notes

For years, every movement, every innovation, every voice has fought for a seat at the table– the opportunity to shape how decisions are made.

Artificial intelligence never asked for a seat.

Rather, it has been building a new table entirely; a table that restructures how decisions are informed, how products are designed, and how we interact with the systems around us. As technology continues to develop at an accelerated pace, the challenge we face is not whether AI belongs at the table, but whether we are prepared to work alongside it.

As someone who grew up in Silicon Valley, I know technology transforms fast. However, the past six months have truly given me a front seat to this. The integration of AI in all facets of life is a marvel but admittedly a little unsettling– seeing the Neo home robot complete household chores while standing taller than me is a frightening reminder that the user experiences we are creating are not confined to screens.

Beyond this, being a beginner of any discipline today is vastly different than it was three years ago. And I am a beginner right now– I’m a student learning UX Research and Design at a tipping point in the industry. Knowing this, it’s important to explore how I can grow with the industry.

So, what is the future of UX Research? It is one that requires responsible and careful collaboration with AI.

Artificial intelligence makes processes easier. It allows for the automation of many tasks that would otherwise take countless hours. However, relying solely on AI to complete phases of the research and design process risks distancing us from truly understanding the people we design for. And well, isn’t understanding people the whole point of user centered design?

While the research and design process can be time consuming, every phase offers opportunities to learn. There are facts and ideas that we can extract from data that AI will simply miss or get wrong. Although AI can recognize patterns, it lacks the ability to grasp context clues and nuances that humans use to understand and empathize with users.

For a long while, I was hesitant to use AI. I was afraid it would dull my own thinking, and cause me to lose my ability to understand and create independently. To some extent, this fear has proven to be valid. However, from my experience in graduate school, in particular the Generative AI & UX course I am currently taking, I am learning that by offloading repetitive tasks to AI, I am able to gain the time to focus on higher-level, strategic work in the user experience research and design process. My hesitation to embrace AI as a teammate comes from recognizing that, although it can automate repetitive tasks, it reduces the hands-on practice that helps us understand the reasoning behind our design decisions.

But my personal struggle with AI aside, I needed to see what this “new table” actually looked like in practice. During a semester-long project designing a VIP mobile app for University of Michigan Football fans, I practiced integrating AI into every phase of the UX research and design process. This project goes beyond a course assignment, it’s a stress test for the human-AI partnership; helping us understand where LLMs fail and excel, and how UX professionals, especially beginners, can overcome barriers to entry.

My experience with AI in the research process revealed several distinct friction points:

(1) Research Problem Framing
Research problem framing is the process of defining the user problem, business goals, and key research questions that guide the study. For the course project, we were asked to use an LLM of our choice to complete this process entirely. Using the provided design brief, we created a problem statement on our own and compared it to the AI output:

“Human-authored: University of Michigan football season ticket holders need a way to seamlessly access exclusive VIP benefits and important game-day logistics once at The Big House because the current season ticket holder experience not only lacks the exclusive, “insider” feel, but it does not provide these VIPs the gameday comfort they are paying for.

Claude-generated:
Prompt: You are an exceptional and thoughtful UX researcher and designer. I am designing a VIP single screen mobile app for University of Michigan football season ticket holders. Our goal is to provide an accessible, personalized experience, increase ticket renewal rates, and provide ticket holders with unique VIP experiences they can brag about. These experiences can be tiered based on the type of season tickets they hold, but they do not need to be. We want to create additional value for season ticket holders that works in tandem with the University of Michigan football experience. Using the project description attached in the brief and provided information, create a UX problem statement that combines emotional status and logistical friction. Attached Design Brief PDF

Output:
The Fragmented VIP Experience
→ University of Michigan football season ticket holders invest in season tickets expecting a premium, insider experience, but currently face both logistical barriers (parking uncertainty, unclear concession benefits, limited access information) and emotional disconnection (no recognition of their VIP status, no shareable exclusive moments). Gerald, who attends games solo after his divorce, and Jennifer, who brings her family, both need a unified digital experience that not only streamlines their game day logistics but also makes them feel valued and provides memorable experiences worth sharing. Without this integrated solution, season ticket holders experience their investment as transactional rather than transformational, leading to decreased renewal rates as the hassle outweighs the prestige.”

I chose to carry Claude.ai’s generated problem statement to the next phase of my research because I felt that it better incorporated the feature sets we were asked to include. However, I did find that it created an unconfirmed narrative at the end when sharing that season ticket holders experience their investment as transactional. Once I recognized this, I adjusted my problem statement and noted that as a point for further exploration.

This particular example illustrates how AI can rapidly expand and structure thinking that might otherwise take longer to articulate. However, it can also over-assert conclusions that have not been validated through user research. In this case, the generated problem statement made a claim about how season ticket holders experience “their investment as transactional rather than transformational”, and proceeded to connect it to decreased renewal rates. Neither of these claims had any evidence to support them.

Sole reliance on LLM generations becomes risky because it directly embeds assumptions into the foundation of the research process. When these assumptions are presented in polished language, as AI outputs typically are, it is very easy for them to be mistaken for validated user insights rather than hypotheses. Especially in the early stages of research, this can be particularly problematic as it not only influences the direction of the research, but it potentially narrows the scope of the project prematurely. Knowing this, it is important for researchers collaborating with AI to actively oversee LLM output, ensuring that hypotheses are not being mistaken for fact.

(2) Research Methodology Creation & Interview Facilitation
Once the table was set– the problem was framed– I had to figure out the points of interest for my interview protocol and research overall.

