I guess I expected AI to be like a college intern. It could handle some of the easier but time-consuming UX activities while I focused on the fun part—ANALYSIS!
Then my last study really surprised me. I gave several AI tools the same dataset and instructions, but most of them barely analyzed anything. They skipped requested analyses, produced surface-level summaries, made calculation and interpretation errors, found patterns the data didn’t support, and drew incorrect conclusions.
What made it worse was how polished the answers sounded. The problems became obvious only when I checked the results against the original data—which meant AI wasn’t saving me nearly as much time as I had expected.
That made me wonder: Was I expecting too much from AI, or were other UX professionals having the same experience?
To find out, I surveyed 34 UX professionals about how they use generative AI, which research activities they trust it to support, and how much human oversight they believe it requires.
At first, those results may look inconsistent. They are not.
X professionals are willing to use AI. They are even willing to put AI-assisted findings in front of stakeholders. But their confidence comes from their own review—not from trusting AI to be right.
Participants rated their trust in AI across nine UX research activities on a 5-point scale.
The highest-rated activities were:
Trust dropped when the work required interpretation:
That divide matters. UX professionals appear more comfortable asking AI to help create the structure around research than asking it to decide what the evidence means.
The open-ended responses made that practical use case even clearer. Among the 33 people who answered what they trusted AI with most:
Participants described using AI for report drafts, summaries, overviews, recommendations, and competitive or industry research.
These are meaningful tasks, but they still leave the researcher in control of what is accepted, revised, or rejected.
The least-trusted responses repeatedly returned to review. Human review was coded in 73% (24 of 33) of the responses.
One participant captured the reason simply: “Trust but verify as my name is on report.”
Others described reviewing everything several times, checking sources, and rechecking analysis for mistakes or strange data. Analysis, patterns, conclusions, bias, and unreliable sources also appeared as concerns.
The message is not that AI has no place in UX research. The message is that accountability cannot be delegated.
AI can assist. It cannot own the judgment.
These findings suggest a practical boundary for AI-assisted UX research: use AI to help draft, organize, summarize, and generate possibilities—but keep interpretation, verification, and final decisions in human hands.
That boundary may shift as the tools improve. For now, the professionals in this survey are not rejecting AI. They are adopting it with conditions.
AI can support the work. The researcher still has to make sense of it—and stand behind it.
Email: theresaw@columbus.rr.com
LinkedIn: theresa-wilkinson