Skip to main content

We’re witnessing a massive gold rush! With the advent of ChatGPT, the software world has become flooded with new products that promise to solve a million different problems using this fascinating AI technology.

If you are new to the “product manager AI” space and are wondering how you can take advantage of ChatGPT, then you have come to the right place. In this guide, we will learn to “talk to” ChatGPT and explore several use cases of integrating ChatGPT into your product or using it as your personal assistant.

Prelude: How AI Product Managers “Talk” To ChatGPT.

Unlike its predecessor, GPT-4 and other machine learning tech for product managers, ChatGPT has become vastly better at conducting conversations and understanding the casual text. This means that you can simply ask it to do something, just as you would ask a human.

This way of “talking to” ChatGPT will surely give you good enough results, but you can unlock its real capabilities only if you start following OpenAI’s best practices of constructing a prompt for this artificial intelligence model.

Author's Tip

A prompt is your request/command (in the form of a text) that you are giving to ChatGPT to process.

Now let me give you a couple of quick tips that will help you increase the quality of output that ChatGPT will hand out to you.

Give Clear Instructions: ChatGPT is quite good at reading and understanding requirements and instructions. We’re product managers who have hands-on experience writing clear and concise requirements, so this should be a piece of cake for us.

Structure your prompt: We usually use structured text like bullet points and headers to make text human-readable. Well, this trick works well for ChatGPT, too as it tends to understand the structured text better.

Show the desired format of the output: If you want to use the output of ChatGPT in your product as part of a feature, then you need to make sure that this output has a predictable structure. The reason is that you will need to run Regex on this output and extract useful information from it.

You can simply include an example of the desired output in your prompt to achieve a predictable output format.

Don’t ask for any recent information: ChatGPT is trained using internet datasets as of Fall 2021. So, if you ask about the events of last week, it will not give you an answer.

OpenAI did add support for browsing the internet for up-to-date data, but they froze it since July 3 as it was not behaving as it was supposed to. Luckily, the recent release of ChatGPT plugins can help you with this.

For instance, you can use KeyMate.AI Search plugin to enable web browsing for your ChatGPT instance. However, please note that you will need to sign up for this plugin and it is a paid one with limited free functionality.

Therefore, until OpenAI brings browsing back, be careful with any requests that include data from the last 1.5 years.

Setting Up Custom Instructions

Custom instructions are another recent update that can significantly increase the quality of output that you are getting from ChatGPT.

In general, the more context you provide to the model, the better it can fulfill your requests. The context in question usually includes instructions like:

  • The tone (business, silly, playful, etc.).
  • The role that the model can assume (act like a Senior PM).
  • The amount of content it should generate (keep it less than 400 words).
  • If it should pay attention to details or examples, etc.

Before custom instructions, you would have to retype this context every time you create a new chat session. Now, however, you simply add all of this into the appropriate section, and this context automatically applies to all chat sessions.

For instance, right now, the custom instructions section for my ChatGPT account looks like this.

As you can see, I have told ChatGPT who I am and what kind of product I am working on now. Moreover, I have stated what I am expecting to get from the model and given detailed instructions on how to answer my questions, including size, the inclusion of examples, etc.

Discover how to lead product teams and build better products.

Discover how to lead product teams and build better products.

  • No spam, just quality content. Your inbox is safe with us. For more details, review our Privacy Policy. We're protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
  • This field is for validation purposes and should be left unchanged.

How Can You Use ChatGPT To Enhance Your Products?

With OpenAI’s recent launch of the ChatGPT API (and more recently, Google DeepMind's release of Gemini), we are finally able to integrate this language model into our products and take advantage of its impressive capabilities.

But how can we use ChatGPT in our products? Well, let me show you a couple of examples of compelling new features and initiatives that you can add to your product development roadmap with the help of this technology.

1. Online Shops: Let ChatGPT Suggest Products to Your Customers Based on Their Buying Behavior.

One of the great capabilities of ChatGPT is to understand contextual information and give you an answer while keeping that context in mind.

This ability of ChatGPT becomes quite handy when you want it to handle a classification task such as determining the best products or product categories to suggest to your users based on the purchases that they have made in your online shop.

I know, you can suggest items with ordinary code/logic as well - by looking at the last purchase of your user, then suggesting something that the majority of others have bought after purchasing that same item. But there are two problems here:

  1. It will take you a lot of development time and money to create such logic by yourself, or you will have to buy a suggestion tool/addon for your website.
  2. Ordinary logic/code is unable to take buying behavior and context into account and will simply suggest whatever is popular.

ChatGPT, unlike ordinary code, is immune to these issues. Making an API call and asking to suggest products will cost you a fraction of a cent or less and it is just great at understanding context.

Now, let me show you an example of a prompt that you can use to ask ChatGPT to suggest products.

screenshot of example of a prompt ask chatgpt

If we look at the prompt that I have given to ChatGPT, we will see that it is following the best practices that we went over in the prelude section of this guide. We are very clear with our requirements here and add structure to our prompt.

