A common complaint about AI tools is that they spit out slop, but this is mostly user error. If you want a document that looks like a product strategy, ChatGPT will create one in about 10 seconds. A PM who tells their team “build something users want” won’t get good results either. Better input yields better output.
Product managers are generally early adopters. Still, seeing how others use AI can inspire their own usage. I put together some example prompts as part of our Spring Cohort that I’ll share here today. They will help you:
Create a feature taxonomy for your own product or a competitor’s.
Turn a product strategy into a compelling narrative to sell your vision.
Frame every PRD within the context of that strategy.
Evaluate your PRDs from a CPO’s perspective
I put together example inputs and outputs for each as well, so you can see how they work and adapt as needed for your own work.
The formula for great AI workflows
Stepping back just briefly, my hope is that you can use these prompts right away, but also that they encourage you to think more deeply about how to use AI in your work. There’s a framework I used to create each of these that can help you find new ways to use AI.
First, you need to understand AI’s capabilities. This is always changing (and improving) and you have to use lots of tools to understand AI’s current strengths and weaknesses. We’ll use a handful of different tools to run the prompts below based on the needs and constraints of each.
Next is task decomposition. When you break a task down into smaller components, it helps you think through the steps and it’ll help you get much better outputs from AI. In general, the bigger the task, the more generic the output will be. And the more narrow the task, the more specific the output will be. As an example, you should separate the creation of PRD from the evaluation of it. The first part requires creativity and narrative. The second part requires critical analysis. Combining this into a single workflow would water down the results.
And lastly, bring conviction, opinions, subject matter expertise, data and non-consensus thinking. You have to fuel the workflow with your own insights to get good results. This is where AI really shines—helping you turn your ideas into something tangible you can advocate for, build and sell. (How you find those unique insights is outside the scope of this article but we have a course on it called Mastering Product Management.)
With that foundation in mind, let’s get into a few prompts that can help you in some of the important work you do as a PM.
Prompt #1: Create a feature taxonomy (from your customer’s eyes)
It’s always helpful to understand your own product from a user’s point of view. One way to do this is to create a feature taxonomy. Tools like ChatGPT or Perplexity are an easy way to start. You can just ask the AI to scrape the website and build the taxonomy. The challenge is that they’ll crawl the marketing site, which will basically mirror the picture you’ve already painted. It’s helpful, but it doesn’t allow you look from the outside in.
To see what your users see, you can use a computer control model like OpenAI Operator or Manus to create an agent that will make an account and login to the product. I used Manus to create a feature taxonomy for Calendly to show how it works.
The full prompt is below. Replace “[Enter site URL here]” with the URL of the site you want to crawl. By default, Manus will create a fake email address to make an account, but this won’t work since you can’t verify the email. Instead, create a temporary, disposable email from a tool like AdGuard or Temp-Mail and include it in the prompt. We do not recommend sharing a real account with these tools.
Manus will mostly handle this on its own, but you may have to step in to handle a reCAPTCHA form. It’ll ask you to take over the screen, then it’ll pick up where it left off.
Here’s the example output from Manus.
<role>
Your mission: **map every customer‑facing capability of the target software product and deliver a structured feature‑taxonomy with clear descriptions and use‑cases.**
</role>
<target_product>
URL: [Enter site URL here] Auth: If a free trial or demo account is available, self‑register with this email address: [example@example.com]. If no trial exists, explore public marketing pages, docs, videos, and help articles.
</target_product>
<high‑level_objectives>
**Explore the product end‑to‑end**
– navigate all menus, settings, and flows a typical user can access.
**Capture every distinct capability**
– note names, UI labels, and where each feature lives in the interface.
**Synthesize a feature‑taxonomy**
– organize capabilities into a 3‑tier hierarchy:
* Module (top‑level area)
* Feature (customer‑valuable function)
* Sub‑feature / Action (granular tasks or settings)
**Describe & contextualize**
– for every (sub‑)feature, write:
* Clear 2–4‑sentence description (what it does & why it matters)
* Key use‑case(s) in bullet form (when a real user would rely on it)
Create a simple table to show the hierarchy of modules and their associated features.
</high‑level_objectives>
<detailed_instructions>
A. **Initial reconnaissance**
Land on the home page ➜ read nav labels & hero copy.
Crawl public pages: “Features”, “Pricing”, “Docs”, “Help Center”, “Blog”.
Build an initial list of modules/features mentioned.
B. **Hands‑on exploration (if trial/demo available)**
Create a trial account using the email address I provided.
Complete onboarding flows and note each screen.
Open every primary navigation item; within each, click all tabs, buttons, or expandable menus.
For complex widgets (dashboards, editors, builders) open tooltips, settings, and advanced menus.
C. **Evidence capture**
* For any non‑obvious feature name, hover/click tooltips and note the explanation.
* If a feature references an external doc page, open it and skim the first two paragraphs for intent.
D. **Avoid these traps**
* Ignore marketing hype; focus on concrete capabilities users can directly access.
