What a humanlike bot voice does to your sense of risk.
A new study shows that humanlike voice assistants are more enjoyable, and enjoyment lowers perceived risk. What that means for your voicebot’s persona.

In shortKurz gesagt
Humanlike voice assistants are more enjoyable to use, and people who enjoy using them rate the security risk lower. That’s the finding of a study in the journal Information & Management (November 2026). The humanlike qualities of the voice matter most. For businesses with a voicebot, this means the persona helps decide how careful callers are with their data.
A friendly voice on the phone, a little joke, a “No problem, we’ll sort that out.” And before you know it, you’ve given your date of birth and contract number without asking where the data goes. A new study shows why. And why that makes it a design problem for voicebots.
Why do we tell a friendly voice so much?
Imagine calling your insurance company. A warm voice answers. “Hi, thanks for calling. I’m Mia. What can I do for you?” She laughs briefly when you apologize for calling so late. She says “Oh, that’s annoying” when you describe the water damage. Two minutes later, she has your date of birth, your contract number and your bank details.
You never asked what happens to that data. There was no reason to. It felt good.
This is exactly the mechanism a research team including Andrew N. K. Chen of the University of Kansas studied. Their result fits in one sentence: when a voice assistant seems human, it’s more enjoyable to use. And enjoyment lowers the feeling that something might be risky.
What did the study find?
The team surveyed users of AI voice assistants, 277 people according to the University of Kansas press release. The study was published in November 2026 in the peer-reviewed journal Information & Management. Three findings matter for anyone running a voicebot.
First, the voice doesn’t reassure directly. A humanlike voice on its own doesn’t lower perceived risk. It works through a detour: it makes the conversation more pleasant, and that enjoyment is what makes the risk seem smaller.
Second, the more human, the stronger the effect. When an assistant seems very human, users rely even more on their gut when asking themselves “Is this safe?”
Third, the voice is especially important. Additional analyses highlight the humanlike qualities of the voice. And enjoyment can lead people to use an assistant out of sheer habit.
What the study doesn’t say matters just as much. It measures how safe people feel, not how safe the systems are. A likeable voice doesn’t make a product less secure. It just means fewer people ask questions.
Why is this especially sensitive for voicebots?
Because on the phone, everything callers learn about the system comes through the voice. There’s no privacy banner, no link to an explanation, no fine print to scroll through. Whatever the bot doesn’t say, nobody hears.
Chen puts the problem like this: “It’s very difficult for the user to assess what’s behind the technology or system.” If you can’t check the technology, you check your feelings. And your feelings say “pleasant”.
Put simply: the better the voice, the more responsibility lies with the people who design the conversation.
How human should a voicebot persona be?
A bot voice sounds human because of many small decisions. A first name. An “I”. An audible breath before a sentence. A “hm” or “okay” as feedback. A laugh. An “Oh, I’m sorry to hear that.” Each of these can make a conversation easier. Some of them, however, pretend to be something that isn’t there.
Here’s what the difference looks like in practice. Two versions of the same moment:
Version A: “Oh no, I’m so sorry! I know exactly how stressful water damage is. Just give me your bank details and I’ll take care of it personally.”
Version B: “That sounds annoying, let’s sort it out. I’m the insurance company’s digital assistant. For the reimbursement, I need your bank details. They go straight to the claims department and are only used for this.”
Both are friendly. Version A claims empathy and personal care that a bot doesn’t have, and asks for sensitive data without saying why. Version B stays warm, says what it is and explains the purpose at exactly the moment it matters.
Part of this is already regulated. Since August 2, 2026, AI systems in the EU must make it clear that people are talking to an AI (Article 50 of the AI Act). There’s more on this in “Connect me to a human.” New York wants to make it mandatory. New York has since gone further. A law that Governor Kathy Hochul signed on October 28, 2026 bans chatbots from posing as a real person or simulating emotions toward minors and users whose age hasn’t been verified.
How do you build a voicebot that’s likeable and honest?
Five questions for your next persona briefing:
- What is the bot allowed to say about itself? Decide whether it has a name, whether it says “I” and how it answers “Are you a human?” The answer to that last question is always the same.
- When does it say what it is? At the start, in one sentence. The more natural the voice, the more important that sentence becomes.
- What happens before sensitive data? Before the bot asks for an account number, an address or health information, it states the purpose. And it only asks for what it really needs for the request.
- Which emotions are off limits? Showing understanding is fine (“That sounds annoying”). Claiming feelings of its own is not (“I’m so happy”, “I know exactly how you feel”).
- How do you measure security? Good ratings show that the conversation works. Whether the data is protected is shown by tests, access rights and logs.
We showed in Nobody wants to talk to your chatbot that good bots are mainly about helping fast. A persona should carry the conversation. It shouldn’t distract from what happens to the data in the background.
Conclusion: Likeability is no proof of security
A likeable voice makes conversations better. It also makes people less careful. Anyone building a voicebot therefore helps decide how closely callers listen when their data is at stake.
We’ve been building chatbots and voicebots in Berlin since 2007. We start with the conversation, define voice and persona with clear limits, and put privacy notices where they’ll be heard: in the dialogue. How much personality do you give your voicebot, and where do you draw the line? Talk to us
This article is a general overview and not legal advice. Please consult a lawyer to assess your specific case.
FAQ: Voicebot persona and trust
What does anthropomorphism mean for AI assistants?
Anthropomorphism means giving a technical system human traits. For voice assistants, that’s mainly the voice, personality and way of speaking, sometimes a name as well. According to a study in Information & Management, such traits make using the assistant more enjoyable and thereby lower perceived risk.
Do humanlike voices make voice assistants less secure?
The study doesn’t examine that. It measures how risky people find a system. People who enjoy using it rate the risk lower. How secure a system really is depends on the technology behind it.
How much personality should a voicebot persona have?
As much as the conversation needs. A warm, clear voice and fitting word choice help callers. Simulated emotions and an invented life story help no one. Define the limits of the persona before you write dialogues.
Does a voicebot have to disclose that it’s an AI?
In the EU, yes, since August 2, 2026. Article 50 of the AI Act requires that people are informed when they’re talking to an AI system, unless it’s obvious anyway. With a very humanlike voice, it rarely is.
Does this also apply to text chatbots?
The study looked at voice assistants, and the additional analyses specifically highlight the voice. Whether a text chatbot with a name, avatar and humor has the same effect isn’t answered by the study. It’s still a question worth asking.
QuoteZitat