I thought NotebookLM was enough — then I tried Claude Projects.

NotebookLM needs no introduction to most people these days — for good reasons. This AI-powered research assistant tool from Google has been a trusted companion for a variety of research-oriented tasks. I have also discussed how I integrate NotebookLM into my workflows, often alongside other tools such as Claude or ChatGPT.

Nevertheless, painting NotebookLM as the single most powerful AI assistant that can do everything could be a mistake. As you will find out sooner or later, there are a few things NotebookLM cannot do right. Depending on what your research entails, these bottlenecks could have you looking for alternate options — and that’s completely alright.

The thing NotebookLM can’t do, no matter how you ask

Retrieval and reasoning aren’t the same thing

As I said, NotebookLM is a truly reliable research assistant! It does an impressive job of providing information grounded in sources and performing many functions related to retrieval. Therefore, when you prioritize source fidelity and want to create content such as an audio overview, infographic, or mind maps, you can set up a near-perfect workflow using NotebookLM. The fact that its basic version is free is another reason to try it out.

However, there’s something NotebookLM hasn’t managed to do despite me asking it countless times: reasoning.

It often feels like source fidelity is the only principle that NotebookLM is based on. When you ask it to reason with something or place your argument in a larger context, NotebookLM tends to fail more often than not. The lack of behavioral control is also a problem. Providing custom instructions for a specific thread or prompt isn’t really an option right now.

I don’t really want to blame NotebookLM here, since it’s just doing what it’s supposed to. Yet, it meant it was time to check out alternative options.

What happened when I set up a Claude Project for the same workflow

The instructions field changed more than I expected

I came across Claude Projects while searching for viable alternatives. Setting up a Claude Project and configuring its multiple options was a refreshing experience after NotebookLM. Here’s how I went about it.

You can access cloud projects from the Claude interface sidebar, which opens a specific page. If you have already created a project, you should see a few options, but you can click New project to get started with Claude Projects. You do it by providing a project name and stating what you are trying to achieve. The possibilities here are endless, and you can set up the project however you like. For example, I was trying to build an assistant for a course I’m teaching.

The moment you enter the project page, you’ll notice two sections on the right side: Instructions and Files. These two customizable sections make Claude Projects stronger across many areas where NotebookLM struggles.

• Instructions are where you can provide text-based ideas on what you want the response to be. For instance, I may want the responses to be formal and suitable for academic interactions. Or I may want to ask Claude to tailor the responses using a template. These instructions apply to all the chats you have under the project and may work alongside the custom instructions you may have set for your profile.
• Files are another way to control how Claude responds to your queries within the project. Just as you can add custom instructions, you can also add files from different locations to the project. For instance, I may want to add the prescribed format for the course plan or a syllabus overview. I can go to Instructions and ask Claude Projects to follow that particular file for the structure. This is also where I would ideally add the project’s references.

From this point onward, Claude Projects will ensure that every response to your queries is based on these instructions and files. This is on top of the reasoning that Claude performs better than NotebookLM in the core models. You can also switch between multiple models from core, Claude, or attach additional attachments depending on the prompt.

Memory and continuity are doing more work than I initially credited

Rebuilding context from scratch gets old fast

Another problem I commonly face with Google’s NotebookLM is that it doesn’t have cross-notebook memory. Every time you create a notebook or start a chat, it’s from scratch. This doesn’t work well when you want coherent responses based on prior memory. Sure, you can select the sources and customize a few things, but it doesn’t result in an efficient workflow.

With Claude Projects, you get the benefit of memory and continuity alike. Memory is populated when you have enough chats within the project, and your preferences and common patterns are recorded as such. Continuity also plays an important role because all your chats are interconnected. If you have requested a specific type of response in a previous chat, Claude Projects will be keen to follow the same approach in the next chat as well.

More importantly, you can set custom instructions to control how your responses are structured. Considering how Claude is almost always better than NotebookLM when it comes to content creation and reasoning, these features actually improve the results that you get.

Where NotebookLM still sits in my workflow — and why that’s not a concession

You may already know that NotebookLM and Claude serve different purposes. When it comes to research, NotebookLM is still part of my workflow, and that’s not really a concession. There are definite times when I have to turn to NotebookLM rather than stick to Claude Projects, and that mostly has to do with source fidelity.

That is, when I know that I cannot base my responses on anything other than the source text, I need to stick to NotebookLM. However, when you can slightly compromise on source fidelity but want better control and continuity in your responses, you should always try Claude Projects.

Developer

Anthropic PBC

Price model

Free, subscription available

Claude is an advanced artificial intelligence assistant developed by Anthropic. Built on Constitutional AI principles, it excels at complex reasoning, sophisticated writing, and professional-grade coding assistance.

OS

Android, iOS, Web-based app

Developer

Google

Pricing model

Free

NotebookLM is Google’s AI-powered research notebook that reads what you upload and helps you transform it into structured summaries, explanations, and visuals.