* Do not include purely internal admin tools unless visible to standard users.
E. **Quality checks before finishing**
* Every feature must belong to exactly one module.
* Descriptions should start with an **active verb** (“Create…”, “Automate…”, “Visualize…”).
* Use‑cases use customer language (“A support manager…” not “the user”).
</detailed_instructions>
<output_format>
Return **one document** that includes the following:
* Product Name
* Url
* Table
* Module Name
* Module Description
* All feature names and feature descriptions that tie to that module
* All sub-feature names, descriptions, and use cases
Do that for all Modules and Features
</output_format>
<termination_condition>
When all primary navigation areas and linked docs have been inspected **or** 30 minutes of wall‑clock time have elapsed—whichever comes first—stop and output the taxonomy immediately
</termination_condition>
You can do this for your own product, but also competitors. And once you’ve run several, you could ask AI to create a feature comparison chart to easily see how your product stacks up against the rest of the market. You can also feed this taxonomy into a product strategy, which brings us to the next prompt.
Prompt #2: Turn a product strategy into a compelling narrative to sell your vision
Most product strategies struggle not because they're wrong, but because they're boring. In the same way your marketing and sales teams need to sell the product, you need to sell the product strategy.
This prompt will help you turn a tightly structured product strategy into a narrative that will galvanize the rest of your team around the vision. The goal here is to make easy to get buy-in for your strategy. A product strategy needs a market analysis, value-prop statement and a feature taxonomy (see above!) but this alone will not help you sell the vision.
I’ve used elements of Andy Raskin’s strategic narrative framework to build this prompt, but there are other narrative structures you can use.
There are a few inputs you’ll need:
**Product Strategy Outline**** (Required):** An outline of the core components of your strategy. The problem you solve, target audience, value prop, business model, growth strategy, competitive advantage, and your core unique insight. Reforge Learning members can get a template and see real strategy documents here.**Target Audience**** (Required):** Who the audience is, a description of buyers vs. end users. Provide the most detailed description possible.**Example Narratives**** (Optional):** Provide example narratives that you like. This isn’t required because I find that it does an 80% job if you just give it the structure of the narrative provided in the prompt.
You can use Claude or ChatGPT to run this prompt. If you use Claude Projects then I suggest you put the below Prompt into project instructions. Put Example Narratives as Project Knowledge. Upload the Product Strategy Outline as an attachment to the chat. That way you can use the project and upload different variations of your Product Strategy as you iterate. You can also use the
A common complaint about AI tools is that they spit out slop, but this is mostly user error. If you want a document that looks like a product strategy, ChatGPT will create one in about 10 seconds. A PM who tells their team “build something users want” won’t get good results either. Better input yields better output.
Product managers are generally early adopters. Still, seeing how others use AI can inspire their own usage. I put together some example prompts as part of our Spring Cohort that I’ll share here today. They will help you:
Create a feature taxonomy for your own product or a competitor’s.
Turn a product strategy into a compelling narrative to sell your vision.
Frame every PRD within the context of that strategy.
Evaluate your PRDs from a CPO’s perspective
I put together example inputs and outputs for each as well, so you can see how they work and adapt as needed for your own work.
The formula for great AI workflows
Stepping back just briefly, my hope is that you can use these prompts right away, but also that they encourage you to think more deeply about how to use AI in your work. There’s a framework I used to create each of these that can help you find new ways to use AI.
First, you need to understand AI’s capabilities. This is always changing (and improving) and you have to use lots of tools to understand AI’s current strengths and weaknesses. We’ll use a handful of different tools to run the prompts below based on the needs and constraints of each.
Next is task decomposition. When you break a task down into smaller components, it helps you think through the steps and it’ll help you get much better outputs from AI. In general, the bigger the task, the more generic the output will be. And the more narrow the task, the more specific the output will be. As an example, you should separate the creation of PRD from the evaluation of it. The first part requires creativity and narrative. The second part requires critical analysis. Combining this into a single workflow would water down the results.
And lastly, bring conviction, opinions, subject matter expertise, data and non-consensus thinking. You have to fuel the workflow with your own insights to get good results. This is where AI really shines—helping you turn your ideas into something tangible you can advocate for, build and sell. (How you find those unique insights is outside the scope of this article but we have a course on it called Mastering Product Management.)
With that foundation in mind, let’s get into a few prompts that can help you in some of the important work you do as a PM.
Prompt #1: Create a feature taxonomy (from your customer’s eyes)
It’s always helpful to understand your own product from a user’s point of view. One way to do this is to create a feature taxonomy. Tools like ChatGPT or Perplexity are an easy way to start. You can just ask the AI to scrape the website and build the taxonomy. The challenge is that they’ll crawl the marketing site, which will basically mirror the picture you’ve already painted. It’s helpful, but it doesn’t allow you look from the outside in.
To see what your users see, you can use a computer control model like OpenAI Operator or Manus to create an agent that will make an account and login to the product. I used Manus to create a feature taxonomy for Calendly to show how it works.
