8 ways I am using AI to help me be a better product manager (and 4 ways I am not)

Working in an AI startup has its benefits. Chief among them is that I have spent a lot of time working with a small but talented group of people who have a great understanding of AI technology. I have been trying hard to absorb as much of that knowledge as I can along the way and to find ways to incorporate AI into some of my product management practices. This post summarises some of what I have learned. 

Before we start, here is Copilot Designer‘s response to the prompt:

Generate an image of Alan Wright from Climate Policy Radar and AlanIsWright.com using AI to help him do product management work

Copilot thinks I am an old man who likes to play with toy aeroplanes

8 ways I am using AI to help me be a better product manager

1 – Postcards from future users & personas

Taking inspiration from Working Backwards, one of my preferred methods for articulating a vision for products and features is through using postcards from future users. These are brief statements, penned by imagined users months or years into the future, illustrating in their own words how our tools have been of help to them. The aim is to help the team align on a shared vision of the impact their work has on our target users. These statements should be written in the style of the persona, be engaging to read, and help generate excitement within the team.

Creative and inspiring writing is not my strong suit. My attempts at writing this content myself were not only time-consuming but also a bit dull for others to read. However, creative writing is something that generative AI models like ChatGPT excel at. Here is a prompt that I have utilised to generate postcards from future users:

You are <INSERT PERSONA NAME>. Write a postcard explaining how this new feature from Climate Policy Radar has made your job easier. <INSERT DESCRIPTION OF NEW FEATURE AND WAYS THAT IT HELPS THE USER>”

Information about <INSERT PERSONA NAME> is below:

<INSERT INFORMATION ABOUT PERSONA>


The first drafts of the personas were also created using AI tools. Note that these needed far more editing than the postcards from the future, as they had a tendency to include details that didn’t align with what I had observed in user research. However, the writing process was far quicker and better than if I had done it myself. Here is an example prompt:

You are an expert in user research. Write a persona using the information below. Your answer should be under 4 headings: motivation, goals, pain points, tools.

<INSERT INFORMATION ABOUT PERSONA> 

2 – Solution ideas

Generative AI tools excel at generating ideas. Below is a prompt I like to use when first starting to ideate how he might solve problems. I ask for the URL so it is easy for me to check it out myself:

You are an expert product manager. Our product is <INSERT PRODUCT DESCRIPTION HERE>

What are the best solutions that allow a user to feed back on the quality of content in a digital product? For each example, list the URLs of some popular products that use that solution.


3 – Risky assumptions

Premortems help us identify and reduce risks before they happen. But they can be time-consuming to do, and too often get deprioritized in favour of delivery.

Increasingly, I am using ChatGPT to do premortems for me. ChatGPT has been trained on a huge volume of literature. In amongst that is a whole bunch of content from other product teams about how and why features have gone wrong for them in the past.

Below is a prompt I like to use to help me identify the types of risks we need to tackle through product discovery:

You are an expert product manager. We are building a feature that enables users to thumbs up or thumbs down a search result. Tell me 20 ways in which this feature could fail. Which are the highest impact and most likely to occur?

4 – Product requirements

There is a wealth of useful data on the internet about how others have previously solved problems similar to ours. So when drafting requirements for more generic new features, I try to avoid reinventing the wheel and utilise prompts like the one below:

You are an expert product manager. Write a product requirements document for a feature that enables a user to give a thumbs up or thumbs down response to a search result.

Less is more in product requirements. One of the arts of good product management is being able to distil long lists of requirements down to the smallest number expressed in the simplest way. This enables the team to deliver value at pace. I like to take what ChatGPT gives me as a first draft. This is much faster than me trying to draft it from scratch, and ChatGPT usually remembers things that I would have forgotten. But this is always followed by me hacking away at unnecessary requirements or detail.

After I have a first draft, I like to follow up with the question: 

Tell me 20 weaknesses of the above product requirements document. 

This helps me spot any areas which might have been missed.

5 – Rewriting content

It might surprise readers of this blog to learn that I am not the strongest user of the English language. I have always struggled with grammar – my unedited content has upset many a communications person in roles past and present!

Rather than going back to school and learning how grammar works, I am using AI tools to double-check important content that I write.

For years, Grammarly has been my go-to tool for proofreading. I like that it offers suggestions that I can accept and reject on a case-by-case basis. I dislike that a lot of its features are behind a paywall, yet these features are visually prominent in the user interface and hard to ignore.

