Transcription
Every AI tool hallucinates legal citations that don't exist, reportedly 10 to 20% of the time. But there's one tool that refuses to answer if it can't cite the exact passage. Every claim linked back to the exact paragraph in your source document. And most law firms have never touched it.
I'm Liam Barnes, founder here at Cyberactive. I've spent hundreds of hours testing AI tools for legal work. I've built workshops around them. And when I show partners Notebook LM for the first time, their reaction is mostly the same. It's, "Why did we not know about this?"
So today, I'm walking you through five Gemini features that are really valuable for legal teams. And again, this is not just a product demo. It's a guide. It's going to show you exactly where each one lives, how to set up each feature, and what it looks like with real documents in real legal workflows. And if your firm runs on Microsoft, as most firms do, don't click away. Four of these five features work with Google Workspace. You just need a free Google account.
So, the first feature or tool I'm going to take you through today is called Notebook LM. Notebook LM is what's called a RAG system, retrieval-augmented generation. Here's what it means in plain English: You upload your documents, you ask questions, and Notebook LM answers only from your uploaded material, not from the internet, not from its training data, only from your sources.
Now, recently, Notebook LM has added the ability to search web sources as well, but as I'm going to show you in a second, it's very easy to distinguish between whether information returned in search results has come from the internet or your source material.
So, let's go through an example right now. Let me give you a quick lay of the land. So, you've got your sources and you've got your chat here in the middle, and then you've got the studio panel. So once you start populating data, you can then use these various different options to generate content either for presentations or or even just to learn the information. So the video overview or the audio overview can be a great way to learn new content.
All right, so let's get into a demo. I've uploaded three documents. Number one, a regulatory guidance document. Number two, a court opinion. And finally, a contract. And as you can see here in the middle of the screen, Notebook LM has processed them and already given me a summary of the documents. Now, I'm about to ask it a question, but firstly, you see over here where you can select from the web or you can select your connected Google Drive. That's for new sources. But because I've got my sources checked, those are the sources that are going to be referenced when I type in any query or any prompt into the chat box right here.
So, let's start with, "What are the notification requirements for data breaches under this regulation?" Okay, so that took 10 seconds or so. Now, let's look at the response. As you can see here, every response has a numbered citation. If I click number three, it takes me to the exact paragraph in the regulatory guidance document.
Now, watch this. I'm going to type in something that is not in my uploaded documents. So, "What are the HIPAA notification requirements?" And then look at the response we get: "Based on the sources provided, there is no specific information detailing the notification requirements under HIPAA."
Now, just think about that for one moment because that's a really key difference. ChatGPT would have answered anyway and possibly even made something up. Claude would have given you a response from its training data. Notebook LM says, "I don't know," because the documents don't contain that information.
So, it sounds pretty good, right? But let me be honest with you about what this means. This is not zero hallucination in absolute terms. Notebook LM can still misinterpret a passage, as an example. It can draw the wrong inference from a correct source, but it will never fabricate a citation that doesn't exist. And every claim is traced back to the exact paragraph. No, verification takes seconds instead of minutes or even hours if you have no idea where something came from.
So, we've got zero fabrication, instant verification for legal work. That's really best-in-class. Now, even if you are a Microsoft organization, you can use Notebook LM for free. You get 50 sources per notebook. So, this is just one notebook. You can create many different notebooks and 500,000 words per individual source. And as with most Google products, they have a premium model, right? So you can start using this today. No subscription, no Google Workspace required.
Second feature I want to show you is an ecosystem play. And obviously, this won't be as relevant potentially if you're a Microsoft organization, but there's many lawyers and in-house legal teams that use Google Drive and the Google ecosystem. And so this is where it gets quite interesting. Think about where your firm's documents live right now. Maybe SharePoint, maybe a local server, maybe someone's desktop. Wherever they are, you probably can't search across them with AI. And Google Drive changes that.
So, as with any Google product, you can easily navigate between them just by clicking the icon up here. And now, we're going to go over to Gemini. So, if I just click Gemini, it's going to open up. And we're going to come down to settings over here. And you're going to click connected apps. And then, just make sure that your Google Workspace is toggled on. Right? So, now it's connected. And you can see here Gemini, Google Calendar, Google Docs, Drive, Keep, Tasks. And so we can now use Gemini to search across all of these different platforms and products.
