How I Employ AI in My UX Work

Note: This blog was written without ANY artificial intelligence; so that you can truly hear my voice in a somewhat casual tone. 

---

In the Beginning

The first time I used AI for work was probably employing Miro’s sort tools to synthesize research notes from a usability study, but I was pretty disappointed with the outcome because the AI organized the cards by semantics, thus as a human looking at the same data I saw the connection between these cards at a very different level. When I did it manually, I grouped the cards by relevance to the specific component and then secondarily by the persona from which I gleaned the feedback.  

This negative experience definitely tainted my view of AI’s capabilities early on, but later I played around with Midjourney’s image generation for fun and heard more and more about the amazing things that AI was doing. 

Later I trained and worked in AWS, crafting and constructing non-visual interfaces for a health insurance phone and chat flow, which would allow members and providers to self-service. This meant that I was designing for AI, but not designing with AI at the time.  

A few short years later I got a brief contract for human-centered content strategy work, and the client offered me use of a ChatGPT subscription for research purposes. It was amazing! I employed ChatGPT daily to pull in immigration statistics and data resources, so that I could build quantitative personas for the types of customers that would engage with my client’s offerings. I combined that quantitative data with the findings from stakeholder and subject matter expert interviews, and then built the entire content growth strategy and information architecture around those users. What’s incredible about this was that I eventually encountered an individual who was the real-life embodiment of one of my personas.  

That got me more actively using ChatGPT on a regular basis in my personal life. Typically, I used it to answer questions and gather knowledge resources on topics: what are the top 5 podcasts for data visualization or how do I calculate how much sourdough I need to create by gram for a .5 pullman vs a single loaf pullman.  

Fast Forward to Today

In my current role, I do full service UX, Dynamic Content Management, and own the Accessibility (AX) program for a behavioral health insurance company and so the ChatGPT usage has become more prominent. I may use it clarify ambiguity in WCAG compliance guidelines or find specifics to fill the gaps in end-user perspectives and preferences that we wouldn’t be able to gather without usability studies with disabled persons. I’ve employed CoPilot to provide me with code for ServiceNow tickets or to get the desired visual effect that I was going for in the source code of our WYSIWYG (LifeRay).  

All of these AI tools are used to speed up and clean up processes that I personally could do manually, but it would take so much longer. For example, I could manually update a page with markdown and it might take an hour, but doing it in ChatGPT took seconds and I can ask it to convert the document into JSON too!  

I find that for writing tasks I initially used AI to identify the right word or to name things or read this and summarize, but then I used it to expand my writing, because I tend to be quite concise in my writing style. Now I have shifted from using the ChatGPT or WIX onboard AI writing assistant for authoring to using Claude, but no matter which tool I use I always review the end product and make some updates to finalize the work. In that way AI speeds up the ideation process. 

How I Don’t Use AI  

I don’t trust AI to do the work for me. I never feel like AI content is perfect out-the-gate, so I generally feel the need to review anything that it puts out. I often look at the AI’s sited sources to ensure that I trust them. There have been instances where I didn’t like the small number of references used by the AI to give me an answer; so, I accounted for that in my decision making.  

Current Preferences in Personal AI Usage 

• I’m leaning more towards Venice for research, because it’s supposed to be committed to less censored, unfiltered information gathering.  

• I'm leaning more towards Claude for writing.

• I probably leverage ChatGPT for general Q&A because I like the voice chat feature, so I can learn while multi-tasking.

• I use the image generator in WIX to create thumbnails for my blog

• I've tried using some photo retouching GPTs, but I haven't seen much need to have a lot of images of myself.

I do find that Claude, Venice and Magisterium are verbose in their responses, so I have to ask them to keep their response to a number of paraphs or line lengths, but it is nice to see how robust their sited references are. 

Looking forward....

this blog is already too long. I'll talk about this in the next one.