After the latest NotebookLM update, I'm rethinking how much I trust AI

Every AI chatbot greets you with some variation of the same warning: “AI can make mistakes.” And if you’ve spent enough time using these tools, you’ve probably seen exactly why that disclaimer is there. Obviously, I’ve seen enough AI hallucinations over the years to become increasingly skeptical of taking anything an AI says at face value.

That’s why NotebookLM’s latest update has caught my attention. Even when I feed it nothing more than rough ideas, it manages to pull together surprisingly structured research. But that’s only a part of what’s new. The more I explored, the more I realized Google has turned NotebookLM into something much more adept than I expected.

Finally, someone read all 87 pages for me

Having used NotebookLM for everything from college projects to work-related research, I’ve seen firsthand how much effort it took to get useful results. Before this update, I had to do most of the tedious work myself — gather sources, provide context, explain exactly what I was trying to understand, and carefully steer the conversation toward something useful.

Now, the process feels very different. Instead of spending time hunting for articles, videos, and references before I even begin, I can start with a rough idea and simply talk it through. As the conversation unfolds, Gemini pulls in relevant sources that I can import directly into my notebook, review at my own pace, and use to shape the direction of my research.

What impressed me even more was how well it handled large documents. I uploaded my psychology thesis and started asking questions about specific concepts, arguments, and findings. Usually, that kind of task sends me down a rabbit hole of scrolling through pages, opening different sections, and trying to remember where I mentioned something in the first place. This time, NotebookLM did most of the navigation for me.

For instance, if I wanted to revisit a particular theory or check how I had interpreted a finding, I didn’t have to hunt through dozens of pages. I could simply ask a question, and NotebookLM would surface the relevant information and synthesize it into a clear, concise answer. The whole experience felt remarkably natural — almost like discussing the document with someone who had just finished reading it cover to cover. Needless to say, the first few times I used it, I went through everything with a fine-tooth comb, checking paragraph after paragraph to make sure it wasn’t making things up. To my surprise, it was consistently on the right track.

I keep wanting to say it does this really well. In fact, I’ve probably already thought the word really about five times while writing this paragraph. But that’s because it really does. If you’ve ever tried to find a specific insight buried in a massive PDF, you’ll understand why I’m so enthusiastic. NotebookLM doesn’t do my reading for me, but it definitely makes finding the information I need feel suspiciously easy.

AI, please present my own data to me

Outside of writing and research, it gets even better. I work as a freelance social media strategist, which means I spend a startling amount of time staring at spreadsheets full of numbers. Engagement rates, follower growth, reach, impressions, month-over-month performance — all the fun things that make marketers excited and make me reach for a calculator.

The problem is that numbers and I have never had a particularly healthy relationship. I can spot a typo from a mile away, but ask me to calculate percentage growth without double-checking it three times, and we’re entering dangerous territory.

That’s where NotebookLM’s presentation tools have become useful. Instead of spending an afternoon wrestling with slides, charts, and numbers that may or may not betray me at any moment, I can simply dump all the metrics into NotebookLM, explain what I want the presentation to focus on, and let it get to work.

Say, for instance, I’m preparing a monthly client report — I’ll throw in everything from follower growth and engagement rates to reach, impressions, and top-performing posts. Then I’ll give it a prompt that explains the story I want the data to tell. A few minutes later (I won’t lie — sometimes even more), it comes back with a slide deck that’s more organized than anything I would have created. What I like most is that it doesn’t just scatter numbers across a dozen slides and call it a day. It actually identifies trends, highlights key takeaways, and structures the information in a way that feels like a proper presentation. It’s almost annoying how competent it is.

And when it’s finished, I can export everything as a presentation or PDF, make a few tweaks, and send it to a client. That export option is particularly important because, as much as I enjoy AI tools, I don’t necessarily want my clients opening a link that screams, “An AI made this.” The final result looks like a polished presentation that I spent hours carefully assembling myself. Between us, of course, we know that wasn’t entirely true.

I’m glad my notebooks aren’t homebodies anymore

One of the most frustrating parts of using NotebookLM, at least previously, was how limited the exporting options felt. When I’m deep into research, I don’t just want everything in one tool — I want to take it elsewhere when required. A Word document for editing, a spreadsheet for breakdowns, and a presentation for sharing. Basically, I want my work to move with me.

But for the longest time, that wasn’t really the case. Google largely kept exports tied to its own ecosystem, which meant you could comfortably move things into Docs, but anything beyond that felt a bit restrictive. If you wanted more flexibility, you were basically stuck juggling workarounds.

That’s starting to change, though. The latest update adds native export support for formats like Word, Excel, Markdown, and AI-generated images via Nano Banana. This finally feels like my work isn’t trapped inside one app anymore. Now, I can simply take a structured research summary, export it straight into a Word file for editing, drop data into Excel for analysis, or move notes into Markdown if I’m working in a different writing setup.

The catch, of course, is access. These features are currently tied to higher-tier setups like Google AI Ultra and certain Workspace plans, which means most everyday users won’t see them right away.

So, for standard users, it’s still a mixed experience. You either rely on limited native integrations or deal with browser extensions that don’t always behave the way you want them to. And while it’s tempting to feel left out, this is probably one of those features that will slowly trickle down to more users over time. Until then, it’s best to wait for Google to expand the rollout.

And that’s really the point I’m trying to make here. I still don’t fully trust AI to just take over my research end-to-end, and honestly, I don’t think I’m ready for that level of confidence yet. But NotebookLM is definitely starting to blur that line in a way that’s hard to brush off.

Because think about it — an AI tool that doesn’t just answer your question, but actually helps you build the entire process around it. It pulls in sources when you need them, helps you make sense of messy ideas, explains things in simple language when you’re stuck, and even turns all of that into something usable, like a slide deck or a structured document. And the important part is, you’re still not out of the loop. You can bring things in, take things out, double-check everything, and decide what actually makes sense.

That balance is what makes it interesting for me — it feels like having something sit beside you, handling the boring part of the process while you still retain full control over the final output. And honestly, that’s the line I don’t want AI to cross. The idea of it doing all my thinking still makes me uneasy. But a tool that carries the load and leaves the thinking to me? That, I’ll take every single time.