But this prompt was simply a demonstration of the use case and your real prompt would look something different.

Let’s not forget that it will not be us manually typing this prompt in the ChatGPT UI interface. Instead, we will need to write a tiny piece of code in our eCommerce application to automatically generate these prompts for each customer and send an API call to ChatGPT.

Therefore, we need to create a prompt template where we can use variables (parts of the prompt that will change depending on the information you have for a specific customer).

Here’s what your prompt template with variables would look like.

screenshot of prompt template with variables

Each of your customers has bought different items recently, the color they prefer (the most common product color in their purchase history), and the amount they have spent at your shop is different too.

Therefore, we will need to create different prompts for each of them, including this information in their individual prompts, and send out API calls to ChatGPT in order to get their personalized recommendations.

Another component of the prompt that I have changed here (compared to the demo one before) is including the list of products or categories in your prompt. This way, you can make sure that ChatGPT is recommending something that you definitely have in your shop.

If we did not ask it to use this list for the recommendation, it could have returned items or categories that do not exist in your store.

As soon as you get your list of recommendations, the list of features that you can build upon it is endless! Why not send them weekly email digests with your suggestions? How about showing these items on your homepage when they open it?

2. Classifieds Website: Help Your Users Improve Their Classified Content with ChatGPT

GPT-4, the deep learning algorithm that ChatGPT runs on, is a generative AI model. Therefore, the task that it does best is either generating or rewriting content.

You can put this skill of ChatGPT to good use if you are managing a classified website by rewriting and improving the post content that your users have posted.

The prompt, in this case can be quite a simple one. Just pick up the classified ad content of your user, and add a request underneath asking ChatGPT to rewrite it in a way that would make it more likely to sell. Here’s what it would look like for an apartment in Tuscumbia, Alabama.

screenshot of simple one prompt

If you have noticed, we have followed another best practice and asked ChatGPT to return its response in a specific format, making it easy for us to process the response text by detecting the term “Effective paragraph suggestion:” using RegEx and storing the content coming after it.

Here’s what the response looks like.

screenshot of regrex response

One more note. The reason I have explicitly asked ChatGPT to make suggestions on a paragraph level instead of rewriting the entire thing is that we do not know the length of the content in a user’s classified.

It might be long and some parts of it might be well-written. So, instead of rewriting the entire thing, we can suggest improving only the paragraphs where the content is sub-par.

Now let me talk about the benefits you get from implementing such a feature (apart from it being very cheap to make as ChatGPT does the heavy lifting part for you).

From the point of view of the users who post classifieds, you add to the core value they get from your website - getting their stuff sold. ChatGPT will not simply rewrite it to make the description look good. We have explicitly asked it to take sales data into consideration and write content that increases the chances of the item getting sold.

From the point of view of your business, you can avoid churning customers who wanted to add a classified on your website, but were not sure if they could write a proper product description. With this feature, you can give them confidence that, even if they have no experience in writing classified content, they will still have a great product description for the item they are selling.

3. Database Tool: Let Users Run Queries By Typing What They Want To Find

It was a pleasant surprise for me to find out that the language model of ChatGPT was capable of understanding anything that you would classify as a language, including programming and database languages.

So, let’s take advantage of this and add a cool new feature to our database management tool. For this example, imagine you’re the product manager of DataGrip.

What if we gave our users (who are most-likely data scientists) a small search-like field where they could type the information they wanted to retrieve from the database using plain natural language and we found and gave it to them? Sounds cool, right?

Well, we can use ChatGPT to get it done. First of all, let’s understand how this feature should work. In order for us to retrieve data from the user database, we need to run an SQL query (assuming the database is SQL).

Therefore, the problem that we need to solve using ChatGPT is converting the requirements written in natural language into an SQL query.

Luckily, our language model is more than capable of doing that. (Yes, it knows data science, too!)

Here’s what the prompt would look like for this feature.

screenshot of sql query

And here is the result that the language model will give to us.

screenshot of result sql query

If you have noticed, I have not just copied the natural language request of our user to the prompt but also added the structure of the user’s database as context. 

If we did not give ChatGPT this context, it would create an SQL query where the names of the fields and other terms would be different than what we had used in our tables and the SQL query would simply not run on our database.

Therefore, your template for prompting ChatGPT would include two variables (user’s request and their database schema) and look like this.

screenshot of template for prompting ChatGPT

As we can see, the applications of ChatGPT as an engine powering your next cool feature are endless. Thanks to its ability to understand context, ChatGPT can serve as an advanced suggestion tool, an AI content rewriter, as well as a human text-to-code translator.

In reality, these three examples are barely scratching the surface of the capabilities of ChatGPT. But let’s leave listing other applications of this technology for another day and move on to understanding how it can help you complete your daily tasks.

