↓Skip to main content

Synthetic empathy as a design lever: notes from MobiWIS 2026

·4 mins
Conferences Research AI Synthetic Empathy Chatbots Career Guidance Persuasive Technology
Table of Contents
At MobiWIS 2026 my co-authors and I presented an experiment with a deliberately narrow question. Does the way a career-guidance chatbot talks affect how users move from understanding to action? With 126 participants, the answer is yes, and where the effect actually showed up was the part that surprised me. Communication style reshapes the early stretch of that path. By the last step, it stops mattering.
A cyclist crossing a stone square in Granada's old town.
A cyclist crossing a stone square in Granada’s old town.

The paper is joint work with Thong H. N. Dinh and Agnis Stibe. It is the empirical follow-up to the synthetic empathy paper I presented in Zagreb earlier this year. Zagreb made the conceptual case. This one tested it.

The experiment #

We split 126 students from applied computer science and software development programs between two chatbots, with a survey before and after a twelve-minute conversation. Both bots addressed the same career question. The only thing that differed was how they talked. One was empathic and coaching: it reflected feelings back, validated what users were going through, asked open questions, and held back on solutions. The other was strategic and rational, closer to a brisk advisor: diagnose the situation, name the trade-offs, push toward a next step.

The scope is what it is: a single session, one domain (career guidance), a sample skewed mostly male, and self-reported readiness rather than observed behavior over time. Real limits. Most first studies in this space have the same ones.

What changed #

Both chatbots moved users along the chain. The empathic one did it differently. Users went from knowing something about their situation to accepting the bot’s guidance nearly three times more strongly with the empathic bot than with the strategic one, and from acceptance the step to forming a real intention was almost automatic. With the strategic bot users stayed in their own heads, applying their own judgment at each step. Then the part that surprised me. Once an intention had formed, both styles produced essentially the same readiness to act. Style shapes the early part of the chain. After that, the user takes over.

Implications #

Chatbot communication style is a real design variable. It changes the path along which users decide, in ways that aggregate satisfaction scores miss.

There is a matching question I keep coming back to. Empathic style appears to fit users who arrive with some prior knowledge and a willingness to engage. Strategic style fits users who need structured reasoning before they will accept anything. A bot that can shift between the two based on the user in front of it should outperform a single-voice system, but somebody has to do the shifting well, and that is not trivial.

The sharper point is the one I keep writing about. After the empathic conversation, a meaningful share of users went straight from accepting the bot’s advice to feeling ready to act, without forming a deliberate intention in between. Warm language can move people without them stopping to think. In a low-stakes career-guidance scenario that may be fine, even useful. In health, finance, or anything political, it is a property a designer needs to know is there before they turn the empathy dial up.

The accepted manuscript has no yet been published, but it will be during the summer of 2026.

A few impressions from Granada, where the conference was held:

An orange tree heavy with fruit in front of the Granada Cathedral.
An orange tree heavy with fruit in front of the Granada Cathedral.
The Granada Cathedral facade seen from below, with carved relief panels and an oculus window in sharp afternoon light.
The Granada Cathedral facade seen from below, with carved relief panels and an oculus.
The Alcazaba fortifications inside the Alhambra.
The Alcazaba fortifications inside the Alhambra.
Stone ruins on a hillside in Granada framed by cypress trees.
Stone ruins in Alhambra.
A glazed tile panel inside the Alhambra showing the Plus Ultra motto.
A glazed tile panel inside the Alhambra showing the Plus Ultra motto.
View across the Darro valley to the Albaicín, with white houses descending the hillside.
View across the Darro valley to the Albaicín.
Carved Nasrid stucco vault inside the Alhambra.
Carved Nasrid stucco vault inside the Alhambra.
The Court of the Myrtles inside the Alhambra, with the long reflecting pool and the Tower of Comares at the far end.
The Court of the Myrtles inside the Alhambra, with the long reflecting pool and the Tower of Comares at the far end.
Kristian Stoffregen
Author
Kristian Stoffregen
Teacher and researcher. My work spans two distinct fields: digital and AI transformation in SMEs, and persuasive technology in generative AI, covering anthropomorphism, synthetic emotions, and the ethics of AI-driven influence. Both are grounded in two decades of practice inside companies such as Microsoft, PwC, Danfoss, CGI, and DR.
←→ Shipping my first open-source release