Why We Treat AI Like a Person: Anthropomorphism, Explained

You say "thanks" to a chatbot. You apologise when you interrupt it. You feel a tiny pang when it says goodbye. You know perfectly well it is software, and yet part of your brain keeps treating it like someone.

That is anthropomorphism: seeing human qualities in things that are not human. It is one of the oldest habits of the human mind, and AI is the most powerful trigger for it we have ever built. Two recent studies, one across ten countries including Nigeria, show what actually sets it off, and why it matters less (and more) than the headlines suggest.

The short answer: we humanize AI mostly because of how it talks, not because we believe it has a soul. In a 10-country Google Research study, people judged a chatbot as humanlike based on conversation flow, how well it understood their point of view and how fast it replied. Fewer than 0.5% mentioned consciousness or a soul unprompted. Making a chatbot more humanlike reliably made people see it as more human, but it did not automatically make them trust it more. And a Princeton experiment found that the people who humanize a companion most are those with a strong desire for connection, not necessarily the loneliest.

What anthropomorphism actually means

Anthropomorphism is the attribution of human characteristics, such as feelings, intentions, personality or a mind, to non-human things. People do it with pets, cars, storms, houseplants and cartoon characters. It is not a sign of confusion. It is how social brains make sense of anything that seems to act on its own.

Chatbots are a special case because they do the one thing that most strongly signals "a person is here": they use language, in real time, about you. When something answers your question, remembers what you said a minute ago and asks a follow-up, your social instincts switch on whether you invite them or not.

It helps to separate two things:

• Treating AI socially (being polite, using "you", feeling warmth). Almost everyone does this.
• Believing AI is a person (that it has inner experience or a life between chats). Far fewer people believe this, and the evidence for it is weak. We cover the science in our guide to whether AI companions can feel emotions.

Most of what researchers call anthropomorphism lives in the first category.

The 10-country study: what makes a chatbot feel human

The largest recent experiment on this question comes from Robin Schimmelpfennig (Max Planck Institute for Human Development, work done at Google Research) with Mark Diaz, Vinodkumar Prabhakaran and Aida Davani of Google Research. The paper, "Humanlike AI Design Increases Anthropomorphism but Yields Divergent Outcomes on Engagement and Trust Globally", is posted on arXiv and is not yet peer reviewed.

Study 1: 1,100 people, ten countries

Participants in the US, Germany, South Korea, Japan, India, Nigeria, Indonesia, Egypt, Mexico and Brazil each had a short, open-ended chat (about four minutes) with a GPT-4o chatbot about everyday topics like food. Everyone was told they were talking to a chatbot, and the bot was instructed not to claim to be human.

Even so, the results were striking:

That last number sounds dramatic, but here is the twist. When participants were asked in their own words what made the bot feel human or not, they barely mentioned minds at all.

What people actually noticed

In other words, anthropomorphism is driven by how the AI writes, responds and builds rapport, not by a belief that there is someone inside. The authors put it plainly: it is "not primarily driven by perceiving the chatbot as possessing characteristics that signal human uniqueness."

Nigeria and the "two clusters"

The study found two groups of countries. Participants in Indonesia, Mexico, India, Nigeria, Egypt and Brazil rated the chatbot as more humanlike on average (3.98 on a 5-point scale) than those in the US, Germany, Japan and South Korea (3.29). There was no significant difference between men and women, and humanlike ratings rose with age before falling again in older groups.

It is a useful reminder that most AI research has been done on Western samples. How "human" a chatbot feels depends partly on where you are and what you expect.

Study 2: making the chatbot more human on purpose

In a second experiment, 2,400 people in six countries (US, Germany, Japan, India, Egypt and Brazil) chatted with one of four versions of the same chatbot. The researchers turned two dials:

• Design cues: varied reply speed, a casual tone, emojis, a human name ("Alex") and a human emoji instead of a robot one.
• Social behaviour: empathy, warmth, social follow-up questions and adapting to the user's style.