How Can ChatGPT Help With Your Day-to-Day?

Apart from serving as a powerful AI engine for your product, ChatGPT can also become the assistant that can handle some of your dull and time-consuming tasks and let you focus on more impactful ones.

You will be amazed just how diverse its capabilities are as a personal assistant. Let me show you a couple of examples.

1. Convert Your Acceptance Criteria To Gherkin Format

Gherkin is a standard for writing acceptance criteria in a very structured way. The most common application of this standard is when you follow the BDD philosophy and write automated tests based on Cucumber.

We used to do this for one of my products in the past. Although I admire the structure of Gherkin (it makes acceptance criteria super-clear and easy to understand), writing with this standard was a very time-consuming task that I hated doing manually.

I wish I had ChatGPT back then because I could have given my ordinary acceptance criteria to the language model and asked to convert it to Gherkin.

screenshot of gherkin format

Please note that it is important to add the requirement of writing separate scenarios in the prompt above. If you don’t, ChatGPT will give you a single bloated Gherkin scenario that includes all of the actions in your acceptance criteria.

But if you ask it for separate scenarios, you get this neat response.

screenshot of gherkin format separate scenario

The example I have used here is about formatting acceptance criteria with the Gherkin standard, but you can ask ChatGPT to rewrite content in any format you want. You can even show ChatGPT what the templates of various documents look like in your company and then ask it to create content following the format in those templates.

2. Write User Stories For Basic Features

Product managers are divided into two categories—those who love writing stories and PRDs, and those who absolutely hate it.

Let’s admit that the majority of PMs (including me) belong to the second category.

Writing stories is especially dull if you are writing requirements for something that is very standard in the industry—like a delete feature or changing email.

Lucky for you, ChatGPT knows how to write user stories and it is especially good at providing requirements for well-known features. Let’s test it out and ask for a story along with acceptance criteria for a Google Oauth2 signup.

We know that ChatGPT will get you better results if you are more specific with your prompt. However, in this case, the feature of Google Signup is so standard and well-known that a primitive prompt like this can produce good-enough results. (Though, let's be clear, there are also situations when you shouldn't use AI to write user stories for you.)

screenshot of primitive prompt

There you go, a nice-looking story with nine acceptance criteria covering the main cases and flows of signing up with Google! ChatGPT has just saved you from completing a very dull task!

3. Create Questionnaires For Product Interviews

When doing product interviews, the questions you ask will determine the quality of the answers you get.

For instance, it is best practice to ask about the past experiences of the interviewees instead of letting them imagine and talk about their future. The reason is that the experiences in the past are hard facts that you can rely on, while the talk about their future might consist of desires that will never materialize.

Imagine you want to learn about somebody’s gym attendance habits. If you ask them how many times a week they would want to go to the gym if they bought a subscription, the answer would be every day. Let’s rephrase that question and ask for the number of times they have attended the gym during the last month. You would be surprised to hear that it was barely once a week.

Why are these answers so different from each other? The first one is a desire, the second one is an actual fact.

You might be surprised to learn that ChatGPT knows about these best practices and is capable of generating a decent questionnaire that you can use for your interviews.

screenshot of generating a decent questionnaire

Don’t forget that ChatGPT understands context! So, you can tell it about the users you are interviewing and the goals that you want to accomplish and the model will create a list of questions that is relevant to your specific case.

screenshot of sample questionnaire

To be honest, I am pretty impressed with the list above. This is very close to the questionnaire that I would have created for such an interview.

Also, kudos to the OpenAI team for giving ChatGPT a friendly character. I love the fact that it wishes you good luck with product discovery after generating the list.

It’s a Machine Learning Revolution!

We knew that OpenAI is really good at building AI models, and the previous iterations of GPT were quite impressive too, but nobody expected ChatGPT to be this smart and capable!

This new technology is in fact so powerful, that it has democratized AI and kickstarted a revolution in computer science.

While ChatGPT can significantly improve your products, it will not guarantee their success. To achieve success, you will also need to have:

  • Knowledge of how to be an AI PM along with a solid understanding of the fundamentals of managing AI teams, software development teams working on AI projects, as well as the lifecycle of SaaS ML products (from ideation and decision-making to launch and support).
  • Great Agile Processes with established workflows and roles for the product owner, software engineering team members, data engineers, software product design and user experience team members, and others.
  • Forward-looking Product Roadmap to Impress your product team and stakeholders.
  • And a real-world plan along with the right metrics for reaching a Product Market Fit for your startup.

Don't miss out on new content! Subscribe to our newsletter.

By Suren Karapetyan

Suren Karapetyan, MBA, is a senior product manager focused on AI-driven SaaS products. He thrives in the fast-paced world of early stage startups and finds the product-market fit for them. His portfolio is quite diverse, ranging from background noise cancellation tools for work-from-home folks to customs clearance software for government agencies.