How would you feel about a robot taking your order at a fancy restaurant? (AI Image by © Facundo - stock.adobe.com)
Diners Warm to Robotic Servers With Deeper Voices, Not Humanlike Faces
In a Nutshell
- Across two experiments with 463 participants each, a robot server’s voice pitch shaped how appealing diners found it, while a humanlike appearance and an assigned gender had little to no effect.
- Lower-pitched voices made robots seem more socially approachable in both experiments, and more competent or more physically appealing depending on the study; that appeal, in turn, shaped how willing people were to interact.
- Likely reasons: a confident-sounding voice signals competence, familiarity blunts the charm of a humanlike look, and short restaurant visits leave little room for gender cues to register.
Restaurants are betting big on robot servers, and most of that bet has gone into making them look right: friendlier faces, more humanlike bodies, the suggestion of a gender. A new study says much of that money may be aimed at the wrong target. What actually won diners over was not the machine’s appearance at all. It was the sound of its voice.
Researchers at three American universities ran two experiments to see which features make a restaurant robot appealing enough that customers want to deal with it. They tested three things people often assume matter: how humanlike the robot looked, whether it read as male or female, and the pitch of its voice. Only one of those moved the needle. Diners consistently warmed to robots with lower-pitched voices, while a humanlike face and an assigned gender barely registered.
For an industry that the authors note was valued at $1.26 billion in 2024 and is projected to climb toward $3.25 billion by 2032, that finding lands close to the wallet. A lifelike android costs far more to build than a boxy machine on wheels. If voice pitch sways customer impressions more than appearance does, restaurants may be paying a premium for a feature their customers shrug at.
Why Voice Pitch Beat a Humanlike Face
To test the idea, the team built realistic restaurant scenarios rather than wheeling actual robots into a dining room. Participants saw AI-generated images of robot servers paired with a recorded welcome speech, then answered survey questions about how appealing the robot was and whether they would want it serving them. Everyone was a U.S. resident, 18 or older, recruited through Prolific, a research platform used for online studies. Each person saw only one version of the robot, a setup that lets researchers compare groups without one robot coloring opinions of the next.
Study 1 put 463 people through a matchup of voice pitch (high or low) against humanlike appearance (high or low). Visually, the gap was real: a sleek android in a vest and bow tie on one end, a plain machine with a screen for a torso on the other. Yet appearance changed nothing about how attractive diners found the robot. Voice pitch did. People rated the low-pitched robots higher on what the researchers call task attraction, the sense that the robot is competent and can get the job done, and on social attraction, the feeling that the robot is approachable and could even be friendly.
Study 2 swapped in a new question, testing voice pitch against the robot’s perceived gender among another 463 participants. Gender made no difference to the robot’s appeal. Low pitch won again, though the pattern shifted: this time it boosted how physically appealing and how socially approachable the robot seemed, while leaving the sense of competence untouched. Study 1 had shown the reverse, with pitch lifting competence and approachability but not physical appeal.
One result held steady across both experiments: a lower voice made the robot feel more socially attractive, no matter what the robot looked like or which gender it seemed to be. Every other gain traded places depending on the study.
A Lower-Pitched Robot Voice Reads as Competent
A deeper voice’s pull on diners is not random. Earlier research the authors lean on associates lower pitch with authority, competence, and trustworthiness, the qualities a hungry customer wants in whoever is taking the order. A voice that sounds confident reads as capable, and a server that seems capable is one people are willing to engage. Separately, the study found that all three forms of appeal, physical, task, and social, predicted a diner’s willingness to interact with the robot, a link that held in both experiments. A catch follows: voice pitch did not reliably raise every one of those forms, since it depended on the study, with only social appeal rising consistently.
So why did the humanlike face fall flat? Terrah and his co-authors offer a plausible read. Diners may already be used to seeing machines in service roles, and familiarity dulls the novelty of a humanlike look. There is also a long-running worry in robotics called the uncanny valley, the idea that a robot which looks almost but not quite human can unsettle people rather than charm them. A lifelike android risks tripping that wire. A voice carries none of that baggage.