For this phase, I brainstormed questions centered around the proposed feature sets and separately leveraged AI to see if there were avenues of research that I could yield powerful insights. Doing so helped me capture the nuances that I would have potentially missed otherwise. To explore this, as well as my overall project scope, with Claude, I asked:

Prompt: Building on this project on the University of Michigan football VIP Mobile App, create a project scope. Include these sections: 1. In-Scope: Focus on the mobile 'Digital ID' experience. Include features for parking availability, concession/merch discounts, personalized merch, potential pregame field access, and social sharing 'bragging rights'. Specify the need for a 'single main view' as requested. 2. Out-of-Scope: Explicitly exclude web-based platforms (mobile-only) and other university sports (football-only). 3. Key Constraints: Accessibility: Must meet AA compliance standards; Branding: Must adhere strictly to the University of Michigan style guide 4. Open Questions: Generate 3-5 research questions regarding real-time data for parking, the technical feasibility of 'one-tap' social sharing from the stadium, the University of Michigan football fan community in general, and anything else that you may see fit.

Output:
[Claude provided a helpful perspective on the project scope and even provided preliminary design solutions that tackled the desired feature sets included in the design brief. In terms of research methodology creation, the following 3 questions are a sample of what it produced:]

> What percentage of season ticket holders currently attend games alone versus with groups, and how does this impact their desired experience? Should we design primarily for individual VIP experiences or group/family experiences, or do we need distinct pathways for both?
> How many season ticket tiers currently exist, and what are the meaningful differentiators between them? What additional VIP benefits would be compelling enough to motivate upgrades or renewals without devaluing existing tier benefits?
> What are the cellular data bandwidth limitations within Michigan Stadium during games, and will this impact users' ability to upload photos or share content to social media in real-time? Should we implement a queue system for delayed posting after users leave the congested area?

For me, this particular output highlighted the power of partnering with AI. While I explored asking human-centered questions aiming to create memorable experiences, Claude challenged me to consider the “how” behind the experience I am creating.

While no part of real user interviews can ever be replaced with AI, collaborating with AI tools beforehand can improve how we prepare for them. AI can help us clarify our objectives, expand on themes we want to explore, and refine questions we plan to ask to ensure that they are insight-driven.

By using AI tools to check and expand on our interview protocols, we open up opportunities for additional exploration. This preparation potentially equips us to enter interviews with greater curiosity. However, it is important to note that AI cannot replace humans especially in this stage.

AI cannot understand human context; it lacks the ability to build real connection and rapport.

You can follow a script word-for-word and learn a lot from those interviews, but humans are unpredictable. They have stories and insights that we could never prepare for; and those sometimes require additional probing and prodding that simply cannot be done by AI.

(3) Research Synthesis & Insight Generation
Following interviews, I dove into my gathered data. Once I sorted through the interview data on my own, I inputted the anonymized transcripts into Claude. The analysis was back “in my hands” within seconds.

I will say, however, this does not mean that it captured every nuance or detail that could be found within the uploaded transcripts. It shed light on common points that needed human verification and exploration. As Liu and Moran (2023) of the Nielsen Norman Group warn, relying on these tools to interpret research risks treating AI identified patterns as truth, while overlooking the subtle behaviors and insights that only humans can capture.

This particular realization was amplified as I moved into the insight generation phase.

At first glance, it was mind-blowing. With prompting, Claude generated 3 different personas in, again, seconds: The Social Connector, The Family Tradition Builder, and The Loyal Insider.

Each different from the other, with their own unique University of Michigan Football “brag-worthy moment”.

But, as I looked closer at them, the foundation wasn’t built on my collected data. AI had filled in the gaps where it saw fit. This can create some serious problems.

Especially when it comes to human-centered design, it is important that we understand the lens our teammate, AI, uses. AI tools are trained on massive datasets. While this can be powerful, those datasets do not represent a diverse range of voices. AI reflects the majority; inadvertently amplifying the loudest voice in the room. And as a result, proto personas lack unique, for lack of a better word, personalities.

Of course we’d love to have the complete picture when creating and designing, but people don’t check every box or have answers to every question. That’s OK!

----------------

Not only has AI entirely disrupted industries, it has redefined how the world operates. While we should not leave AI to do end-to-end UX research and design work, we can integrate AI as a teammate to increase efficiency in some processes. Doing so will allow us to direct our attention to parts of the research process that require more human care and nuance. When including AI in our processes, it’s imperative that we double check outputs, and keep in mind who we are designing for and whose voices we might be excluding.

The idea that we need to fight for a seat at the table has been instilled in each and every one of us. But perhaps the future of UX Research is not about fighting for a seat at the table AI is building, it’s about learning how we might build the table together. As a beginner in this field, I am learning that growing with the industry does not mean competing with AI or resisting it, rather, it means developing the judgement to know when to rely on it and when to lean into my own abilities. While artificial intelligence pushes the boundaries of what is possible, it is my responsibility to ensure that empathy and ethical awareness maintain their pace alongside it.

----------------

References
Anthropic. (2026). Claude (Claude Sonnet 4.6) [Large language model]. https://claude.ai

Google. (2026). Gemini (3 Flash) [Image generation AI]. https://gemini.google.com/

Jariwala, R. (2026, January 27). Assignment #1 [Unpublished manuscript]. University of Michigan.

Jariwala, R. (2026, February 3). Assignment #2 [Unpublished manuscript]. University of Michigan.

Liu, F., & Moran, K. (2023, July 2). AI-powered tools for UX research: Issues and limitations. Nielsen Norman Group. https://www.nngroup.com/articles/ai-powered-tools-limitations/