the stated incompatibilities between our legacy system and the new platform?"
* The AI will scour *only* those 52 pages I uploaded. Its answer might look like this:
* "The legacy system uses stored procedures for all business logic (Source: Legacy_Architecture.pdf, p. 32), while the new platform's spec states it only supports user-defined functions and views for security reasons (Source: Cloud_Spec.pdf, p. 1). This is a stated incompatibility requiring a logic migration."
* Every claim is tied to a source. If the answer isn't in your documents, it should say so.
The pitfall I've seen teams fall into is using the regular chat for document-specific questions. They'll ask, "What does the contract say about service level agreements?" and the AI, in regular chat mode, might give a beautifully written, perfectly formatted, and completely fictional SLA clause because it's pulling from its training on what SLAs *usually* look like, not what your document *actually* says.
**When to use which:**
* Use **"Ask about sources"** when you need verifiable, document-specific answers. You're acting like a lawyer or an auditor. You need proof.
* Use **Regular Chat** when you're ideating, drafting, or need to bring in outside concepts. You're acting like a strategist or a writer.
The value—and the risk—is entirely in that distinction. If you're not using "Ask about sources" for analysis of your materials, you're not using NotebookLM; you're just using a chat bot with extra steps.
—BW
Migrate once, test twice.