Asked what a winning robot voice actually sounds like, lead author Abraham Terrah of Oklahoma State University reached for two famous baritones. A low pitch does not mean a quiet one, he said in an interview with StudyFinds, rather it is “a deep, calm and smooth voice,” the register of a Morgan Freeman or a Johnny Cash. A higher voice, by contrast, can read as excited or urgent, not what most people want from whoever is taking their order.
Gender cut the same direction. A restaurant visit is brief and task-focused: greet, seat, serve, done. In an exchange that short, the authors reason, there is little room for gender cues to matter the way they might over a longer relationship. Diners are watching the clock and their appetite, not sizing up the robot’s identity.
Designing Restaurant Robots That People Actually Want
For restaurant owners and the companies that build their machines, the takeaway is refreshingly concrete. Money funneled into hyper-realistic faces and bodies may be money spent on a feature customers do not reward. A well-chosen voice, by contrast, is cheap to program and seems to do the heavy lifting.
Terrah put the math bluntly, noting that even non-humanoid service robots commonly run $15,000 to $20,000 to acquire. “I would definitely recommend restaurant owners to not overspend on robots for the sake of appearance, and invest in the robots’ communicative capabilities,” he said. He added a reality check: most commercial robots today handle a narrow set of tasks rather than anything close to general-purpose work.
There is a workplace angle too. Restaurant staff often eye new robots warily, wondering whether the machine is there to help or to replace them. Terrah is direct that the study is no cause for alarm: the robots now in restaurants suit predictable, repetitive jobs like carrying food or greeting guests, and the research does not show them taking over the rest. “Human workers remain essential for these parts of hospitality service that require judgment, flexibility, and also empathy,” he said, citing the ability to notice when something has gone wrong and fix it. His one caution is for owners, not staff: watch whether a robot is added as a support tool or slotted in to push people out of the service entirely.
A few cautions keep the finding in its lane. Nobody in the study ate a real meal served by a real robot; they reacted to pictures and recordings online, which is not the same as a busy Friday night. Researchers set the study in casual dining, not fine dining or fast food, so the lesson may not travel to every kind of restaurant. And the statistical gaps between high and low pitch, while real, were modest rather than dramatic. Its signal is clear, even if the size is smaller than a headline might imply.
Still, the direction is hard to miss. For years, the race to build a likable service robot has been a race to make it look more like a person. This study points to a quieter lever that designers have mostly ignored. Going by how people responded to its robots, customers may care less about a convincing human face than about a server that sounds like it knows what it is doing.
Author Q&A: A Conversation With OSU’s Abraham Terrah
Lead author Abraham Terrah, of the Spears School of Business at Oklahoma State University, answered questions about the study, what surprised the team, and what it means for restaurants and their staff. His responses appear in full below, lightly formatted for readability.
Your experiments suggest a robot waiter’s voice does more to win over diners than its face. Did that surprise you and your co-authors, given how much the industry spends on making robots look human?
Yes, to some extent. We expected voice to matter because it is one of the main ways through which a customer experiences a social entity. We think of hospitality robots as social actors because they embody distinct and recognizable roles, like servers, baristas, or bartenders. So the ability of speech, carried out through voice, is a characteristic of a social entity, in the sense that they interact with humans in the course of providing a service.
So we definitely expected voice to play a strong part. We also expected a more humanlike appearance to play a larger role, given how much attention is given to faces and bodies in robotic design. But our results showed that what really matters to robots’ attractiveness within a restaurant setting lies in its communicative capabilities.
Think about Siri, for example. It mostly communicates using voice technology, but is physically associated with a phone. But still, as interactions unfold, a personality emerges, which can be perceptible by users. So for a service robot used in restaurants where the service interactions tend to be short and standardized to some degree, voice is what makes that interaction feel natural and intuitive, regardless of the looks. Now, this does not mean that physical design is irrelevant. Rather, making a robot more humanlike on the outside is not really the most important way to make people comfortable interacting with service robots.