The full prompt is below. Replace “[Enter site URL here]” with the URL of the site you want to crawl. By default, Manus will create a fake email address to make an account, but this won’t work since you can’t verify the email. Instead, create a temporary, disposable email from a tool like AdGuard or Temp-Mail and include it in the prompt. We do not recommend sharing a real account with these tools.
Manus will mostly handle this on its own, but you may have to step in to handle a reCAPTCHA form. It’ll ask you to take over the screen, then it’ll pick up where it left off.
Here’s the example output from Manus.
<role>
Your mission: **map every customer‑facing capability of the target software product and deliver a structured feature‑taxonomy with clear descriptions and use‑cases.**
</role>
<target_product>
URL: [Enter site URL here] Auth: If a free trial or demo account is available, self‑register with this email address: [example@example.com]. If no trial exists, explore public marketing pages, docs, videos, and help articles.
</target_product>
<high‑level_objectives>
**Explore the product end‑to‑end**
– navigate all menus, settings, and flows a typical user can access.
**Capture every distinct capability**
– note names, UI labels, and where each feature lives in the interface.
**Synthesize a feature‑taxonomy**
– organize capabilities into a 3‑tier hierarchy:
* Module (top‑level area)
* Feature (customer‑valuable function)
* Sub‑feature / Action (granular tasks or settings)
**Describe & contextualize**
– for every (sub‑)feature, write:
* Clear 2–4‑sentence description (what it does & why it matters)
* Key use‑case(s) in bullet form (when a real user would rely on it)
Create a simple table to show the hierarchy of modules and their associated features.
</high‑level_objectives>
<detailed_instructions>
A. **Initial reconnaissance**
Land on the home page ➜ read nav labels & hero copy.
Crawl public pages: “Features”, “Pricing”, “Docs”, “Help Center”, “Blog”.
Build an initial list of modules/features mentioned.
B. **Hands‑on exploration (if trial/demo available)**
Create a trial account using the email address I provided.
Complete onboarding flows and note each screen.
Open every primary navigation item; within each, click all tabs, buttons, or expandable menus.
For complex widgets (dashboards, editors, builders) open tooltips, settings, and advanced menus.
C. **Evidence capture**
* For any non‑obvious feature name, hover/click tooltips and note the explanation.
* If a feature references an external doc page, open it and skim the first two paragraphs for intent.
D. **Avoid these traps**
* Ignore marketing hype; focus on concrete capabilities users can directly access.
* Do not include purely internal admin tools unless visible to standard users.
E. **Quality checks before finishing**
* Every feature must belong to exactly one module.
* Descriptions should start with an **active verb** (“Create…”, “Automate…”, “Visualize…”).
* Use‑cases use customer language (“A support manager…” not “the user”).
</detailed_instructions>
<output_format>
Return **one document** that includes the following:
* Product Name
* Url
* Table
* Module Name
* Module Description
* All feature names and feature descriptions that tie to that module
* All sub-feature names, descriptions, and use cases
Do that for all Modules and Features
</output_format>
<termination_condition>
When all primary navigation areas and linked docs have been inspected **or** 30 minutes of wall‑clock time have elapsed—whichever comes first—stop and output the taxonomy immediately
</termination_condition>
You can do this for your own product, but also competitors. And once you’ve run several, you could ask AI to create a feature comparison chart to easily see how your product stacks up against the rest of the market. You can also feed this taxonomy into a product strategy, which brings us to the next prompt.
Prompt #2: Turn a product strategy into a compelling narrative to sell your vision
Most product strategies struggle not because they're wrong, but because they're boring. In the same way your marketing and sales teams need to sell the product, you need to sell the product strategy.
This prompt will help you turn a tightly structured product strategy into a narrative that will galvanize the rest of your team around the vision. The goal here is to make easy to get buy-in for your strategy. A product strategy needs a market analysis, value-prop statement and a feature taxonomy (see above!) but this alone will not help you sell the vision.
I’ve used elements of Andy Raskin’s strategic narrative framework to build this prompt, but there are other narrative structures you can use.
There are a few inputs you’ll need:
**Product Strategy Outline**** (Required):** An outline of the core components of your strategy. The problem you solve, target audience, value prop, business model, growth strategy, competitive advantage, and your core unique insight. Reforge Learning members can get a template and see real strategy documents here.**Target Audience**** (Required):** Who the audience is, a description of buyers vs. end users. Provide the most detailed description possible.**Example Narratives**** (Optional):** Provide example narratives that you like. This isn’t required because I find that it does an 80% job if you just give it the structure of the narrative provided in the prompt.
You can use Claude or ChatGPT to run this prompt. If you use Claude Projects then I suggest you put the below Prompt into project instructions. Put Example Narratives as Project Knowledge. Upload the Product Strategy Outline as an attachment to the chat. That way you can use the project and upload different variations of your Product Strategy as you iterate. You can also use the