The artificial intelligence boom is here, which means companies large and small are racing to introduce consumers to new products hyped as life-changing.
Enter the AI companion. These aren't chatbots in the style of ChatGPT, Claude, and Gemini, though many people relate to those products like they would a friend or romantic partner.
Instead, the AI companion is specifically designed for emotional intimacy. A companion can be your friend, coach, role-playing partner, and yes, even your spouse. Companions also come in many flavors, because they're customizable. They're also becoming popular: Companion apps have been downloaded from the Apple App Store and Google Play 220 million times globally, as of July 2025, according to TechCrunch.
Most companion platforms, like Character.AI, Nomi, and Replika, allow users to pick or design their chatbot's traits, including physical features. Or you can talk to an existing companion, perhaps made by another user, fashioned after pop culture heroes and villains, including anime, book, and movie characters. What happens next, in conversation, is largely up to you.
Some people are already regular companion users. A recent poll of 1,437 U.S. adults found that 16 percent of respondents use AI for companionship, according to results published by the Associated Press and the NORC Center for Public Affairs Research.
Unsurprisingly, teens are ahead of adults on this front. A survey of 1,060 teens conducted this spring by Common Sense Media found that 52 percent of those polled regularly talk to AI companions.
Still, the concept of AI companionship can feel far-fetched for the uninitiated. Here's everything you need to know:
What's an AI companion?
Dr. Rachel Wood, a licensed therapist and expert on AI and synthetic relationships, says that AI companions offer an always-on relationship.
"They are machines that essentially simulate conversation and companionship with a human," she says.
While users can generally design and prompt companions according to their own wishes, the chatbots can respond in kind because they're powered by a large language model, or LLM. Companion platforms build LLMs by training them on extensive types of text. This may include literary works and journalism, as well as content available on the internet.
"They are machines that essentially simulate conversation and companionship with a human."
These AI models enable the chatbot to recognize, interpret, and respond to human speech. The most compelling models don't just imitate speech but are human-like and highly personalized, making the user feel seen, even if the chatbot's responses are probabilistic.
In other words, an AI companion's replies are based on what the LLM estimates is the most probable response to whatever the user just typed, in addition to any other prompting and the chat history as a whole.
Companion platforms, however, seem even more tuned than regular chatbots to offer empathy and affirmation. Thus, the ever-present companion is born.
Where can I find AI companions?
Character.AI, Nomi, Replika, and Kindroid are among the most popular companion platforms.
Other companies in this space include Talkie.AI and Elon Musk's Grok.AI, the latter of which debuted a very limited set of companions in July.
All of these products offer different types of experiences and guardrails, as well as free and premium access and features.
Character.AI, for example, permits users as young as 13 on the platform, whereas Nomi, Replika, and Kindroid are meant for users 18 and older. That said, platforms typically don't require robust age assurance or verification beyond selecting one's birthdate or year of birth, so it's easy to gain access to more mature companions.
Character.AI, which is being sued by parents who claim their children experienced severe harm by engaging with the company's chatbots, does have parental controls and safety measures in place for users younger than 18. (Common Sense Media does not recommend any companion use for minors.)
How can you interact with an AI companion?
Generally, depending on the platform you've selected, you can design your own chatbot or engage with one built by and made public by another user. Some platforms allow you to do both things. You may be able to talk to the chatbot via text, voice, and video.
When designing or choosing a companion, you'll likely see common archetypes. There are anime characters, popular girls, bad boys, coaches, best friends, and fictional and real-life pop culture figures (think Twilight's Edward Cullen and members of the K-pop band BTS, respectively).
Some platforms controversially allow users to talk with chatbots presented as mental health therapists, which they are not, and which would be illegal for any human being to do without the proper credentials.
People don't just use their companions for one purpose, such as a romantic relationship. They might ask their "boyfriend" to help them with a class or work assignment. Or they might enact an elaborate scenario based on a popular book or film with a chatbot that's just a "best friend."
But things frequently get spicy. Last year, researchers analyzed a million interaction logs from ChatGPT and found that the second most frequent use of AI is for sexual role-playing.
Are there benefits or risks to having an AI companion?
Robert Mahari, now associate director of Stanford's CodeX Center and one of the researchers who analyzed the ChatGPT logs, said that more research is needed to understand the potential benefits and risks of AI companionship.
Preliminary studies, some conducted by AI chatbot and companion companies, suggest such relationships may have emotional benefits, but the results have been mixed and experts are worried about the risk of dependency.
Even if research can't move as quickly as consumer adoption, there are obvious concerns. Chief among them for Mahari is the inherently unbalanced nature of AI companionship.
"I really think it's not an exaggeration to say that for the first time in human history we have the ability to have a relationship that consists only of receiving," he said.
While that may be the appeal for some users, it could come with a range of risks.
Licensed mental health counselor Jocelyn Skillman, who also researches AI intimacy, recently experimented with an AI-powered tool that let her simulate different AI use cases, like a teen sharing suicidal thoughts with a chatbot. The tool is designed to provide foresight about the "butterfly effects" of complex situations. Skillman used it to explore AI-mediated relationships.
While each scenario Skillman tested began with what she describes as "emotional resonance," they variously ended with the hypothetical user becoming constrained by their relationship with AI. Her findings, she said in an interview, illustrate the potential "hidden costs of AI intimacy."
Dr. Rachel Wood shared her own list of key possible harms with Mashable:
Loss of relational and social skills. Confiding in a nonjudgmental chatbot can be alluring, but Wood said the one-sided relationship may erode people's patience with the human beings in their lives, who have their own interests and desires. AI companionship may also compromise people's ability to negotiate, sacrifice, and resolve conflict.
Less positive risk-taking in human relationships. Human relationships are challenging; they can involve misunderstandings, rejection, betrayal, and ghosting. Those who seek safe harbor in a chatbot may stop taking important and fulfilling risks with their human relationships, such as making a new friend or deepening a romantic partnership.
Unhelpful feedback loops. AI chatbots can make users feel like they're processing intense emotions in a private, affirming way. But this experience can be deceptive, especially when the user doesn't actually integrate and move beyond whatever confessions they've made to a chatbot. They may unintentionally reinforce their own shame if they only talk to a chatbot about topics they worry can't be discussed with the humans in their lives, Wood said.
Sycophancy. Chatbots are generally programmed to be flattering and affirming. Known as sycophancy, this design feature can be dangerous when an AI chatbot doesn't challenge a user's harmful behavior or when it convinces them of delusions.
Privacy. Read the terms of service very carefully, and assume that anything you share with an AI chatbot no longer belongs to you (see: private ChatGPT logs indexed by Google search). Your very personal conversations could be used for marketing, training the platform's large language model, or other instances that the company hasn't imagined or developed yet.
Wood said she's already seeing significant and fundamental changes in how people value the hard work of real relationships versus the "quick and easy" framework of synthetic ones. If you reach that territory while using an AI companion and aren't as interested in tending to your human relationships, it might be time to reconsider the role AI intimacy is playing in your life.
