A critical and empathetic exploration of how synthetic users, created from qualitative research, can help us design better products.
This post, inspired by Lucho Domínguez's #SiTuMeDicesLearn, explores how synthetic users can become strategic allies — provided we start from a non-negotiable place: human empathy.
Speaking with our users every week: that is the mantra of continuous discovery so passionately championed by Teresa Torres, and also one of the most complex challenges to sustain in practice. Time pressures, lack of resources, internal bureaucracy, fear of making mistakes, and simple cognitive biases all hinder that constant interaction with the people we design for.
In this context, a controversial yet deeply thought-provoking figure emerges: the synthetic user.
Can an AI simulate a human being with enough depth to guide us in design decisions? Can it help us think better?
From the Desire to Research to the Fear of Getting It Wrong
Those of us who work in product design or UX Research have all felt that tension between what we know we should be doing (talking to users, validating, exploring) and what we actually manage to achieve. The barriers are not always technical. They are often mental:
- Fear of discovering we are on the wrong track.
- Overconfidence in intuitions dressed up as hypotheses.
- Numerical bias that dismisses qualitative evidence.
- And above all: the feeling that research is expensive or slow.
As Lucho reminds us, failing because you didn't research is far more costly. More than 50% of development teams' time is spent reworking features that could have been avoided with proper discovery. This is where synthetic users can help us — not to avoid fieldwork, but to facilitate continuous validation.
What Is a Synthetic User?
According to Nielsen Norman Group, a synthetic user is an AI-generated simulation that attempts to behave like a real person. It can be used for quick tests of value propositions, user flows, prototypes, or surveys before making direct contact with real users.
But — and this is crucial — it cannot generate new insights. It does not replace interviews, ethnographic testing, or generative research. It is useful after that work has been done, as a means of iterating more frequently and with less friction.
How Synthetic Users Are Built: Two Approaches Under Development
The industry is primarily developing two methods for generating synthetic users:
1. Psychology-based approach (OCEAN)
This method draws on the Big Five personality traits model: openness, conscientiousness, extraversion, agreeableness, and neuroticism. By combining these attributes, AI models generate psychologically plausible profiles with differentiated responses according to personality.
While useful for testing typical behaviours, it lacks contextual grounding and may miss key social and cultural nuances needed to understand a real segment.
2. Demographic data-based approach (UX Agent)
Here, profiles are generated from variables such as age, gender, income, or location. It is fast, but also limited. As Lucho points out, King Charles III and Ozzy Osbourne share demographic variables, yet they could not be more different.
Both approaches, if used without context, can produce models that "hallucinate": the AI fills in gaps with assumptions, creating coherent responses that have no real connection to a specific human group.
The Interactius Methodology: From the Field to Simulation
In response to these limitations, Lucho proposes an approach centred on empathy and ethnographic rigour. The key is not to start with AI, but with real people. His methodology is structured in five phases:
1. In-depth qualitative research
At least 20 in-depth interviews are conducted with a homogeneous segment. This number ensures theoretical saturation and allows patterns, contradictions, and complex emotional elements to be identified.
2. Analysis and coding
Data is analysed to extract recurring themes, tensions, symbolic barriers, and forms of expression. It is not just about knowing what people do, but how they feel about it and how they express it.
3. Transfer to the AI model
A conversational prompt is built, tailored to the language, context, and emergent personality of the interviewed group. Hallucination is avoided by closing all ambiguity gaps.
4. Conversational validation
The synthetic user is put through multiple tests: concepts, interfaces, flows, or value propositions are presented, and their responses are analysed. It is verified that they react with realism and consistency, rather than behaving like a sycophantic or neutralised model.
5. Support and updates
The team delivers tools to the client (ethical prompts, training) and supports the correct use of the avatar. A process of periodic review and updating is also established, since people change, and synthetic users must evolve with them.
Synthetic User Example: Montse, the Voice of Many, Synthesised
Montse is one of the avatars developed by Interactius following this methodology. She represents a segment of older women living alone. Her profile was built from 20 qualitative interviews, and her responses reflect an authentic voice, rich in nuance, emotion, and common sense.
When asked to introduce herself, Montse says: "I've been a widow for some time now, and although my children want to move me into a care home, I refuse. As long as I can look after myself, I want to stay in my own home, with my things, my silence, and my memories…"
When evaluating the Cruz Roja telecare service website, Montse notes: "The first thing that catches my eye is that red button and the fact that they can always locate you. I immediately feel a bit put off… It makes me think I'm being watched all the time. I'd prefer they offered it to me with more human, more approachable language."
And when presented with a smartwatch proposal that contacts emergency services and family members in the event of a fall, she responds: "I think it could be useful… only if the watch is comfortable, doesn't look like a medical device, and doesn't give false alarms. (…) And above all, that it doesn't make my daughter feel she has to keep an eye on me all day long."
But the most revealing moment comes when she is asked how she can help a product team: "So often, things are designed for older people without really asking us how we live or how we feel. What I do is say in my own words whether something is useful to me, whether it makes me feel confident, or whether, on the contrary, it makes me feel controlled, clumsy, or as if I no longer matter.
I can say whether something fits into my daily life, whether I'd understand it without help, or whether using it would stress me out. I can also explain what puts me at ease, what bothers me, and how I'd like to be spoken to or have a service offered to me.
In the end, I'm not technical or an expert in anything modern — but I am an expert in my own life. And if a product doesn't respect me or makes me feel out of place, I notice that straight away."
Montse does not just answer: she teaches.
She reminds us that knowledge does not always come from technology, but from lived experience.
When to Use a Synthetic User?
When well constructed, synthetic users can be valuable allies at key moments in the design process:
- Validating value propositions at early stages.
- Getting immediate feedback on wireframes, flows, or copy.
- Refining questionnaires or discussion guides before real fieldwork.
- Evaluating ideas in ideation sessions with quick inputs.
- Facilitating the onboarding of new team members.
Their value lies in agility and consistency. They allow the user's voice to be brought to the centre of the process without sacrificing quality or context.
Considerations for Responsible Use
Innovating with AI demands responsibility. Some key principles for ethical and effective use:
- They do not replace human contact. They complement; they do not substitute.
- Quality in = quality out. Without deep research, the model fills gaps with assumptions.
- Anonymisation and consent. All data must be handled ethically.
- Periodic updates. A synthetic user cannot remain frozen in time.
- Prompt training. Knowing how to ask is key to getting useful answers.
- Contextual use. An avatar must not be used outside the domain for which it was created.
3 Key Takeaways
- Well-constructed synthetic users enable agile iteration without losing depth.
- Empathy is irreplaceable, but it can scale if we encode it properly.
- Designing from a human perspective remains the only path to creating meaningful products.
Montse is not real, but everything about her was born from real people. Her words, emotions, and contradictions emerge from deep listening, empathetic observation, and respect for human complexity. By transforming that knowledge into a useful conversational model, we are not replacing anyone. We are putting more tools at the service of human-centred design.
Perhaps the greatest value of synthetic users lies not in what they tell us, but in how they remind us — time and again — that we design for human beings. And that, even in the age of AI, remains what matters most.
Don't Miss the Full Talk
If you are interested in exploring this practical, critical, and human vision of designing with AI, don't miss the full talk by Lucho Domínguez.
Request Information About Our Synthetic Users Service
If you would like to learn more about how synthetic users can benefit your research and design process, get in touch with us and fill in the form for more information — we will be delighted to walk you through the entire process.
