UX Caffè

AI and UX Research: How the Way We Understand Users Is Changing

From interview analysis to synthetic users: artificial intelligence is accelerating UX Research, but truly understanding people remains a human responsibility

2026-10-08

Artificial intelligence is entering practically every stage of digital product design. It generates text, images, prototypes, code, and interfaces. So it was inevitable that it would also reach one of the most delicate areas of UX: User Research.

And this is exactly where things become particularly interesting.

Asking AI to generate ten alternatives for a button label is one thing. Using it to interpret what people think, want, or do is something entirely different.

UX Research exists to reduce the distance between the people designing a product and the people who will actually use it. Interviews, usability tests, surveys, behavioral analysis, and observation help us avoid designing products based only on our own assumptions.

Now, however, AI promises to accelerate a significant part of this process.

It can transcribe an interview in minutes, summarize dozens of conversations, identify recurring themes, classify hundreds of pieces of feedback, suggest research questions, and even simulate potential users.

That is a major transformation.

According to Maze’s Future of User Research Report 2025, 58% of surveyed professionals said they were already using AI tools in research. Among those using AI, 74% used it for research-data analysis and 58% for transcription. The main perceived benefits included greater efficiency, faster research cycles, and optimized workflows.

By 2026, the phenomenon has become even more visible: Lyssna’s research identifies AI-assisted analysis and synthesis as the most significant UX Research trend, selected by 88% of the researchers surveyed.

So the question is no longer whether AI will enter User Research.

It already has.

The more interesting question is: how much can we delegate to AI without losing our ability to truly understand users?

From Transcription to Insight: Where AI Is Already Changing UX Research

One of the most immediate applications of AI concerns all the necessary but extremely time-consuming work that happens after a research session.

Imagine ten one-hour interviews.

Traditionally, that means hours of recordings, transcripts, notes, highlights, classifications, affinity mapping, and comparison between participants’ responses.

Today, a significant part of this process can be accelerated by artificial intelligence.

An AI system can analyze transcripts, identify recurring topics, group similar observations, extract relevant quotes, and create an initial summary of the findings.

This does not necessarily mean eliminating the researcher.

It means eliminating part of the mechanical work so the researcher can focus more on interpretation.

And this is probably where AI currently offers the most tangible value.

Nielsen Norman Group has highlighted how artificial intelligence can be particularly useful during the planning and analysis stages of research: it can help prepare materials, structure questions, organize information, and accelerate repetitive activities.

The researcher therefore spends less time moving information around and more time asking what that information actually means.

Preparing Research with AI

Even before meeting users, artificial intelligence can become a kind of research assistant.

We can use it to create an initial research-plan draft, identify possible interview questions, review a screener, highlight questions that may be too leading, or suggest hypotheses worth testing.

This can be particularly useful for small teams where there is no dedicated UX Researcher and the designer is also responsible for research.

However, there is a fundamental difference between using AI to help prepare research and allowing AI to independently decide how that research should be conducted.

A generative model can suggest a perfectly written question that is methodologically inappropriate for the study.

It can also fabricate references, misunderstand context, or suggest a methodology that sounds convincing but does not actually fit the problem.

AI can therefore be an excellent sparring partner.

But it should not automatically become the methodological owner of the research.

Analyzing Hundreds of Feedback Items Without Losing Your Mind

There is another area where the advantage becomes even clearer: large volumes of unstructured feedback.

Reviews, support tickets, comments, open-ended surveys, customer-service conversations, and feedback collected directly through the product can become a goldmine of information.

The problem is that someone has to read it all.

When the volume reaches hundreds or thousands of entries, doing this manually becomes difficult.

AI can classify this information, identify recurring themes, detect emerging problems, and help teams quickly discover patterns that might otherwise remain hidden.

The real opportunity is not simply about “doing things faster.”

It is about being able to analyze information that, because of time and resource limitations, we probably would not have analyzed at all.

This is where AI and UX Research can become a genuinely powerful combination.

But a Summary Is Not Automatically an Insight

This is where the first major risk appears.

When an AI system produces an extremely organized summary of twenty interviews, it is easy to perceive it as authoritative.

Five main themes. Three recurring problems. Four recommendations.

Everything looks perfectly structured.

But User Research is rarely that clean.

People contradict themselves. They say one thing and do another. Their tone changes. They hesitate. They misunderstand questions. They tell seemingly secondary stories that suddenly reveal an entirely new perspective on the problem.

An important part of a researcher’s work consists precisely in recognizing these nuances.

AI can find patterns in the data we provide, but researchers still need to verify those patterns, return to the original evidence, and determine whether what appears important actually is.

Delegating synthesis cannot mean delegating judgment.

The Big 2026 Topic: Synthetic Users

And now we come to probably the most controversial part.

If we can use AI to analyze users, why not use it to simulate them as well?

This is where synthetic users enter the picture: AI-generated profiles designed to simulate the characteristics, behaviors, and responses of particular types of users.

In theory, the advantage is enormous.

No recruitment. No calendars to coordinate. No participant incentives. We can interview one hundred synthetic users in minutes.

It is therefore not surprising that the subject is attracting significant attention.

Research published by User Interviews in 2026 shows that synthetic users are already being experimented with for designing surveys and screeners, obtaining preliminary guidance, and conducting certain exploratory activities.

But the same research also reveals considerable caution.

Forty-seven percent of surveyed researchers describe themselves as skeptical and want more evidence before trusting these tools, while 24% say they are cautiously optimistic. Major concerns include the quality and accuracy of insights, the risk that stakeholders may place too much trust in AI-generated results, and the possibility of amplifying bias against underrepresented groups.

