After the Honeymoon: The Novelty Effect and the Long Game of Social Robots

Almost every social robot is a hit in its first week, which is exactly why the field has learnt to distrust first weeks and to ask what keeps people coming back after the magic wears off.

The first-week problem

Put a new robot in a school corridor, a hospital lobby or a living room and something reliable happens: people crowd round it. They talk to it, test it, film it and tell their friends. Then, with equal reliability, most of them drift away. Researchers call this the novelty effect, and it is one of the most important confounds in human-robot interaction. A short lab study can easily mistake curiosity about a new gadget for genuine engagement with a social partner.

Anyone who works with audiences will recognise the pattern. A good trick is astonishing the first time and pleasant the second. By the fifth viewing you are watching the performer, not the effect. Robots that rely on surprise alone run into the same wall, and the only honest way to find out whether a robot is more than a trick is to leave it in place for a long time and keep measuring.

Robots in the classroom

Some of the earliest serious attempts to do that came from Takayuki Kanda, Hiroshi Ishiguro and colleagues in Japan. In a field trial reported in 2004, two Robovie humanoid robots spent about two and a half weeks in a Japanese elementary school, speaking English with the pupils. The aim was partly to see whether the robots could act as social partners and informal peer tutors. Interest was high at first and then fell away; later summaries of the work note that most children lost interest, perhaps partly because they struggled to understand the robot. There was no clear overall learning gain, although children who kept interacting with the robot into the second week tended to do better on the English test afterwards.

Kanda and colleagues returned with a longer study, a two-month trial in an elementary school classroom reported in 2007. This time the team reported that it kept interacting with many children over the full two months and appeared to form friendly relationships with them. The contrast between the two studies is instructive. Duration alone does not doom a robot; what the robot does with that time matters.

Thirty seconds at the front desk

A similar story came from Carnegie Mellon University, where Gockley and colleagues installed Valerie, a robot receptionist with a scripted character, in the entrance of a university building. Over nine months many people interacted with her every day, yet few individual exchanges lasted longer than about 30 seconds. People came back, but mostly for a quick greeting or a practical question. That is not failure, exactly, but it shows that daily contact and deep engagement are different things, and that a robot can be woven into a routine without ever becoming much of a companion.

What the reviews say

In 2013 Iolanda Leite, Carlos Martinho and Ana Paiva published a survey of social robots built for long-term interaction in the International Journal of Social Robotics. Reviewing the long-term studies then available, they described the main features of such robots, drew together what seemed to help robots stay interesting and proposed directions for future work. A recurring theme in the literature they gathered was that robots and agents without strong social abilities tended to lose their users over time. Novelty buys a few sessions; something else has to carry the rest.

A much larger review appeared in 2025, when Kayla Matheus, Rebecca Ramnauth, Brian Scassellati and Nicole Salomons examined 120 long-term studies published between 2003 and 2023. Their working definition of long-term was modest: the same users meeting a robot for more than three sessions on at least three separate days. Even by that bar, they found the field had made progress, with more real-world deployments, a wider range of ages and more autonomous robots rather than ones secretly steered by researchers. They also found gaps, including few studies with teenagers or in workplaces.

Designing for the second month

What does this mean for anyone building a social robot? First, that early enthusiasm should be treated as noise, not signal. A pilot that runs for a single afternoon says little about whether a product will survive contact with daily life. Second, that variety and memory matter. A robot that greets you the same way on day forty as on day one is announcing that it has not noticed you. Third, that usefulness and sociability need each other. Valerie was useful and short-lived as a conversation partner; a purely chatty robot with no function struggles to justify the effort of talking to it.

Large language models have changed one part of this equation, since conversation no longer has to be scripted line by line. But fluent talk is not the same as a relationship, and a robot that can say anything may still have nothing to remember. The novelty effect is a reminder that people are not only evaluating a machine; they are deciding whether it is worth a place in their lives. That decision takes weeks, not minutes, and the research that respects that timescale is still the exception rather than the rule.

References

1. Kanda, T., Hirano, T., Eaton, D. and Ishiguro, H. (2004). Interactive robots as social partners and peer tutors for children: A field trial. Human-Computer Interaction, 19(1-2), 61-84. Link
2. Kanda, T., Sato, R., Saiwaki, N. and Ishiguro, H. (2007). A two-month field trial in an elementary school for long-term human-robot interaction. IEEE Transactions on Robotics, 23(5), 962-971. Link
3. Gockley, R., Bruce, A., Forlizzi, J., Michalowski, M., Mundell, A., Rosenthal, S., Sellner, B., Simmons, R., Snipes, K., Schultz, A. C. and Wang, J. (2005). Designing robots for long-term social interaction. Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 1338-1343. Link
4. Leite, I., Martinho, C. and Paiva, A. (2013). Social robots for long-term interaction: A survey. International Journal of Social Robotics, 5(2), 291-308. Link
5. Matheus, K., Ramnauth, R., Scassellati, B. and Salomons, N. (2025). Long-term interactions with social robots: Trends, insights, and recommendations. ACM Transactions on Human-Robot Interaction, 14(3), 1-42. Link

Feature drafted with AI from the published research listed above, for social-robotics.com.