Low-pitched voices kept coming out ahead. When you were choosing the voices for the study, could you hear the difference yourself? What does a “low-pitched” robot server actually sound like to a customer walking in?
Absolutely! The difference is immediately noticeable and not subtle. And a low-pitched voice does not mean a quiet one. But it’s a deep, calm, and smooth voice. I am thinking, for example, about Morgan Freeman or Johnny Cash. To a customer walking in a restaurant, that kind of voice sounds grounded and calm; rather than sharp. Compare that to a high-pitched voice, which sounds more excited or urgent, and where the speakers’ energy is noticeably elevated. It’s fine for some contexts, but not necessarily what you want from someone or something taking your order.
The deeper voice made robots seem more competent and more approachable, depending on the study. Why do you think a lower voice carries that kind of weight with people, even when they know they’re dealing with a machine?
It’s worth being precise here; we are talking about pitch, not volume. A low-pitched voice can still be loud. When we think about humans, prior research on evolutionary psychology suggests that men are, in general, attracted to women with higher-pitched voices, while women are attracted to men with lower-pitched voices. So here we asked ourselves, would those hold true when it comes to service robots? Prior research on voice perception also suggested that lower-pitched voices are often associated with competence and authority. We associate such voices with confidence and composure. Based on our results, these associations appear to transfer directly onto service robots.
I would also add that lower-pitched voices for robots do not necessarily work for all contexts. But in our restaurant scenarios, voice pitch was a powerful cue, with low-pitched voices found more appropriate for the restaurant service context. So when it comes to robot servers, low-pitched voices were preferred regardless of the robot’s appearance and conferred gender, and this may be, in fact, because the robot is a machine. In the end, it’s not as if people would think about getting in romantic relationships with the robot servers.
A humanlike face barely moved the needle. There’s a long-standing idea in robotics called the uncanny valley, where an almost-human robot starts to unsettle people. Did you see any sign of that discomfort in how participants reacted to your most lifelike robot?
We were attentive to that risk when going into the design of this research. The uncanny valley effect was first theorized by Masahiro Mori in the 1970s. It is used to describe how robots that are perceived as almost-human can feel unsettling rather than reassuring. That effect shows up most strongly with very fine humanlike detail robots, like hair and facial hair, or skin texture.
In our experiments, the humanlike conditions were designed to stop short of these kinds of markers, and we kept some mechanical cues visible so participants always knew they were presented with a machine. So we did not see a negative reaction even for the robots depicted as almost-human in appearance.
That said, one limitation related to our design was that our research participants responded to images and recordings of robot servers, and did not encounter an actual robot in person. So it is also likely that some uncanny valley effects might not fully surface unless people are interacting face-to-face with a robot.
Whether the robot read as male or female made no difference to diners. That cuts against a lot of assumptions about gendered service roles. Were you expecting gender to matter more, and what do you make of it not mattering here?
Yes, we expected perceived gender could matter more. People often bring gender expectations into service settings, and robot designers themselves frequently signal gender through voice, appearance, or naming. So we wanted to test whether diners would respond more positively when a robot seemed to match familiar expectations about service roles. And then what we found was that perceived gender did not significantly change how attractive the robot came across. That was interesting because it implies that, for services as those provided by casual dining restaurants, customers may be evaluating the robot less as a gendered social actor and more as a service tool: How does it perform the service act? Is it easy to interact with?
You point out that a deeper voice is cheap to program while a lifelike body is expensive to build. If a restaurant owner asked you today where to put their money, what would you tell them?
This really goes to return on investment, especially since the costs of acquisition for robotic servers today remain relatively high (common models are currently $15,000 to $20,000). And those are for models that are not even humanoids. So, focusing on the shape and physical embodiment may be costly in the long-term. Voice, by comparison, is not as expensive to design and implement, but it’s also easy to reprogram. I would say that the main lesson of our research is not simply about picking a low-pitched voice. It is mostly about prioritizing a robot that performs consistently and can be adapted to the restaurant’s service style, all the while making the interaction easy for customers.