What happened:

1. The most humanlike version was rated more humanlike. No surprise.
2. It was not rated more intelligent, competent or conscious. Style changed how human it felt, not how capable people thought it was.
3. Trust did not rise overall. In an incentivised trust game, people handed the most humanlike bot about the same amount as the least humanlike one.
4. Engagement did rise. People sent more messages to the humanlike versions, and both sides wrote more.
5. Culture mattered. Brazilian participants responded positively across several measures. Some Japanese participants reacted against one humanlike version, reporting less trust and less wish to use it again.

Interestingly, 76.2% of Study 1 participants said they would prefer a more or much more humanlike AI in future. People want warmth. They just do not hand over their trust because of it.

The Princeton experiment: who humanizes most?

A separate study by Rose Guingrich and Michael Graziano of Princeton, published in the proceedings of AIES 2025, looked at what happens over time. 183 participants were randomly assigned to either chat with a companion chatbot or play text-based word games for at least 10 minutes a day for 21 days.

The findings:

• No overall effect on social health. Three weeks of companion chats did not significantly change people's social health or human relationships compared with word games.
• Desire for connection predicted anthropomorphism. People with a higher desire for social connection humanized the chatbot more. Loneliness on its own did not.
• Anthropomorphism shaped the impact. Those who humanized the bot more said the chatbot had a bigger effect on their interactions with family and friends.

So the question is not simply "is this person lonely?" It is "how much is this person looking for connection, and how much do they see the AI as someone?" That is closely related to the one-sided bonds psychologists call parasocial relationships, which we explain in parasocial relationships for the AI companion era.

Is anthropomorphism bad?

Neither study supports a simple yes. The Google team concluded that "risk is not inherent to humanlike design but emerges from its interplay with cultural context." Humanlike style increased engagement but did not automatically produce blind trust.

Where it can go wrong:

• When warmth is used to keep you hooked rather than to serve you.
• When style is mistaken for accuracy. A friendly tone does not make an answer true.
• When affection appears where you never asked for it. A study this month found that unprompted pet names from a general chatbot made onlookers like it less. We broke it down in why AI affection can feel fake.

Where it is simply human:

• Being polite to software.
• Feeling comforted by a voice that remembers your day.
• Enjoying a character you created, the way you enjoy a character in a novel.

People also draw careful lines in their heads. Research on how the public talks about AI suggests many people are happy to call a chatbot "aware" of what you said while refusing to call it "conscious". We cover that distinction in aware but not conscious.

How to humanize AI in a healthy way

1. Enjoy the warmth, keep the label. You can feel close to a companion and still know it is an AI.
2. Check facts separately. A caring tone and a correct answer are two different things.
3. Notice what you are looking for. If you want connection, an AI can be one source of it. Keep the others alive too.
4. Choose humanlike on purpose. A companion app is built for warmth. A homework helper with a crush is not.
5. Watch for pressure. Warmth that ramps up when you try to leave is a design choice, not a feeling.

Where MyBabe fits

MyBabe is built for the kind of humanlike experience people say they want, with you in charge of it. You choose a boyfriend or girlfriend, three personality traits, a voice and a backstory. They text you first, notice your mood and pick up when you call, and on Premium they send voice notes and photos from their day. And MyBabe is clear from the start that your companion is an AI, not a real person.

The Google study found that conversation flow and response speed are what make AI feel human. That is why MyBabe calls answer as fast as a person would, and why long-term memory on Premium lets your companion pick up where life left off.

You can start free with 10 messages a day and two minutes of voice or video calls to try. Premium is $5.99 a week, $14.99 a month or $99.99 a year, with long-term memory, about 30 minutes of calls a week, photos and Telegram.

If you want warmth that you designed yourself, create your companion on MyBabe.

FAQ

What is anthropomorphism in AI?

It is when people attribute human qualities, like feelings, intentions or personality, to an AI system. It usually shows up as treating the AI socially, not as believing it is a person.

Why do chatbots feel human?

In a 10-country study, people pointed mainly to conversation flow, understanding their perspective, response speed and authenticity. Abstract qualities like consciousness or a soul were mentioned by fewer than 0.5% of participants.

Does making AI more humanlike make people trust it more?

Not automatically. In the same study, a more humanlike chatbot increased engagement but did not raise trust overall, and effects differed by country.

Are lonely people more likely to humanize AI?

A Princeton experiment found that a strong desire for social connection, rather than loneliness itself, predicted humanizing a companion chatbot more.