Skip to main content
Back

Google's NotebookLM Finally Shows Which Sources Built Your AI Output

NotebookLM's new Source Attribution feature exposes the exact prompt and sources behind every Studio artifact, and lets you tweak and re-run them.

  • Source Attribution launched: NotebookLM now shows the exact prompt and sources used to generate every Studio artifact via a three-dot menu.
  • Iterate button: Users can edit the prompt or source selection and regenerate; each iteration saves as a separate artifact, preserving the original.
  • Free for all users: The feature is part of the core Studio experience, not locked behind paid plans; rolling out gradually across accounts.
  • Fills a key gap: Inline citations already traced individual claims; Source Attribution now exposes the full generative recipe behind entire artifacts like Audio Overviews and reports.
  • Professional use-case unlock: Compliance, legal, and research workflows can now audit and reproduce exactly how a document was generated.
  • Limitations remain: Some artifact types (Video Overviews, Infographics) still cannot be edited in-place; mobile rollout may lag the web app. Try it at notebooklm.google.com.

NotebookLM has always been Google's most source-honest AI tool. Every answer in chat comes with inline citations that jump you to the exact passage in your uploaded documents. But there was always a blind spot: the Studio. Generate an Audio Overview, a study guide, or a briefing doc, and you had no idea which prompt drove it or which of your forty uploaded sources actually made it into the synthesis. If the output was off, your only move was to regenerate and hope. That changes now.

The black box cracks open

Google has rolled out two new features for NotebookLM: "View prompt and sources" and "Iterate." Together, they let you review the exact prompts and sources used to generate any piece of Studio content, adjust them, and regenerate the output. The announcement landed with a characteristically blunt tagline: "No more guessing."

To access it, click the three-dot menu on any generated artifact and select "View prompt and sources." A window appears showing the original prompt and the specific sources used to build that piece of content. That last part is the key distinction: you see the exact formula, prompts plus sources, used to make each artifact, not just the full list of everything in your notebook. It is the difference between knowing what is in your pantry and knowing what ended up on the plate.

Iterate: a testbed, not a one-shot generator

Transparency alone would be a read-only feature. The "Iterate" button is what makes it actionable. From the prompts-and-sources window, you can rewrite the prompt, narrow or expand the source selection, and fire off a new generation with updated parameters. Critically, not only are the answers anchored to your documents, but now you can also inspect and tune the way that anchoring happens.

One design decision Google got right: every iteration saves as a separate artifact. Your original output stays intact. You can compare a concise variant of an Audio Overview against the original and keep whichever works, without destroying the previous result. This removes the biggest friction in iterative AI work: the fear of losing a good output while chasing a better one.

Why this matters beyond convenience

For anyone who uses AI occasionally, it is one more convenience. For anyone who uses it at work, where every claim has to be justified and every result has to be defensible, it is the kind of feature that decides whether a tool is fit for professional use or stays a clever toy.

This is the real gap Source Attribution fills. NotebookLM's inline citations, which scroll the original document to the exact relevant passage when clicked, were already considered revolutionary for academic integrity and corporate compliance. But that was transparency about individual claims. Source Attribution adds a second, higher layer: transparency about the generative process itself, meaning the prompt and the weighted mix of sources behind an entire artifact.

There are real limits to keep in mind, though:

  • There is still limited ability to make adjustments after generating certain content assets. For example, you cannot make edits to Video Overviews or Infographics after the initial generation step. Iterate helps you re-run with new parameters, but it is not in-place editing.
  • New features that launch on the web app can be slow to roll out on mobile, limiting their usefulness until addressed.
  • The feature applies to content generated after activation. Older artifacts may not have full recipe data available depending on when your account received the update.

Who should use this, and when

The practical use-cases map directly to situations where auditability matters:

  • Research and academia: Verify which papers actually fed a literature summary before citing it downstream.
  • Legal and compliance work: Reconstruct the exact source mix behind a briefing doc if a claim is challenged.
  • Content production: Systematically A/B test prompt variations for Audio Overviews or reports without losing prior versions.
  • Team workflows: Share not just the artifact but the recipe, so colleagues can reproduce or extend it.

Where it fits in NotebookLM's trajectory

Source Attribution arrives during a period of rapid expansion for the platform. Since October 2025, NotebookLM has shipped eight major upgrades, including Deep Research, a Gemini 3 engine upgrade, built-in slide editing with PPTX export, and Cinematic Video Overviews. Each of those features added generative power. Source Attribution is the first that adds control over that power rather than more of it.

NotebookLM is an AI research tool created by Google that provides responses grounded in user-provided sources, powered by the latest Gemini model. Its core identity has always been source-fidelity over raw capability. Source Attribution deepens that identity by making the fidelity verifiable, not just claimed.

Availability and pricing

The feature is rolling out gradually across accounts, as is standard for NotebookLM updates. If you open the three-dot menu on a Studio artifact and do not yet see "View prompt and sources," check back in a few days. NotebookLM Standard is free forever with 100 notebooks, 50 sources per notebook, and 50 daily chats. Paid tiers run from Plus at $7.99/month through Pro at $19.99/month and Ultra at $99.99 or $200/month, each bundled with the corresponding Google AI subscription tier. Source Attribution itself is not gated behind a paid plan; it is part of the core Studio experience available to all users.

The broader signal here is that as NotebookLM becomes more capable, Google is investing in making it more legible. Powerful and auditable is a harder combination to build than powerful alone, and for the use cases where NotebookLM is already winning, it is the combination that actually matters.