How I use ChatGPT to write PRDs as a Product Manager?
One thing that I’ve learned over the years is a well written PRD brings a lot more clarity to the team while solving specific customer problem and it also reduces back and forth between product & engineering teams.
Remember, this image?
That’s why most CPOs or VPs of Product share certain templates, and PMs have to mostly fill them with all the details.
Honestly, this helps in a couple of ways:
- PRDs follow certain standard practices.
- You don’t miss anything critical while writing PRDs.
- Reviewing PRDs for the leadership team also becomes easier.
But, it takes around 2–3 hours to write a comprehensive PRD covering all the scenarios and edge cases — and as PMs, we don’t have time :)
We’re always jumping from one meeting to another. I face this too, and it’s hard to get focused time without interruptions.
“Hey, can we connect for 5 mins over huddle?” or “Quick Huddle?”
[we are all in the same boat]
This is where AI tools helped me and we all know it will get better and better from here.
I’ve been using ChatGPT from almost the time when they launched it. Over time, I figured out a better way to use it for my day to day product manager tasks starting from crafting well written PRDs within 15–20 mins provided I already have all the necessary details.
Let’s jump to How I do this?
1. Create the PRD Template and Break It into 3–4 Parts
First, I start with a template and break it into different sections, such as:
- One-pager document
- Writing the Introduction or Objective section with Success Metrics, Estimated Impact etc.
- User Stories and Acceptance Criteria
- GTM (Go-to-Market)
2. Next, I Share the Template in ChatGPT with a detailed prompt
Quality of content depends a lot on the prompt, context and specific detail that you share about the feature. More details within a structured format drastically improves the response quality.
With different iteration, I also realised that adding the company, pod or charter gives LLM initial context to understand the feature in depth, drafting better quality response.
Act as an expert Product Manager at <company_name> and help me write the PRD for <specific task in detail> by following this template. Let’s do it step by step, first we will focus on <part of PRD eg: Introduction> using the template below.
[More details about feature]
[Template]
Here is the example prompt for writing PRD for a newly launched product feature by Zomato around “Recommendation from Friends”.
You are the Lead Product Manager at Zomato working in personalisation and recommendation charter. Help me in help me write the PRD for “Recommendation from friends” by following this template. Let’s do it step by step, first we will focus on Introduction and Observation section using the template below. More details about Recommendations from Friends — The idea is to allow users to share recommendations with their friends if they like something. Initially, users will be asked for permission to access their contacts. Once granted, it will show recommendations from their contacts list across each restaurant listing. This will also improve the discovery of restaurants within filters.
[PRD Template]
Highlighting again on the importance of providing contextual information about the feature you are working on. Try to share a lot of details — what’s the problem or opportunity that you are trying to solve, who is it for and what would you improve. If it’s an experiment, highlight that it’s an experiment and you are testing multiple variants.
You can even use some dictation app or speech to text app just to speak your thoughts and paste the transcription here (more details on some other post).
Important note: You do not have to share any sensitive details with LLMs. You can also turn off the training data inside ChatGPT and trust them that they will not use any of the information for training the models.
3. Iterate if it Misses something
In certain cases you may need to go back and forth to refine specific parts. Alternatively, you can use the output as a reference rather than starting from a completely blank slate and make it your own. You will have to proof read everything before directly copying it because sometimes AI will just fill in random data confidently around Observation & Insights section.
Here is an example conversation where I played around “Recommendation feature of Zomato”
https://chatgpt.com/share/675d7ec6-0434-800f-92a0-07e105f88f20
Why This Method Works
Essentially, I’m doing three core things here:
- Providing as much context as possible in terms of the company, role, and product feature I’m working on, so the LLMs can follow a set of guidelines.
- Breaking the PRD into smaller chunks instead of asking it to generate everything at once. Models has limited context windows and it can hallucinate or take shortcuts.
- Sharing an exact template in which I need the output.
You can also experiment with prompts by adding your sample writing templates and asking ChatGPT to follow the same.
I’m confident this method will save you at least 1–1.5 hours every time you write your next PRD.
Here is the PRD Template that you can duplicate or refine based on your use case or even use your company’s PRD template https://docs.google.com/document/d/13jrXdK4yinP7qv0lw6GcQXvB4JuDfq6zZZOEhg4yQ14/edit?tab=t.0
Happy writing!
