UXPrompts.ai

SUMMARY

I developed UXPrompts.ai, a prompt library to integrate Large Language Models into the UX process, focusing on crafting prompts and transforming it into a digital product. Throughout this journey, I explored the challenges and learning experiences of using generative AI for product development, marketing, and user engagement.

Well, much has changed in these past few years. The quality of the LLMs has improved exponentially. Large Language Models (LLMs) have evolved from text completion to powerful chatbots that are able to execute code, use tools, access external knowledge and search the web.  Also, our ability to operate them in a meaningful way has also increased, through better understanding of prompt engineering and capabilities of the toolset.  A capable prompt engineer can create prompts that deliver a variety of different and useful outputs. The better the prompt engineer understands the industry, and the capabilities of the LLM, the better prompts they can create.

UXPrompts.ai is a LLM prompt library purpose-built to serve and support the UX industry. This library was born out of my own collection of prompts that I had been using as part of my day job running a large, globally distributed digital product team. 

Through countless hours of trial and error, we had figured out the best ways to incorporate LLMs into our own workflow. The way we did it was to first define and document the different stages of our own process.  Our team roughly follows the Double Diamond design thinking process and with a bit of tweaking, this can be aligned to the five stages of the UX Process.

1. Empathize
The Empathize stage focuses on understanding the needs, behaviors, and motivations of our target users through various research methods. This stage involves planning and conducting research, analyzing data, and synthesizing findings to create a deep understanding of the user’s perspective.

2. Define
The Define stage clarifies the problem statements, defines the scope and user journeys, gathers requirements, and sets business goals and priorities to create a clear project direction.

3. Ideate
In the ideation stage, creative solutions are brainstormed, developed, and refined through structured sessions  and feedback loops to form a solid foundation for prototyping.

4. Prototype
The Prototype stage involves creating, reviewing, and refining prototypes based on different fidelity levels to explore and iterate on design solutions and integrate them into the user flow.

5. Test
The Test stage involves conducting sessions with users to gather insights, analyzing the data to identify usability issues, and refining the product based on feedback to ensure it meets user needs effectively.

Now that we have the different stages defined, I can look to see what are the most frequently encountered problems and/or expected deliverables at each stage.  

The fun part of this is that I can use a LLM as a sparring partner. For each stage above, I created a list of sub-stages (e.g. 1.1 Research Preparation, 2.2 Scope Definition, 3.2 Idea Selection, 4.1 Prototype planning, etc..) with their own tasks that need to be completed before moving onto the next stage (e.g. 1.1.a Define objective and goals, identify patterns and insights, map out known constraints, etc..). This list was created from my own experience running UX teams over the past 15 years. However, even with that experience I’m sure that I missed something. So I asked ChatGPT 4o to review and help me identify any missing stages and/or tasks.

I then asked the same of Gemini 1.5-Pro and was able to compare the results and combine them to round out any gaps that I had in my initial process. Now that I had a (long) list of specific tasks, and an understanding of where they fit into the overarching process, I was ready to get started creating the prompts.

At my day job, my team and I had been running several corporate training sessions on how different teams could efficiently adopt and implement LLMs into their own team’s process. Mostly, this was centered around how to craft really good and purposeful prompts. We had even started to run internal Prompt-a-thons, where different teams would try to craft the best prompt to common departmental challenges.  This taught us a LOT about prompt patterns, prompt structure and the level of detail needed for a successful prompt.

For this prompt library, given the variety of roles at play through the UX process (e.g. UX Researcher, UX Designer, Project Manager,  Business Analyst, etc..) I chose to use a persona prompt pattern with the following structure:

Persona the LLM should adopt + Key knowledge, experience, or background for the persona + Detailed task for the LLM + Task objectives + Desired output or deliverables

Reaching into my personal library I selected several hand-crafted prompts and fed them through the LLM to “train” it on what a successful prompt structure would look like. Then, going back to my detailed tasks list from above I asked the LLM, stage by stage, to help me craft new prompts in the provided format. To keep the LLM from becoming redundant, I made sure to provide the complete UX process and reference previous prompts that either I had provided or it had created.

I revised all of the generated prompts for accuracy, usefulness and adaptability, making tweaks where necessary.  Often, after long conversations with the LLM it would drift away from the provided structure, so I would have to course correct or start a new chat.

The final prompts were then added to a detailed spreadsheet, and cataloged according to stage of the UX process and output generated.  Also, any part of the prompt that was configurable was highlighted, making it easier for library users to swap in their own unique project context or details.

The prompt library was then shared out to my professional network and multiple UX & Product communities to add their own prompts and give constructive feedback.

With the library created, I could now dive into the world of digital product sales. I wanted to take this opportunity to learn as much as I could about the different digital-only ecommerce platforms and what it takes to market and promote a digital product. I wanted to get this library launched as quickly as possible so that I could experiment with different pricing models and CPC campaigns, to test and learn.

The first step was exploring how to put the library behind a paywall. I explored different platforms like Notion, Airtable, Google sheets and Zoho and what options they had to paywall their content. I settled on using Google Sheets + a custom app script as it was the quickest and cheapest option and I figured that virtually all of my potential users would be familiar with Google Sheets.

Next, I evaluated different payment processors like PayPal, Stripe and also integrated marketplaces like Etsy, Gumroad,  Shopify and Sellfy.  I settled on Gumroad because of it’s ease of use and ability to have tiered subscriptions. 

Once I had the Google sheets + App Script + Gumroad integrations complete, I created a product landing page using Carrd and, of course, used Midjourney to create all of the key visuals.

With the landing page setup and the paywall integrations complete, I was now ready to acquire customers.  The initial feedback I had received from my professional network and UX communities were overwhelmingly positive. That market validation is what gave me assurance to proceed into productizing the library in the first place.

From that initial outreach I created a waitlist and had been in contact with them about their interest and how they expected to use the library in their own workflows.  This qualitative feedback allowed me to hone the unique value proposition of my product and be crystal-clear about the problems it could solve.

Working as part of a large team means that individual roles are often quite specialized. It is rare that one person gets to be completely hands-on throughout all stages of product development, product positioning, product launch, marketing and growth. That is the appeal of side projects – they provide opportunities to learn the things that we don’t get to learn on the job.

I also wanted to see how useful Generative AI could be at various stages of the product development process. I found it to be a useful sparring partner for generating prompts and acting as a challenger against my defined processes and tasks. I found it useful in quickly generating variations of advertising taglines and copy that I had created, and was also a brilliant creative partner in generating key website and advertising visuals. I also got to understand its limitations and usefulness.

However, nothing is perfect right out of the gate, and often it can be quicker to just do it myself than to describe and teach a chatbot to deliver the results I want.