July 13, 2026
Sabina's first notebook was bad and she says so out loud, to a room of forty people. Two live sessions, one recursive trick worth stealing, and the question nobody else is asking.
Sabina's first notebook was bad.
She'll tell you so herself, out loud, to a room of forty people. She was building a research base on European AI legislation, dumped in everything that looked relevant, and got back mush. Not because the tool broke. Because she hadn't done the thinking yet. She tore it down and rebuilt it after going somewhere else first to figure out which sources actually deserved to be in there.
That's the whole lesson and she gets to it faster than anyone we've watched teach this.
Google renamed the thing recently. NotebookLM is Gemini Notebook now, same tool, new sign on the door, and half the room still calls it the old name. What hasn't changed is the one design choice that makes it different from everything else in your stack: it only tells you what your own sources say, with citations pointing straight back at them. (Google)
Closed room. That's the feature. Sabina uses it because nothing wanders in off the internet uninvited.
Now the move worth stealing.
She built a notebook out of a photographer's public teaching material on light and color. Asked it for a report on camera basics. Then she took that report, fed it back into the same notebook as a brand new source, and generated a presentation off it. Then a short video off that. Each pass tighter, more specific, more hers. Google didn't name that feature. Sabina found it by pushing buttons.
She also stopped mid-session and asked the room a question nobody asked her to ask. Is it actually fine to build on somebody's free teaching material this way? To route around a video somebody makes a living from? She didn't have a clean answer. She called it a gray area and left it sitting there. Google's own terms say don't upload what you don't have rights to, and repeat infringement kills your account, so the instinct is worth having. (Google)
Most AI content skips that part. We're not going to.
Numbers that change what's worth trying: fifty sources per notebook free, more on a paid Google plan. Eighty-plus languages, and Sabina says it handles hers better than the big chat models do, which is not a small thing if you don't work in English. Linked sources are snapshots, not live feeds. What you added is what you added.
It will not automate anything. It will not check your data on the first of the month and email your team. Somebody asked. It doesn't.
It's a room built for one job. Understanding material you already trust, all the way down, in whatever format finally makes it stick.
Lynda's way in is the simplest we've heard. Gather what you've got. Listen to the audio overview first, cold. Then go back and add more once you can feel where the holes are.
You don't need the whole library to start. You need a frame, and the nerve to keep feeding it.