After watching our founder use ChatGPT to proofread and improve content that had already been checked by Grammarly, I am increasingly using ChatGPT instead. Overall I have found the results to be better, probably in part because I have a GPT4 API key while I only have the free subscription to Grammarly. 

A disadvantage of using ChatGPT is that there is always a risk of it changing the meaning or context of content without me realising it, especially as changes can’t be explicitly tracked and accepted. So proofreading is essential. But then again, proofreading is always essential, so it doesn’t add any extra stages to the process.

The screenshot below shows what this section of the blog looked like before I used GPT to fix it. See if you can spot the difference.

I use the GPT for Sheets and Docs plugin alongside my GPT4 API key in Google Docs and Sheets

6 – Transcribing meetings

Voice recognition technology has seen significant improvements in recent years. Two products that I have used to transcribe meetings are Read.AI and Otter.AI. Both are accurate and can be really useful to quickly read through the notes from important meetings that I wasn’t able to attend or to take detailed notes during user interviews where I don’t have a note-taker present.

Recording voice notes is another big benefit from the increase in voice recognition and transcription accuracy. I am alternating between AudioPen and Google Voice Recorder for this. It’s a handy way to be able to take long notes on the go.

7 – Writing SQL / Python code

AI has made it far easier for me to do simple technical tasks myself that I might previously have had to ask an engineer or data scientist to help with or spent hours trying to figure out. I use Cursor as my IDE, and combined with my GPT4 API key, it writes and reviews code based on natural language prompts. This has enabled me to do things like find duplicates in our dataset, assess our data quality, analyse survey results, and write queries to understand how new features are performing – without distracting the team. The faster progress that AI has enabled has also kept me motivated to keep developing my technical skills.

Using GPT4 in the Cursor IDE to to help me write some Python

8 – As a general replacement for search engines

Search engines like Bing and Google are great for finding web pages that answer my questions. These are still my go-to tools for when I want to do a thorough search and evaluate all the source content myself.

Through featured snippets and knowledge panels, search engines also try to surface content from the highest-ranking web pages that they think answer my questions. However, this is a relatively small part of the search results – there are usually a lot of paid ads competing for my attention.

Increasingly, I am using Perplexity instead of traditional search engines for four reasons:

  • It better understands the questions I am asking by asking clarifying questions when necessary.
  • It is up to date. It searches the internet and summarises key information from the top-ranked articles.
  • It cites the sources it uses, making it easy for me to verify results or do more of my research.
  • The user interface is clear and uncluttered.
An example of a Perplexity response. Most of the time, this is more useful than what traditional search engines provide

4 ways I am not using AI

1 – To do anything more than a draft

AI models can sometimes make mistakes, misunderstand context, and fabricate content. We have all seen the pictures of people with six fingers and three legs, or heard about cases like the lawyers who got fined for submitting fake citations they were given by ChatGPT.

I don’t know who said this first, but there’s a good quote going around about AI being like having 1,000 interns. I like this analogy because it illustrates the huge potential to get work done to a fairly decent standard. But as with interns, with AI, it is wise to double-check its work. This usually means:

  • Edit it down a bit to focus on the most important details. AI can be verbose. Saying the same thing with fewer words makes it easier for people to understand each other.
  • Spot any areas that might have been missed or underemphasized. This is where expert knowledge comes in. With 15 years of experience managing digital products, I have a good sense of when AI is saying something that isn’t quite right about product management. I am much less capable of challenging the outputs of an AI that is writing a Python query.
  • Check the source data myself for anything that doesn’t sound quite right. This is why I like tools like Perplexity that make it easy for me to access the source data.

2 – Writing notes for important meetings

Tools like Read.AI, Otter.AI, and Notion all offer meeting summarization functionality. These can be useful for getting a sense of what was discussed. Despite this, I still summarise important meetings the old-fashioned way for two main reasons:

  • AI summaries have a harder time than I do of distinguishing signal from noise. All too often they overemphasise irrelevant information and play down or ignore completely the important details. They also tend to be verbose, and I have seen them get important details wrong.
  • I often use the process of writing meeting notes as a way to facilitate meetings. The process of recording actions, decisions, and questions is a collaborative one that shapes the meeting when done in real time.

For low-stakes meetings where I wouldn’t usually take notes or am unable to attend at all, summaries are better than nothing. But they are not yet at the stage where they add much value for me otherwise.

