Early enthusiasm for a social robot does not predict whether people keep using it

What it is

Early enthusiasm for a social robot — frequent interaction, positive ratings, users showing it off to others — largely reflects one fact: the thing is new. That is the novelty effect, and it is a separate question from whether the product delivers durable value. The novelty effect only requires unfamiliarity; durable value only becomes visible once unfamiliarity wears off. The practical warning here is that early engagement, early praise, and early interaction frequency cannot be used to predict use weeks or months later — treating them as evidence of long-term adoption is one of the most common methodological mistakes in this area.

Why it happens

The novelty effect rests on the fact that a stimulus carries less information the more it is repeated. The first time someone encounters an object that talks, turns its head, and expresses emotion, simply watching it and probing what it can do is rewarding in itself — not because the robot has solved a problem, but because not knowing what happens next carries its own exploratory value.

That exploratory value has a hidden precondition: the interaction space has to still be unexplored. Most social robots offer a fairly limited set of moves, phrases, and response patterns, and a handful of sessions is usually enough for a user to exhaust them — once someone has seen that the same few responses keep repeating, further interaction stops producing new information. At that point the portion of engagement driven by exploration collapses quickly, and continued use depends entirely on whether the robot supplies stable functional value on its own. This is why first-week or first-month engagement data is systematically inflated: it mostly captures behavior from before the exploration phase has ended, not behavior after the product's value has stabilized.

Studying it

Confirming whether a novelty effect is present, and how much it decays, requires longitudinal measurement rather than a single assessment: log the same cohort's interaction frequency or session length at day 1, week 1, month 1, and month 3, and check whether the curve shows a steep or sustained early peak followed by a clear decline that settles at a lower stable level. A flat or steadily rising curve is evidence against novelty being the dominant factor in that deployment.

Beyond behavioral logs, standardized instruments such as the Godspeed questionnaire can be administered repeatedly at different time points to track how subscale scores — likeability, perceived intelligence — change over time, rather than collecting a single total score at launch or at the end of a trial. A one-time score already has the novelty effect baked into it; without a comparison point there is no way to separate it out.

Where it stops holding

The speed and size of the decline depend on how much depth the interaction actually has. Products with a single function and a fixed interaction pattern decay fast, because their interaction space is exhausted quickly; products that offer personalized feedback or progressively unlock new content decay more slowly. There is no evidence, though, that any social robot avoids this decline entirely — the difference is only in rate.

This boundary does not apply to products explicitly designed for one-off or short-term use, such as trade-show or mall guide robots, where manufacturing a momentary sense of novelty is the actual design goal. Judging that kind of deployment by long-term retention standards misapplies the concept.

Applying it

Do not use first-contact or first-week satisfaction and engagement data to decide whether to scale up a deployment or keep funding it — that data mostly reflects exploration-phase behavior. The evaluation window needs to extend past the point where users have exhausted the interaction pattern, which varies with product complexity — commonly a few weeks to a month, longer for products with more content to explore.

The concrete check is to take the same core behavioral metrics (daily interaction count, self-initiated interactions) and re-measure them at multiple points after deployment, plotting a trend line, instead of taking a single snapshot at launch, in press coverage, or at the end of a trial period. A single data point cannot answer whether novelty is driving the numbers; only a trend can.

Related

• Same group: X6.04.2 A social robot that provides no lasting practical value gets abandoned once the novelty fades · X6.04.3 Judging whether a social robot succeeds requires evaluating it over a long deployment, not a short trial
• Nearby: X6.05 Use by special populations · X4.07 Operator situation awareness
• Search terms: novelty effect · longitudinal HRI · Godspeed questionnaire · habituation