Across the tech industry, people are asking the same question about their role: can this be done by an LLM? UX Research is no exception. Product managers, designers and other team members who once relied on researchers for user data are now turning to ChatGPT for answers. It is faster, it is always available, and it does not ask for a research brief.
This shift is not hypothetical. It is happening now, in product teams, in roadmap conversations, in the quiet moments when a PM decides not to file a research request and types a prompt instead. The question is not whether AI is changing UX Research. The question is how deep that change goes — and whether researchers are thinking about it clearly enough.
To find out, I ran a workshop with colleagues across UX Design, UX Research and Research Operations. The goal was to understand not just how AI was affecting our day-to-day work, but how it was affecting our sense of professional identity — because those are not the same thing.
The question is not whether AI is changing UX Research. The question is how deep that change goes — and whether researchers are thinking about it clearly enough.
The Workshop
The session was structured in two parts. The first introduced the scope of change already visible across UX Research, Design and Engineering. The second was an ideation session where participants were encouraged to imagine the future — bold ideas, unconstrained by current reality — across three areas: skills and responsibilities, processes and tools, and collaboration and communication.
What came back was not what I expected.
Rather than projecting forward, participants described the present. The ideas that surfaced were current-looking — grounded in immediate experience, shaped by what was already happening around them rather than what might happen next. People ideated on tools they were already using, frustrations they were already feeling, workarounds they had already built.
This is not a failure of imagination. It is a known cognitive pattern — availability heuristic, the tendency to evaluate future probability based on how easily examples come to mind. When you are living inside a change, the change itself becomes the ceiling of your thinking.
The more important insight is what this pattern reveals: we are in the middle of something significant, and most of us have not yet stepped far enough outside it to see its full shape.
The challenge of systems thinking is to move from first-order ideas — AI synthesises data faster — to second-order consequences: if AI synthesises data faster, the role of the UX researcher shifts from data processor to strategic consultant. That requires different skills, a different presence in the organisation, and a different way of measuring value.
Essential Skills
More than 80% of UX Research professionals are already using AI in their work (User Interviews, State of User Research Report 2025). It has moved from experimental to operational — from something researchers tried to something they depend on. Transcription was the first obvious candidate for delegation. Synthesis tools followed. Now researchers are asking a harder question: not what AI can take over, but what it fundamentally cannot do.
The answer to that question starts with data. Privacy and data protection sit at the top of every Research Ops agenda, and for good reason. Processing participant data through third-party AI systems creates legal exposure that operations teams cannot absorb lightly. The risk of regulatory fines, the loss of organisational control over sensitive knowledge, the difficulty of auditing what an LLM does with what it receives — these are not theoretical concerns. They are active blockers to adoption, and they will not disappear as AI tools mature.
Alongside the privacy question sits a deeper operational need: the self-service insights machine. The vision is compelling — a system where any product team member can ask a question and receive a research-backed answer instantly, in context, without filing a request or waiting for a report. That vision has not yet been realised. And until it is, every research finding that sits in an unread document represents value the organisation paid for and never received.
What Researchers Need
The workshop surfaced a set of practical, specific needs that pointed clearly at where the real friction lives.
The first was repetition fatigue. Running six to eight interviews and asking substantially the same questions across all of them is a real cost — not just in time, but in cognitive engagement. Researchers want to minimise the repetitive surface of their work without losing the depth that makes each individual interview valuable. The nuance of a participant’s hesitation, the way an answer shifts when the question is slightly reframed, the follow-up that opens a door no one expected — these are not things that can be automated. But the scaffolding around them can be.
The second was verification — ensuring that survey responses and unmoderated testing data come from real humans rather than bots. As synthetic participation becomes more accessible, the integrity of research data becomes a more active concern rather than a background assumption.
The third was timing. Researchers frequently arrive at a problem after the critical moment has already passed — when the design decision has been made, the feature has shipped, the user has already developed a workaround. Capturing the moment of frustration as it happens, in real time, rather than reconstructing it through retrospective interview, is a persistent gap that no current tool fully closes.
