For most of my career, UX research has followed a familiar rhythm. Recruit participants, schedule sessions, sit through hours of interviews, spend days manually coding transcripts, then distill it all into a deck that lands in someone’s inbox and maybe — maybe — influences a roadmap decision. It was rigorous. It was important. And it was painfully slow.
That rhythm is breaking. Not because research matters less, but because AI is collapsing the parts of the process that never required human judgment in the first place — leaving us more time for the parts that do.
Here’s what’s actually changing, from someone who thinks about this every day.
The 80% Problem
A striking statistic has been circulating in research circles: AI tools can cut qualitative analysis time by up to 80%. That’s not a marginal efficiency gain — that’s a structural shift in how research teams operate.
Think about what that 80% looked like before. Transcribing recordings. Tagging themes. Cross-referencing sessions. Building affinity diagrams on virtual whiteboards at 11pm. These were the taxes we paid to get to the insight. Necessary, but not where the real thinking happened.
Tools like Dovetail and Looppanel are now handling this work automatically — transcribing interviews, detecting themes, surfacing patterns across sessions, and generating summaries that used to take days. Dovetail’s AI theme detection, significantly refined in 2024, can process hundreds of interview transcripts and return meaningful clusters in minutes. Looppanel users report cutting their analysis time to roughly 30% of what it used to be.
The shift isn’t about doing less research. It’s about doing more of the research that actually requires a human brain.
From Reactive to Predictive
The old model of UX research was largely reactive. Something ships, users complain, research investigates. By then, the cost — in engineering time, in user frustration, in lost trust — has already been paid.
AI is making research predictive. Tools can now observe user sessions at scale, detect friction patterns, and flag potential drop-offs before a product team even reviews the metrics. Instead of reacting to churn reports, product leads can identify where problems are likely to occur and intervene earlier in the design cycle.
This is a meaningful change for design managers like me. Research used to be a gate in the process — something you did before you built, or after you shipped, but rarely in between. Now it can run continuously in the background, surfacing signals in real time. The best teams I’ve seen aren’t pausing to do research; they’re embedding it into every stage of development.
Sprig, for example, lets teams deploy micro-surveys directly inside their product — catching users at the exact moment of interaction rather than asking them to recall an experience days later. That’s not just faster. It’s more accurate.
Democratisation — And Its Risks
One of the most significant shifts is who gets to do research. AI-powered tools have dramatically lowered the barrier to running studies. A product manager can now set up an unmoderated usability test in Maze, collect 300 responses in under 48 hours, and get AI-generated themes and sentiment scores — no research background required.
This is mostly good. Research reaching more decisions, faster, is the goal. But it comes with a risk that anyone in a research leadership role should take seriously: the gap between running a study and understanding what it means.
AI can identify patterns. It cannot tell you why those patterns matter in the context of your users’ lives, your business strategy, or the ethical implications of acting on them. It can cluster themes from 200 interviews. It cannot hold the nuance of a single participant’s hesitation before answering a question — the kind of signal that changes everything.
The risk isn’t that AI replaces researchers. The risk is that people mistake AI outputs for research conclusions, and skip the critical thinking that turns data into direction.
What the Tools Are Actually Good At
Having spent time with most of the major platforms, here’s my honest read on where AI genuinely adds value in research right now:
Transcription and tagging — This is the most mature use case and where the ROI is clearest. Every tool does this now. Dovetail, Looppanel, and Notably all handle it well. Pick the one that fits your repository workflow.
Synthesis at scale — If you’re running continuous discovery or have a large corpus of past research, AI can find connections across studies that humans would never have the bandwidth to surface. This is genuinely powerful and underused.
Unmoderated testing — Maze and UserTesting have matured significantly. For validation tasks — does this flow make sense? does this copy land? — AI-assisted unmoderated testing is fast, affordable, and reliable enough to act on.
Sentiment analysis — Useful, but treat it as a signal, not a verdict. Context matters enormously in qualitative research, and sentiment models trained on general text don’t always read specialist or domain-specific language accurately.
AI-simulated users — This is the frontier, and I’d proceed with real caution. A handful of platforms now offer synthetic personas that simulate user behavior without recruiting real participants. The speed is extraordinary. The validity questions are equally extraordinary. For early-stage concept testing, maybe. For anything consequential, the jury is very much still out.
The Role That Survives and Thrives
At Insight Out 2025, Maze CEO Jonathan Widawski put it well: the ROI of research often lives in the negative space — in the things you didn’t build, the products you didn’t launch because you learned early that you shouldn’t. AI doesn’t change that. If anything, it amplifies it. You can now run that early-signal research in days instead of months, which means the case for doing it becomes easier to make.
What changes is where researchers spend their time. The work shifts toward strategy, toward asking better questions, toward translating AI-surfaced patterns into decisions that require judgment, empathy, and organizational context. Toward being, as Ryan Glasgow of Sprig put it, transformational for the business rather than just answering what color the button should be.
That’s a better job. It’s also a harder one to justify if you’re still spending most of your time on tasks a model can do in minutes.
What I’m Watching
A few things I think will define the next phase of AI in UX research:
Multimodal analysis — Combining transcript data with facial expression, eye tracking, and physiological signals to build richer behavioral models. Early, but directionally interesting.
Research in the design tool — The gap between designing and testing is closing. Figma integrations are already making it possible to run validation studies without leaving the canvas. This will become standard.
Ethical guardrails — As AI takes on more of the research process, questions around bias in training data, consent in synthetic personas, and transparency in AI-generated insights will move from academic to regulatory. Teams that build ethical frameworks now will be ahead of the curve.
The research ops role — Managing AI tools, ensuring data quality, and building the infrastructure for continuous research is becoming a discipline in its own right. Research ops is no longer optional at scale.
AI won’t make UX research obsolete. It will make bad UX research — the kind that’s slow, siloed, and treated as a phase rather than a practice — much harder to justify. And it will make the researchers who adapt into something closer to what the function was always supposed to be: the connective tissue between what users need and what organizations build.
That seems like a trade worth making.
What AI tools have made the biggest difference in your research practice? I’d genuinely like to know — find me on LinkedIn.
