Google quietly turned another crank on the automation engine inside NotebookLM this week: instead of exporting a static deck and tweaking slides by hand, you can now tell the model what to change and let it rebuild the whole presentation. That sounds small. It’s actually a sign of how AI tooling is moving from one-shot generation to iterative, conversation-driven document creation – with all the benefits and annoyances that implies.
Here are the concrete updates Google shipped (public details preserved): Gemini 3.1 Pro is available in the Gemini app and NotebookLM for Google AI Pro and Ultra subscribers. NotebookLM-generated slides on the web now support Prompt-Based Revisions: open the deck in the Studio panel, tap the pencil icon next to the title and you’ll get a fullscreen interface with a ”Change Slide 1” prompt box to edit text, color, visuals, and more; after making edits you tap ”Generate new deck” to create a refreshed version. The Prompt-Based Revisions feature is now rolled out to all NotebookLM users. Slides can be exported in the PPTX format, with Google Slides coming next.
The NotebookLM app on Android and iOS now lets you Customize Video Overviews. Options exposed to users include Format: Explainer (default) or Brief; the ability to Select sources; Visual style choices listed as Auto-select, Custom, Classic, Whiteboard, Heritage, Paper-craft, Watercolor, Anime, Retro print, Kawaii; a ”What should the AI hosts focus on” prompt box; and Language. The mobile app is also rolling out the ability to open Google Docs, Sheets, Slides, and other Drive files in the corresponding app (or a new browser tab if it’s not installed). Chat Suggestions are coming as well: to get the conversation going, you will now see suggested questions appear in your Chat; these questions will continue to customize based on your conversation with NotebookLM.
Why this matters
Prompt-Based Revisions change the mental model. Instead of exporting and patching a deck, users can iterate inside a single prompt loop: ask for a tone shift, swap visuals, or reflow content and let the model regenerate the deck. For fast, one- or two-person workflows this is a huge time saver. For teams that rely on strict brand templates or meticulous slide craft, it introduces unpredictability – a regenerated deck can subtly change structure, spacing, or phrasing in ways that require manual fixes.
This move also underscores Google’s product strategy: pay tiers get access to the newer reasoning model (Gemini 3.1 Pro) while other users remain on older models. That’s an explicit monetization path for advanced AI features, and it favors customers already invested in Google’s ecosystem.
Where this fits in the market
Competitors have been pushing presentation automation for years. Microsoft’s PowerPoint with Copilot offers automated slide generation, design suggestions, and iterative editing tied to Office assets. Canva and standalone tools such as Beautiful.ai let users generate and export PPTX files from templates and AI prompts. What’s different here is NotebookLM’s combination of an assistant that’s already ingesting your documents and a prompt-first edit flow: the model has the context, and you can refine output conversationally.
That proximity to personal documents is powerful – and risky. Organizations using NotebookLM for research or client work will want clear admin controls and data-handling guarantees before letting sensitive files feed into prompt-driven generation. Expect enterprise IT teams to press Google for policies on model training, retention, and admin visibility if NotebookLM’s slide tools creep into business workflows.
What’s missing and what’s next
Two things are noticeably absent: first, firm controls to lock layouts and brand tokens during a regenerate operation; second, deep collaboration primitives so multiple editors can review the regenerated deck without stepping on each other. Both are obvious next steps if Google wants NotebookLM decks to replace professionally produced slides rather than supplement them.
Practically, I expect Google Slides integration to arrive quickly (it’s already promised), followed by template-locking for Workspace admins and better change-tracking when decks are regenerated. Competitors will respond by tightening enterprise controls or by making their own prompt-based flows more deterministic – because predictability matters to the people who sign the invoices.
For users, the takeaway is simple: Prompt-Based Revisions make it faster to turn notes into presentations, but if your work needs pixel-perfect slides, treat the regenerate button like a helpful junior designer – it moves fast and saves time, but you’ll still want to give the deck a final pass.
And if you’re testing NotebookLM for the first time: play with the ”Change Slide 1” prompt box and the Visual style presets. They show where Google hopes most users will spend their time – iterating until the AI lands the right combination of copy and visuals, rather than polishing every element by hand.
Watch for Google to lean into this conversational editing pattern across other document types next: slides are just the most visible experiment in making AI-driven iteration feel natural.