So, we'll just click out of this. And so we'll just start a new chat. And because it's using AI, it's much more powerful than a regular search. And so I'm just going to say, "Find our NDA with Vert.Ex Technologies." And you can see here it's connecting to the workspace. The AI knows to search our internal storage space. And here's the results: "Found a mutual non-disclosure agreement between Vert.Ex Technologies and Whitaker Reeves and Associates LLP. You can access the full document here."
I can click to see the document to make sure it's the correct document. And I can actually just try again without apps. It might be a little bit hard to see on the screen, but that means it wouldn't be searching from our connected Google Workspace apps. So even though this was just a simple example, hopefully you start to get a sense for what your firm library can look like when it's AI searchable. For example, "What indemnification terms have we used with healthcare clients?" It would find every contract in Drive that matches that search query. "What were the security obligations in our last SaaS agreement?" It surfaces the clause from the exact document. So it gets really, really powerful and saves a whole lot of time.
And for many firms, when we're talking about a per-seat cost of $15 to $50, cost is not a major contributing factor. But for many legal teams and law firms, it is. And so you can see here on the screen that Google Workspace, the Business plan is much more cost-competitive and actually delivers a better feature set for a lower price. And so for this price, you could have your firm's entire precedent library, template bank, and policy documents all searchable by AI for a fraction of what you're probably paying right now.
Okay, so the third thing I want to show you is deep research with Gemini. And this is where Gemini goes from useful to genuinely really powerful. Gemini has caught up incredibly fast over the last couple of years since the Bard fiasco, to the point where Gemini is now taking market share away from OpenAI at the enterprise level. Now, when we're talking about using Gemini or any LLM for that matter, their regular model, so when we're not talking about a deep research or a reasoning model, it gives you one answer from one search. But deep research breaks your question into sub-questions, runs dozens of searches, reads hundreds of sources, and produces a structured report with citations.
So, let's see a working example. And I've just switched it to light mode. I think that some of the buttons were not as visible in dark mode. And so we're going to come down here. We're going to select deep research. Okay. And you'll, depending on your plan, you can select different models, but we're going to do thinking as well. Okay. Okay. And we're going to type, "What is the regulatory landscape for AI in healthcare across the US, EU, and the UK?" And we're going to submit that.
So, we can see here that Gemini is generating its research plan instead of just answering immediately. So we can evaluate the plan, we can even edit it, or we can click to start the research. So let's start that research. It's now opened up a panel on the right-hand side. So we can see the process it's going through. It's identifying regulatory gaps and divergences, strategic research path. We can see the websites, so the sources that it's referencing, and you can go back in and look at the research once the output has been generated, which obviously gives you confidence in the quality of the research. If you're not happy with some of the sources, then you can obviously run the search again and give it that feedback. You could tell Gemini to omit certain sources or focus on specific types of sources.
And so that took about four or five minutes. It was a really deep research that is not your average research that you get on any of these LLMs' basic introductory model. It's not training data from last year. If a regulation changed yesterday, deep research would have found it. And in my experience, I've found Claude and also OpenAI ChatGPT to show less up-to-date information. And so real-time research is generally of a lower quality than Gemini. And it makes sense because Gemini and Google are indexing search results, and Gemini can immediately get access to those search results given that they're in the same ecosystem. The other thing about ChatGPT is even though it does have web search, it doesn't show you the research plan first. And with Gemini, as you saw there, you see exactly what it's investigating before it starts the research process.
And just a couple of other things, you can see the entire contents of the outputted research report. You can share, export it, but you can also create some of those media assets that I was referring to earlier when we were looking at Notebook LM. So you could turn this into a web page, an infographic, an audio overview, a bunch of different things. Let's see what it does when I select the web page. And after just a couple of minutes, it's produced the research report in the format of a web page. Now, how useful that's going to be to you obviously depends on your specific use case. But if you're a visual learner or you prefer to listen, then you could turn it into an audio note or a podcast conversation. And so again, it's a really good way to not only visualize information, conduct research, but also learn about specific subject matter as well.
So, feature five, I'm going to show you two tools that change how you consume legal information. So first, we've got the audio overview in Notebook LM. So you upload documents as we did earlier, and Notebook LM generates a podcast-style conversation between two AI voices summarizing the key points. And to do it, it couldn't be easier. Essentially, you just come over here and you click audio overview. Now, you've got to select the sources that you want to be contained as part of the audio overview. So, we're going to select all of them. I'm going to click audio overview. And you'll see here it says, "Generating audio overview. Come back in a few minutes."