The fundamental problem is simple: a synthetic user has never actually experienced anything.

It has never waited twenty minutes at a service counter.

It has never used an application while holding a child.

It does not have a disability.

It has never felt frustrated because it needs to complete a task before entering a meeting.

It does not experience embarrassment, fear, urgency, or fatigue.

It can linguistically simulate these situations based on the data on which it was trained, but it does not possess those experiences.

And it is often within the imperfections of real people that we find the most valuable insights.

Does That Make Synthetic Users Useless? Not Necessarily

Completely dismissing synthetic users would probably be just as misguided.

They can be useful for preliminary, low-risk activities.

For example, they might help us stress-test a questionnaire before distributing it, rapidly explore hypotheses, identify obvious scenarios we may have overlooked, or better prepare for subsequent research involving real people.

In this sense, they become a kind of simulator.

They do not replace the final test, but they can help us arrive better prepared.

The fundamental distinction is therefore between generating hypotheses and generating evidence.

A synthetic user can help us formulate a question.

It should not automatically be considered the answer.

The Risk of Bias Becomes Even More Important

Every research process contains potential bias.

Artificial intelligence does not eliminate it.

In fact, it can make bias less visible.

Generative models work with enormous quantities of existing information. This means they can reflect imbalances, stereotypes, and gaps contained within their training data.

The risk becomes particularly sensitive when we attempt to use AI to simulate specific groups, minorities, or people with particular social, cultural, or physical circumstances.

An output may sound plausible without being representative.

And that plausibility is precisely what can make it dangerous.

An obviously absurd response is easy to discard.

An incorrect but extremely convincing response can instead become the basis for a product decision.

Privacy: Are We Giving AI Our User Interviews?

There is another less glamorous but essential issue: data.

A user-interview transcript can contain names, personal information, workplace problems, health information, behaviors, opinions, and confidential company details.

Before uploading everything into an AI tool, we therefore need to ask where that data goes, how long it is stored, whether it is used to train models, and what agreements have been made with research participants.

The convenience of a “Summarize” button does not remove the researcher’s responsibilities.

If anything, AI makes research governance even more important.

User Interviews’ research into synthetic users highlights a significant finding: almost 63% of participating researchers report that their teams still have no guidelines governing the use of these tools.

In other words, technology is advancing faster than the rules we use to decide how it should be applied.

The Researcher Does Not Disappear: The Value of the Role Changes

This leads us to perhaps the most interesting transformation.

If AI transcribes, classifies, summarizes, and prepares reports, part of the researcher’s operational workload decreases.

But other skills become increasingly important.

Understanding which questions are worth asking.

Choosing the correct methodology.

Evaluating the quality of evidence.

Identifying bias.

Distinguishing an interesting correlation from a genuinely useful insight.

Contextualizing findings.

Turning research into product decisions.

And, above all, maintaining a real relationship with people.

Nielsen Norman Group expressed this idea very clearly in its 2026 discussion of AI and UX work: even if AI were able to produce findings comparable to those of a researcher, direct observation of users would still provide fundamental value. Teams learn by seeing people use their products, hearing about their difficulties, and directly encountering behaviors that no report can fully reproduce.

And perhaps this is the most important point.

UX Research does not simply produce documents.

It produces shared knowledge about users.

A Possible Rule: Use AI to Accelerate, People to Understand

We can therefore imagine a modern UX Research process in which AI has a very specific role.

It can help us prepare.

It can transcribe.

It can organize.

It can identify patterns.

It can suggest interpretations.

It can transform large amounts of material into something more manageable.

But when we need to understand why a person behaves in a particular way, determine whether an insight is reliable, or make an important product decision, human oversight must remain central.

The most interesting model is therefore not “AI versus researcher.”

It is researcher with AI.

And the professionals who learn how to manage this collaboration effectively will probably be able to conduct more research, more quickly, and potentially with greater depth than before.

Conclusions

Artificial intelligence is profoundly changing UX Research, but probably not in the simplistic way we sometimes imagine. We are not witnessing the arrival of a technology that suddenly “knows users” on our behalf. Instead, we are gaining access to a new generation of tools capable of eliminating a significant amount of the operational work that has always accompanied research. Transcribing hours of interviews, classifying hundreds of feedback items, identifying recurring themes, organizing notes, and preparing an initial synthesis can all become dramatically faster. This means that a UX Researcher, or a designer who also conducts research, can dedicate more time to what really matters: asking the right questions, observing people, interpreting behaviors, challenging team assumptions, and transforming evidence into better decisions.

The risk begins when we confuse processing capability with the ability to understand. An AI model can produce an excellent summary without ever having used the product. It can simulate a user without having lived any of that person’s experiences, and it can generate an extremely plausible response without that response representing any real person. This is why synthetic users represent both one of the most interesting innovations in UX Research and one of the areas where we need to be most cautious. Using them to explore hypotheses, prepare a study, or identify scenarios worth investigating may make sense. Turning them into substitutes for real people means risking a system in which AI generates the users, interprets their responses, and ultimately suggests what we should design. At that point, we might have an incredibly fast process, but we would have lost exactly what UX Research is supposed to protect: contact with reality.

The most promising future is therefore one of AI-assisted but human-led research. Artificial intelligence can become an extraordinary accelerator, while researchers retain methodological responsibility, critical thinking, source verification, attention to privacy, and the ability to interpret human context. Paradoxically, the better AI becomes at processing information, the more valuable the designer’s and researcher’s ability to determine which information actually deserves attention will become.

For UX professionals, this should not be viewed only as a professional threat. It is also a major opportunity. Removing some repetitive activities can bring research back toward its most important purpose: not producing reports faster, but understanding people better in order to design better products.