Research brief
Automated social presence in artificial agents: A conventional and qualitative comparative meta-analysis
Antoine Juquelier, Simon Hazée, Yves Van Vaerenbergh & Ingrid Poncin · Journal of the Academy of Marketing Science (2026)
In one sentence
Chatbots, voice assistants and robots that make customers feel they are dealing with “someone” rather than “something” go hand in hand with higher satisfaction, engagement and loyalty, but this advantage fades in emotional and long-term relationships, and no single design feature, such as a human face or a name, creates it on its own.
Why this matters
Firms are investing heavily in making their artificial agents more sociable: warmer voices, more natural conversations, names and faces. Researchers call this “automated social presence”, that is the feeling of interacting with a socially present other when the counterpart is a machine. More than 160 empirical studies have examined it, yet their conclusions disagree: some find clear benefits, others none at all. Managers are left without clear answers on whether making AI more social pays off, when it works, and how to design it.
What we did
We conducted a meta-analysis: a study of studies. We combined the results of 165 studies covering 54,917 customers who interacted with chatbots, voice assistants and service robots. We also developed a new method to identify which combinations of design features make an agent feel socially present. Together, these analyses answer four questions:
Question 1
Does it pay off?
Links with satisfaction, engagement, loyalty and well-being
Question 2
Why does it work?
The perceptions that explain the effect
Question 3
When does it fade?
The service situations where it matters less
Question 4
How to design it?
The combinations of features that create it
What we found
Social presence pays off, for most outcomes. Agents that feel socially present are strongly associated with higher customer satisfaction, loyalty and engagement. The link with customers’ psychological well-being is much weaker: a friendly machine does not readily make people feel better about themselves or their lives.
It works through two routes. Customers see socially present agents as warmer and more competent, and they also come to feel a sense of ownership over the interaction: the agent seems more enjoyable, helpful and adaptable to their needs. Warmth, competence and enjoyment are the most consistent drivers across outcomes.
The advantage fades when emotions run high, and over time. Contrary to what many expected, automated social presence matters less in emotionally charged encounters (health, loans, service failures), where customers likely want genuine human support, and in long-term relationships with an agent, where its social cues lose their novelty.
There is no single magic design feature, but several winning combinations. No design feature on its own is required to make an agent feel socially present. Instead, three different combinations of features work, and the very same feature can help or hurt depending on what it is paired with.
Key takeaways for managers
Invest in social capabilities with confidence. Across 165 studies, agents that feel socially present are consistently associated with more satisfied, engaged and loyal customers.
Prioritise warmth, competence and enjoyment. Communicate with emotional warmth, give reliable and helpful answers, and make the interaction pleasant to use.
Hand over to humans when emotions run high. Let the agent detect emotional cues or urgency and proactively offer to connect the customer with an employee.
Refresh agents in long-term relationships. Social cues lose impact with repeated use; update conversational styles and features over time, and combine AI with human contact in ongoing relationships.
Design combinations, not single features. A human face, a name or a gender cue is not enough on its own; it must fit the agent’s capabilities.
Three design recipes that work
We classified agents along two dimensions: their capability (how broad their knowledge is, whether they talk by voice or text, and whether customers can speak freely or choose among predefined options) and their identity (a human-like appearance, a name, a female gender cue). Three combinations consistently created a strong sense of social presence:
Recipe 1
The specialist chat persona
Capability: specialised knowledge, text-based, predefined answer options
Identity: human-like appearance, a name and a female gender cue
Recipe 2
The all-round voice assistant
Capability: broad knowledge, voice-based, free-form conversation
Identity: a name, but no human appearance and no female gender cue
Recipe 3
The embodied voice specialist
Capability: specialised knowledge, voice-based, free-form conversation
Identity: human-like appearance and a name, but no female gender cue
The same ingredients can also backfire. A robot with a human appearance and a name, for example, felt less socially present when it could only offer scripted, predefined voice replies. Managers can choose the recipe that best fits their brand, technology and budget.
Keep in mind
A meta-analysis combines existing studies, most of which measured customers’ perceptions at a single point in time; the results show strong associations rather than proof of cause and effect in every setting. Studies based on surveys and crowdsourced participants tended to report stronger effects than experiments. Our design analysis covered six features because primary studies rarely describe their agents in more detail. Finally, design choices such as gender cues raise ethical questions, for instance about reinforcing stereotypes, that deserve careful attention.
Citation: Juquelier, A., Hazée, S., Van Vaerenbergh, Y., & Poncin, I. (2026). Automated social presence in artificial agents: A conventional and qualitative comparative meta-analysis. Journal of the Academy of Marketing Science, online first.
