Photo by editor
# Introduction
To create a Document product requirements (PRD) is a common practice. Product management And a common task in fields like software development and the tech industry as a whole. Some of the difficulties and stringent requirements commonly encountered in creating a PRD include ensuring clarity, preventing scope narrowing, and preserving stakeholder alignment.
Thankfully, AI tools have emerged to help navigate these challenges more effectively, without completely delegating the strategic decision-making underlying the PRD creation process – in other words, humans are still in the loop. Here is an example. Google’s Notebook LMwhich synthesizes on-the-ground raw data or content to answer questions, thereby turbo-charging the workflow to create on-the-ground, useful PRDs.
Through the process of using Notebook LM’s features, this article will navigate you through the process of converting raw, sometimes chaotic information into a grounded PRD in minutes, based on a beginner-friendly use case. Spoiler: It won’t just be about chatting with an AI assistant.
# From messy notes to structured PRD drafts
Let’s consider the following scenario. You are a newly hired product manager for a startup that wants to develop a new mobile app. Flora Friend. The app aims to help people stop accidentally killing their houseplants.
The team, including you, has put together a set of three “mess” documents detailing what a potential app should look like:
interview_transcript_matt.txt: A 30-minute interview with a user named Matt, who owns over 50 plants. In these interview notes, Matt says that existing apps are “overcomplicated” and make it difficult to keep track of things like “what fertilizer to use.”competitor_research_notes.txt: A rough list of bullet points created after analyzing competitor apps like “PictureThis” and “Planta”, highlighting their paywalls and interface flaws.brainstorming_whiteboard.jpg: Random but somewhat “cool” ideas mentioned by the team during lunch breaks and other casual conversations, like “Spotify playlists for plants”, “watering reminders”, etc.
Imagine a complete document containing all the content mentioned above. Manually converting them into a clean PRD that nicely brings them all together can seem like a pain, right? Enter the Notebook LM!
Log in to Notebook LM with your Google account and click on “Create a new notebook“Give your new notebook a name, something like”Flora Friend PRD“
Once a new notebook is created, you will be welcomed to the main Notebook LM interface, which looks like this:
Notebook LM interface
A word of caution: This newly created notebook is not intelligent. It is not regular. Large language model (LLM); It does not know plant care or any other specific topic. But we’re going to teach him an “express” master’s degree about it with our messy—yet enlightening for tool—notes.
Let’s say you have the above three files containing some content related to a plant care app, or any other raw data files of your own. You can upload them to Notebook LM Canvas using the Upload button in the main, central section.
Once uploaded, you can think of your notebook as a small, toy-sized one. Breeding increased by retrieval (RAG) system that can begin to think and behave like an AI based on the information it has access to. In fact, without asking, by clicking on one of the uploaded files on the left, NotebookLM generates a comprehensive, well-organized summary of the contents of that file: it’s called the file’s summary. Source Guide.
Now comes the important part. We can ask something like “write PRD” in the chat box below, and that’s it. But we want to do it right and provide clear, specific instructions, and that involves some quick engineering, namely forcing the nascent AI to prioritize what we want our PRD to reflect: prioritizing user problems over random ideas generated by the team (without completely ignoring them). Here’s a well-crafted tip that works:
I am the Product Manager at Florafriend. Draft the PRD based on these sources only.
Important Restrictions:
1. Prioritize features that address the pain points described in interview_transcript_matt.txt.
2. Exclude any ‘brainstorming’ ideas that do not directly address the user’s problem.
3. Structure the output with these headers: Problem Statement, Core Features, Functional Requirements (UI/UX), and Success Measurements.
Try adapting this prompt to your business problem or use case. Once submitted, you are likely to get a nice and clean PRD with key sections like problem statement, core features, non-functional (UI/UX) requirements, success metrics, etc.
Interestingly, PRD has what looks like numeric references that you can hover over. If you do, you’ll see the source (one of the source files) pop up:
Before accepting that first PRD, remember that a first draft is rarely perfect. Engage in conversations to improve it incrementally, for example if you see a monetizing section missing, ask: “Based on competitor_research_notes.txt, what monetization models are our competitors using, and what should we avoid?Then, manually check the outputs, make sure they remain the first PRD draft, and add key monetization insights, either manually or by asking NotebookLM’s AI to do it — if you choose the latter, always check what you get before blindly approving: AIs can make mistakes!
There is icing on the cake. Audio review Section (Studio) on the right-hand panel. By simply clicking on it, you will generate an audio overview of the information contained in the source files. It’s a great way to absorb information when reading might be less engaging, such as when you’re on your daily commute.
# Next Steps
This article introduces the capabilities of Notebook LM to generate grounded PRD specifications from raw, messy documents in a matter of minutes with very simple steps. From here, a valuable next step can be taken. Google’s Anti-Gravity To convert your PRD specifications into a functional software prototype.
Iván Palomares Carrascosa Leader, Author, Speaker, and Consultant in AI, Machine Learning, Deep Learning and LLM. He trains and guides others in using AI in the real world.