And so, you can see here that it's now completed. And so, let's have a listen to it. And like I said, having it on your phone, so you can just download the Notebook LM app, allows you to listen to whatever's generated on the go. And so, I'm going to play it from the phone.
"Today we are, uh, we're going to make you look at your company's servers and probably your legal team very differently."
"Oh, absolutely."
"You know, we usually hear people say that, um, data is the new oil, right? It's the standard cliche, meaning it's this incredibly valuable asset that powers the modern economy."
"Right, it's the fuel, but if you actually look at the stack of..."
And I can even increase that to 2x speed if you like to listen to things quickly out on the go. And so, another really useful feature inside of the Google ecosystem. And so what you've got there is essentially two AI voices breaking down a regulatory guidance document in a conversational format. Key provisions, implications, what matters in about 10 minutes of audio. It's just a really efficient way to learn things. A 100-page brief distilled into a 10-minute podcast format audio file that you can listen to anywhere.
Okay, now I'm going to show you a cool feature of Google Chrome. So, open up any web page. It could be a court opinion, an SEC filing, a regulatory guidance document. And so, let's just, as an example, let's click into this here. And what you're going to do is you're going to click, it depends on your Google Chrome. Sometimes Google displays things differently. Sometimes you'll have a little Gemini icon. Sometimes you won't. And then what you need to do is just click into the, uh, the search bar or the address bar. Come down. And then see here where it says, "Ask Google about this page." Click on that. Then you've got a pop-up, right? Right? And then you can interact with what's on the page, what's on the screen. And so then we're going to ask some information. So, "What is the holding in this case?"
You can see this panel open up on the left-hand side. So, what we can see here is Gemini is analyzing the page content and it's going to provide a concise answer about the case's holding based on what's on the specific page. And so you can just start to picture your workflow where you can leverage AI across so many different touchpoints and use cases that you really can increase your productivity and your overall efficiency throughout an entire workday.
Now, as with any LLM, Gemini has its drawbacks. It has its limitations. So let's talk about what Gemini can't do because a balanced view matters, right? More than just a sales pitch. At the end of the day, you need to choose the tool or tools that are right for your firm or your specific use case.
So, the first thing that I think is worth pointing out is that Gemini with web search still hallucinates. About 10 to 15% of citations from regular Gemini queries may be inaccurate. Notebook LM with your own documents is more reliable, but Gemini searching the open web is not immune. So you've got to verify everything.
Second, there's no legal-specific tools as of the end of February 2026. Claude has a legal plugin with slash commands built for contract review. Gemini has nothing equivalent. No slash review contract, no slash triage NDA. It's a general-purpose tool. But if I were a betting person, I would probably bet that Gemini is going to plug some of these gaps soon.
Third, no Westlaw, no LexisNexis, no paywall databases. Same limitation as every AI platform. Gemini works from free web sources, and you upload your documents.
And then fourth, data privacy varies by tier. Consumer Gemini may use your data for training. Google Workspace Enterprise excludes your data. For client confidential work, use the Enterprise tier. Make sure your IT team knows the difference.
And then finally, Notebook LM doesn't search the web unless you ask it to.
So, let's do a quick wrap-up of what we've gone through. Five features. We've got Notebook LM. So, upload your documents, get cited answers, zero fabrication. Google Drive as your firm's AI searchable corpus. Approximately $14 a user, definitely cheaper than your current DMS. You've got deep research, multi-step investigations with real-time web data and visible research plan. Gemini in your workspace, AI built into Google Docs and Sheets for firms on Google. And then finally, audio overviews as well as Chrome in the side panel.
Notebook LM has a very generous free plan. Deep Research is $20 a month. Google Workspace starts at about $14 bucks a user. You don't need to overhaul your firm's tech stack to try this out. I would say just open Notebook LM, upload a brief, and ask a question. That's your first five minutes, and you'll start to get a sense for the quality of the output and whether you can incorporate it into your existing workflow.
If this was useful, I'd really appreciate hitting the like button. Leave a comment. It helps more legal teams find this type of content. And if you want to see more videos on how legal teams can use AI to increase capacity and drive revenue, please subscribe. I put out a new video or two every single week. Thanks for watching, and I'll see you in the next.