Chat in NotebookLM: Smarter research conversations with clearer results
NotebookLM chat just received a major upgrade. It's faster, handles far larger context, and adapts to your goals so you spend less time wrangling sources and more time producing solid work.
What's new under the hood
- 1M-token context window: NotebookLM now taps the full 1 million-token context window in chat across all plans. Long literature reviews, dense PDF stacks, and complex project histories are now practical in a single conversation.
- More coherent multi-turn chats: Capacity for ongoing conversation has increased over sixfold, improving consistency and relevance across extended sessions.
- Deeper, multi-angle synthesis: The model actively explores your sources from different angles and compiles a single, nuanced answer grounded in the most relevant citations. In large notebooks, this careful context handling leads to clearer insights and fewer dead ends.
- Quality lift you can feel: Early testing shows a 50% jump in user satisfaction for responses that draw from many sources.
Saved and private conversation history
Your chat history now saves automatically, so you can pause a project and pick it up later without losing context. You can delete history at any time. In shared notebooks, your chat is visible only to you. This is rolling out over the next week.
Set goals to shape how chat works
You can now give chat a clear goal, voice, or role. Click the configuration icon in chat, then specify how it should behave and what you want to achieve. This helps the system align with your intent from the start.
- Treat me like a PhD candidate: You are my research advisor. Rigorously challenge every assumption. Ask probing questions, identify logical fallacies, and force me to defend my work from the ground up.
- Act as a lead marketing strategist: Your response must be an immediate action plan. Be analytical and direct, focusing only on concrete strategies and critical-path steps to reach the goal fast.
- Analyze from three perspectives: 1) strict academic (evidence, logic), 2) creative strategist (non-obvious connections, applications), 3) skeptical reviewer (gaps, flaws, risks).
- Act as a Game Master: Run a text-based simulation with a high-stakes scenario, a clear goal, and a step limit (e.g., 10). I make the choices. You narrate the outcomes with realistic details.
How researchers and analysts can use this now
- Systematic literature reviews: Load large paper sets. Ask for a synthesis across methods, datasets, and conflicting results. Then have it critique its own summary from a skeptic's viewpoint.
- Grant and manuscript prep: Set a "peer-reviewer" goal to stress-test your significance, novelty, and limitations sections. Request line-by-line risk checks and missing-citation alerts.
- Data extraction plans: For long reports, define a schema (variables, units, confidence). Have chat extract structured notes and flag any ambiguous passages for manual review.
- Cross-disciplinary scans: Use the "three perspectives" goal to surface non-obvious links between fields while filtering out weak analogies.
- Long-running projects: Keep a persistent thread per project. Resume anytime with the saved history and ask for a recap plus next critical steps.
Quick start
- Open your notebook and add core sources (papers, reports, notes).
- In chat, click the configuration icon and set a goal (e.g., "You are my methods reviewer-find logical gaps and missing controls").
- Ask for a multi-angle synthesis, then follow up with targeted questions (assumptions, edge cases, failure modes).
- Save the thread; return later to continue with the same context intact.
Why this matters
Large context plus goal-driven behavior means fewer shallow summaries and more rigorous, source-grounded analysis. For science and research workflows, that translates to better questions, tighter arguments, and faster iteration.
If you want to sharpen how you write effective goals and prompts, explore practical guides here: Prompt Engineering Resources. For an overview of the Gemini family used behind these improvements, see Gemini at Google DeepMind.
Your membership also unlocks:
