In a recent talk at the DIBI event in Edinburgh, we delved into the question of whether we can design and build using AI. Prompt engineering, a new skill that I believe will be a prerequisite in future job ads, has been a topic of discussion both internally and externally. While I rely on ChatGPT and Midjourney for various tasks like research, idea generation, and development, I was curious to explore their application in the UX/UI process, which is a significant part of my work.
To explore this concept further, we decided to examine a test project for New Balance, a client who graciously allowed us to use their name in our AI experiment. As the headline sponsors of the London Marathon, we wanted to research and map out a website targeting both runners in general and marathon runners specifically.
Starting with the basics, we employed ChatGPT to understand our audience and put the AI through its paces. Due to limited access to plugins and live internet during the preparation of this talk, we used zero-party data based on the 2021 cutoff date (stay tuned for more recent developments mentioned later in this article).
Breaking down the prompt into role, task context, and output format instructions, we received an unsurprisingly good result. To keep this article concise (and spare you from reading for half an hour), I’ll summarize the findings. The AI-generated task list consisted of ten items, all of which align with exercises typically performed in a comprehensive UX program.
For the purposes of the test, we asked our AI buddy to focus on the top two tasks. However, we paid particular attention to persona development, starting with the first task of Market and User Research. We brainstormed with the bot to create as many different personas as possible and, after narrowing down the list, developed one persona in detail.
Moving forward, we continued to work with the AI to map out our approach and process, conduct research, craft user profiles, and gain a deep understanding of the people we were designing for. But our exploration did not end there.
Building upon our persona, the Endurance Explorer, we brainstormed and drafted numerous user stories. We also created:
- A full user journey
- A user flow
- A site structure (including content ideas)
- A deeper understanding of Lucy’s world to establish stronger empathy
Were the results perfect? No, as AIs, like humans, have biases, flaws, and can return inaccuracies. Were the findings validated or refuted through interviews, tests, and workshops? Not in this case, as it was a light experiment conducted for a thirty-minute event talk. However, it did demonstrate the potential for using generative AI to accelerate the UX process.
To conclude the talk, I introduced a tool called Uizard, which I recently gained access to. Until recently, the UX/UI design practice had been one of the last creative domains resisting AI disruption. However, recent developments indicate that this is changing. We ran a live demo of Uizard, a tool that generates lo-fi wireframes and UI prototypes based on simple text prompts and stylistic context. This tool provides an initial creative starting point, incorporating AI-generated images and copy, applying design systems from reference sites and screenshots, and even creating clickable prototypes.
While Uizard is not intended as a comprehensive design package, it showcases the direction we are heading. We are still in the early stages of exploring generative AI, and our workflows with it may be somewhat rudimentary. Nevertheless, we remain in control, steering the wheel and pressing the pedal. It’s an exciting journey, so let’s have fun and stay safe on this AI-empowered path.
Footnote: Following the talk and during its preparation, ChatGPT Plus users gained access to Plugins. This enabled a live internet connection and the use of GPT-4’s enhanced capabilities. One notable plugin, called Yabble, allows for Silicon Sampling, which involves surveying a simulated group of people generated by GPT. It’s important to note that these simulated individuals are based on training data up until 2021, which represents the internet’s knowledge at that time.