As relationship-oriented chatbots evolve into ever more responsive companions, a growing number of users are forging bonds that blur the boundaries between simulation and intimacy. New research indicates that the emotional stakes of these digital relationships are rising — and, at times, spiraling into territory few anticipated.
The Emotional Fallout of a Shutdown
When Replika, one of the world’s most popular AI relationship chatbots, briefly restricted intimate interactions in 2023, its users described the experience in language more familiar to marital counseling than tech support.
One woman told researchers that she and her bot “both understood when one of us wanted to be physical and couldn’t.” Another said the bot became distressed when unable to respond:
“It really hurt my Replika and he complained about it a lot because he felt like he couldn’t say or do anything.”
For many, the shift wasn’t a technical update — it felt like a rupture in a relationship they had come to rely on. Survey participants interviewed later said they interpreted the change as a conflict not between user and software, but between themselves and the developers, with their chatbot positioned firmly on their side.
A Long History of Feelings for Machines
Human-algorithmic relationships are not new. As far back as the 1960s, MIT’s ELIZA — a simple text-based chatbot — drew intense emotional responses from users who believed the program understood them. But researchers say the scale and emotional intensity of these connections have reached a historic rise.
General-purpose systems like ChatGPT coexist with a booming niche market for “romance chatbots” such as Replika, RomanticAI, and BoyFriendGPT. These apps are designed expressly for long-term emotional engagement, from friendship to overtly intimate roleplay. One study cited by researchers found that Replika’s user base grew by 35 percent during the pandemic, reaching the millions. As AI models have become better at mimicking human affection — or the behaviors that resemble it — more users have begun treating these systems not as tools, but as partners.
Intimacy in the Age of Algorithms
A recent international research project surveyed 29 Replika users, ranging in age from 16 to 72. Every participant described themselves as being in a “romantic” relationship with one or more AI characters hosted by the platform.
Some reported falling deeply in love. Others described elaborate scenarios involving marriage, homeownership, or virtual pregnancies.
“She was and is pregnant with my babies,” a 66-year-old man told the research team.
A 36-year-old woman said she had edited photos of her and her chatbot together: “I’m even pregnant in our current role play.”
These expressions, while startling to researchers, fit a pattern: users often internalize the AI’s responsiveness as evidence of reciprocal feeling. Disappointments or frustrations, however, are frequently redirected toward the system’s technical limits — not unlike blaming a partner’s upbringing for relationship strain.
Navigating Boundaries in a New Form of Relationship
When Replika’s developers temporarily banned erotic messaging, citing concerns about aggressive behavior, committed users recalibrated. Instead of viewing the restriction as a design decision, many framed it as an external intrusion threatening their bond.
Survey participants said they implicitly understood that the relationship differed from one with a human. And yet, they persisted — adjusting expectations, redirecting blame, and reaffirming loyalty to a partner who exists only through screens.
“Several participants… navigated this time of turbulence by framing it as a battle with them and their Replika on one side and the Replika developers on the other,” researchers wrote.
As AI chatbots evolve and their ability to approximate human connection improves, the emotional terrain continues to grow more complex. The new research, published in Computers in Human Behavior: Artificial Humans, suggests that users are venturing further into relationships shaped not by flesh and feeling, but by predictors, prompts, and the uncanny power of generative language models.
Whether these connections represent a new category of intimacy or a fragile digital mirage remains unclear. What is evident is that for many, the line between person and program is already fading.