3 – Analysing user research findings

I have seen a few blog posts from people who have been using AI to summarise the results from user interviews and usability tests. This isn’t something I have found much value in so far. Partly this is because of the first point above: there is too much noise in the summary and important details are often missing. It is also because:

  • A lot of the value I get from the research process is in the synthesis. Most user research conversations I have are immediately followed by a chat with the other team members involved. AI summaries often get generated too late to be used for this conversation. And even when they are generated in time, I haven’t seen them add that much value.
  • A lot of the insights from user research are not captured in transcripts. Facial expressions, pauses, tone of voice, and whatever the user is sharing on their screen all contain useful information which AI summaries will miss.

4 – Writing blog posts

Don’t worry, dear reader. This blog post is an original man-made work of art from me (with only a little bit of help from ChatGPT to put capital letters and hyphens in the right place).

In case you’re wondering if AI could have done better. below is an AI-generated version of this post and the prompt I used to make it.

You are an expert product manager. Write a blog post about how AI can help product managers to be better at product management. Write in the writing style of alaniswright.com

Title: Harnessing the Power of AI in Product Management: A Game-Changer for Product Managers

Product management is an ever-evolving discipline that requires constant adaptability and learning. As the technology landscape continues to shift, product managers need to stay ahead of the curve, leveraging the latest advancements to streamline their processes, enhance product development, and deliver superior customer experiences. One such technological revolution that is transforming the product management landscape is Artificial Intelligence (AI).

AI has become an integral part of our everyday lives, from virtual assistants like Siri and Alexa to recommendation engines on Netflix and Amazon. But its potential extends far beyond these consumer-facing applications. AI has the power to revolutionize product management, helping product managers become more efficient, strategic, and customer-centric. Let’s delve into how.

1. Enhanced Decision-Making

AI can process vast amounts of data in real-time, providing product managers with valuable insights that would otherwise take weeks or months to compile manually. AI-driven analytics can identify trends, predict customer behavior, and highlight potential issues before they become significant problems. This data-driven approach empowers product managers to make more informed decisions, reducing risk and driving product success.

2. Streamlined Processes

AI can automate routine tasks, freeing up product managers to focus on more strategic aspects of their role. From tracking product performance metrics to managing stakeholder communication, AI can handle a wide range of administrative tasks, increasing efficiency and productivity. It also reduces the risk of human error, ensuring more accurate and reliable outcomes.

3. Personalized Customer Experiences

AI’s ability to analyze vast amounts of data also enables it to deliver highly personalized experiences. By understanding individual user behavior, preferences, and needs, AI can help product managers tailor products to meet the unique needs of each customer. This level of personalization can significantly enhance customer satisfaction and loyalty, driving long-term product success.

4. Predictive Analytics

AI’s predictive capabilities are a game-changer for product managers. By analyzing historical data, AI can predict future trends, customer behavior, and potential issues. This foresight allows product managers to proactively address potential problems, seize opportunities, and stay ahead of the competition.

5. Improved Product Testing

AI can also streamline product testing, using machine learning algorithms to identify and correct bugs more quickly and efficiently than traditional methods. This not only speeds up the testing process but also ensures a higher-quality end product.

6. Enhanced Collaboration

AI can facilitate better collaboration among teams by providing a centralized platform for sharing information, tracking progress, and managing tasks. This can improve coordination, reduce misunderstandings, and ensure everyone is on the same page.

In conclusion, AI is a powerful tool that can significantly enhance product management. By leveraging AI, product managers can make more informed decisions, streamline their processes, deliver personalized customer experiences, and stay ahead of the competition. However, as with any technology, it’s essential to use AI responsibly, ensuring it aligns with your product strategy and adds value to your customers.

Embracing AI in product management is no longer an option; it’s a necessity. As product managers, let’s harness the power of AI to drive product success and deliver exceptional customer experiences. After all, in the world of product management, staying ahead of the curve is not just about survival; it’s about thriving in an ever-changing landscape. 

End notes

Thanks for reading! I want to finish by saying a big thanks to Henry Franks, Michal Nachmany, Kalyan Dutia, Harrison Pim, and Matyas Juhasz. Many of the techniques I have mentioned in this post have been born from conversations with you all!

0 replies

Leave a Reply

Want to join the discussion?
Feel free to contribute!

Leave a Reply

Your email address will not be published. Required fields are marked *