The Moment Data Misses
Human emotions are not states. They are events. Frustration does not sit patiently waiting to be measured — it arrives, peaks and dissolves, often within seconds. By the time a researcher asks a participant how they felt when they couldn’t complete a task, the participant is already in a different emotional register, reconstructing an experience rather than reporting one.
This is not a limitation of research methodology. It is a limitation of time. The gap between when something is felt and when it can be articulated is where the most honest data lives — and where most research tools cannot reach.
What would it mean to close that gap? Not to eliminate the human interpretation of emotion, but to capture the raw signal — the moment of tension before it becomes a conscious thought — and bring it into the analysis alongside everything else. That is a frontier the field has not yet crossed. It is also one of the clearest cases for human researchers remaining in the process rather than being replaced by it. Emotional data requires emotional literacy to interpret. Pattern recognition can tell you that something happened. It cannot tell you what it meant.
What AI Cannot Hear
UX Research is built on the premise that what people say and what people do are different things. Every researcher knows this. What is less often said is that there is a third layer beneath both: what people communicate without intending to.
Non-verbal communication — the pause before an answer, the slight shift in posture when a question lands uncomfortably, the laugh that signals deflection rather than amusement — carries information that verbal transcription cannot preserve. Irony is routinely missed by literal interpretation. Humour often signals discomfort. Silence can mean confusion, reluctance, or careful thought, and distinguishing between them requires being present in the room, reading the whole person, not just the transcript.
These are not edge cases. They are the texture of human communication. And they are precisely the layer that separates a research finding from a research insight. A finding tells you what happened. An insight tells you why — and the why almost always lives in the parts of a conversation that never make it into a quote.
As AI tools become more capable of processing language at scale, the researcher’s role in attending to what language cannot carry becomes more important, not less. The ability to sit with ambiguity, to notice what is not being said, to follow the thread of a conversation into territory the participant did not plan to go — this is not a soft skill. It is the core competency of the role. And it is the competency most resistant to automation.
As AI tools become more capable of processing language at scale, the researcher’s role in attending to what language cannot carry becomes more important, not less.
How Researchers Learn
One pattern that emerged clearly from the workshop was a shift in how UX Researchers seek knowledge. Increasingly, the first point of contact for a methodological question, a new approach or a workshop design challenge is not a senior colleague — it is an AI assistant. LLMs have become a kind of always-available mentor: non-judgmental, immediately responsive, and free of the political risk that can come with visibly not knowing something in a corporate environment.
This is a significant change in professional culture. The informal knowledge transfer that happened through peer conversation — the corridor question, the shared doubt, the senior researcher who remembered a similar project three years ago — is being partially replaced by a more private, more individual mode of learning. That has real consequences for how expertise develops inside research teams and how institutional knowledge accumulates over time.
The second pattern was technical curiosity directed at AI itself as a subject of study. How do LLMs behave? How does the framing of a prompt change the response? How do users form expectations of AI-driven interfaces, and what happens when those expectations are violated? These are UX research questions applied to a new class of product — and they are becoming central to the work rather than peripheral to it.
This points toward a reframing of what AI literacy means for UX researchers. It is not primarily about learning to code or understanding transformer architecture. It is about developing the critical vocabulary to ask the right questions about AI systems — their limitations, their failure modes, their effects on human behaviour — and to bring that vocabulary into product conversations where it is currently absent.
Where This Leaves Us
The researchers in the workshop were not wrong to focus on the present. The present is genuinely challenging, genuinely interesting, and genuinely worth understanding. But the more important conversation — the one about what UX Research becomes in a world where AI handles the repeatable, the scalable and the fast — is still largely ahead of us.
That conversation needs researchers who can think in systems. Who can trace a first-order change to its second and third-order consequences. Who can articulate the value of what they do in terms that go beyond the deliverable — the report, the synthesis, the persona — and into the territory of strategic judgment, human interpretation and organisational intelligence.
The role is not disappearing. It is becoming more demanding. And the researchers who will thrive in it are the ones who understand the difference between what AI is taking over and what it is making more visible — and who are deliberately building toward the latter.