So I would definitely recommend restaurant owners not overspend on robots for the sake of appearance, and invest in the robots’ communicative capabilities. Most of the commercial robots available today are flexible within a fairly narrow set of tasks, rather than truly general-purpose, so owners should also be realistic about what such available robots can actually do.
Your study describes robots as helpers that handle routine tasks so staff can do the rest. How do you think workers should read this research, as a reassurance or a warning?
I don’t think that this research should be read as a warning that restaurant jobs are about to disappear. It is more about what helps customers feel comfortable interacting with service robots, and the study does not really show that robots can replace the full range of work done by restaurant employees. The kind of robots currently deployed in restaurants are best suited for tasks that are predictable and repetitive, like carrying food or greeting customers. Human workers remain essential for these parts of hospitality service that require judgment, flexibility, and empathy.
For example, noticing when something is wrong and taking a proactive approach toward fixing it, or when there are unusual requests. In their current service act, most robots are bound to follow a script and can’t really go beyond that script.
But, in the end, I do understand workers’ concerns. In my broader research on robotic applications in hospitality, I have heard employees and supervisors express some concerns about what more capable robots could mean for their jobs. I think workers should just be cautious about how robots are implemented. Is it just as a support tool, or are they being placed in such a way that removes human employees from the service experience? As for our study, it doesn’t provide reason to be warned.
Everyone in the study reacted to images and recordings online rather than a real robot at a real table. What’s the one thing you most want to test next once these robots are actually serving meals?
There are already plenty of restaurants with robot servers around the world. In an effort to map such establishments across the nation, I was able to identify more than 150 locations. So I would like to really test the fit between a robot and the restaurant’s actual service system. The robot has to navigate the dining room and possibly take orders. Mostly, it should allow for easy intuitive interactions according to the context of application, like fine dining vs. casual dining. So in testing this fit, I’d like to see whether predictions from this study in particular, and the body of literature in general, hold up across different service roles where the expectations are different; not only food runners but also bartenders or baristas. That kind of field research would help answer the question restaurant operators actually face, which is not just whether customers find a robot appealing, but whether that robot genuinely improves the service experience for both customers and employees.
Paper Notes
Limitations
The study’s design limits how far its conclusions can stretch. Participants reacted to AI-generated images and voice-over recordings in an online survey rather than to a live robot in an actual restaurant, so their responses may not match how real diners behave during a real meal. Because the robot images were AI-generated, the authors note, subtle visual cues may have shaped gender perception beyond the intended manipulations, which made the effect of gender harder to isolate cleanly. Both experiments covered only two levels each of voice pitch and humanlike appearance, a narrow slice of the full range of possible robot designs, and tested a single service role in a casual-dining setting. Findings may differ in fine dining, fast food, hotels, or other contexts, and the measured differences between high and low pitch, though statistically meaningful, were small.
Funding and Disclosures
The authors declare no known competing financial interests or personal relationships that could have appeared to influence the work. No external funding source is listed in the paper. Author contributions are recorded under a CRediT statement, with Abraham Terrah credited for conceptualization, methodology, formal analysis, data curation, and writing, and Luana Nanu and Cortney L. Norris credited for writing, with Nanu also contributing project administration and methodology. Per the paper, data will be made available on request.
Publication Details
The paper, titled “Enhancing human-robot interaction in restaurants: The impact of anthropomorphic features on perceived attractiveness,” was written by Abraham Terrah of Oklahoma State University, Luana Nanu of the University of South Florida, and Cortney L. Norris of the University of South Carolina. It appears in the International Journal of Hospitality Management, Volume 137 (2026), article 104696. It was received on February 2, 2025, revised on March 30, 2026, accepted on April 4, 2026, and made available online on April 8, 2026. DOI: 10.1016/j.ijhm.2026.104696.
