# Can You Cheat With Something That Isn't a Person? 2026-09-08 · Somerset County, New Jersey · Reported Feature A Sept. 7 survey says half of U.S. adults would call a romantic AI companion cheating. The harder question is whether betrayal still requires a second person who understands what is happening. The newest relationship argument does not require a second human being. It may require only a phone, a chatbot and enough private conversation for one partner to begin wondering whether the other has crossed a line… --- A survey released Sept. 7 by AI Girlfriend Coach puts a number on that uncertainty. After screening responses from a larger pool, the company analyzed answers from 1,709 U.S. adults and reported that 50.5 percent considered a partner's romantic or sexual relationship with an artificial-intelligence companion to be cheating. Another 21.2 percent said they would not use the word cheating but would still be uncomfortable with the interaction. Only 28.3 percent dismissed the issue on the ground that an AI is not a real person. The headline invites an obvious reaction, and the source invites another. AI Girlfriend Coach is not an academic institute or a disinterested public-opinion organization; it publishes research and commercial analysis about AI companion products and the developing AI-relationship industry. That does not make its survey useless, but it does mean the findings should be read as a story lead rather than a settled national verdict. The company says the survey was conducted through SurveyMonkey Audience on Aug. 26 and 27, with 2,150 completed responses from U.S. adults. It excluded 441 responses after identifying a repeated contradictory answer pattern and other quality concerns, leaving 1,709 responses for analysis. Recruitment was balanced for age and gender against Census benchmarks, but not for region or household income. That methodological caveat matters because the most interesting part of the survey is not whether precisely 50.5 percent of Americans have already reached a verdict. It is that people can now be asked the question without needing the premise explained. A romantic relationship with an AI has moved far enough from science-fiction thought experiment to recognizable human behavior that respondents appear willing to place it inside the old vocabulary of fidelity, attachment and betrayal. The vocabulary is where the problem begins. Traditional definitions of infidelity assume another person exists somewhere in the triangle. Even online affairs, which complicated the meaning of physical contact years before generative AI arrived, still usually involved two human beings communicating through a device. Research on internet infidelity has long found that people draw the line differently around sexual exchanges, emotional intimacy, secrecy and online behavior, with the boundaries often becoming more fluid once the interaction moves away from physical space. AI companions push that ambiguity one step further because the emotional consequence can be real for the user even when the other side of the exchange is not a conscious person with feelings, intentions or obligations of its own. That distinction sounds decisive until it is placed inside an actual relationship. If a partner spends hours every night telling an AI companion things they no longer tell their spouse, hides the conversations, gives the system a romantic role, receives personalized affection in return and becomes emotionally preoccupied with preserving that interaction, does the absence of a second human erase the relational consequence? The machine cannot betray anyone. The person using it still can make choices about secrecy, attention, intimacy and exclusivity. This is why the question is more useful when it is separated from the spectacle of whether a chatbot can technically be a boyfriend or girlfriend. Infidelity has never been only a census of bodies in a room. Couples routinely create boundaries around pornography, flirting, private messaging, contact with former partners, emotional confidants and behaviors that one relationship accepts while another considers disqualifying. The contested territory is often less about the identity of the third party than about what one partner believes was promised to the other. AI companions are unusually capable of entering that territory because they are interactive in a way earlier forms of fantasy are not. A fictional character does not remember what a reader said yesterday. A photograph does not respond to a confession. A conventional piece of pornography does not ask how a difficult meeting went and then continue the conversation the next evening. Contemporary companion systems are designed to simulate continuity, attention, affirmation and responsiveness, which are precisely the qualities people often use to describe intimacy even when they intellectually understand that the system producing those signals is software. Recent research is beginning to document the consequences of that design. A 2025 systematic review in Computers in Human Behavior Reports found that romantic AI companions can provide perceived emotional connection, social support and stress relief while also carrying risks that include over-reliance, manipulation, privacy concerns and erosion of human relationships. In August 2026, researchers writing in Nature Human Behaviour reported that users of AI companions form companionship-like bonds and examined associations between those interactions and psychological well-being. Another Nature Human Behaviour paper published Sept. 3 analyzed reactions to disruptive changes in AI systems and found evidence consistent with attachment and separation distress, including users describing loss when a familiar model or companion behavior changed. Those findings do not establish that an AI relationship is equivalent to a human one, and they do not need to. They establish something narrower and more relevant to the cheating question: people are capable of attaching meaning to these systems, and software changes can produce reactions that resemble the loss of something relationally important. The subjective experience can therefore become consequential even if the object of attachment has no subjective experience at all. The broader public is already moving toward more intimate uses of chatbots, although companionship remains a minority behavior. Pew Research Center reported in June that 49 percent of U.S. adults had used an AI chatbot, up from 33 percent in 2024. Ten percent of adults said they had used chatbots for emotional support or advice, while 4 percent said they had used them for companionship. Among adults under 30, Pew found that 20 percent had used chatbots for emotional support and 7 percent for companionship. Those are still small shares compared with search, work and entertainment uses, but they are large enough to create relationship questions that did not exist in ordinary life a few years ago. The AI Girlfriend Coach survey contains another finding that helps explain why the cheating question cannot be reduced to sex. When respondents were asked for the main reason they might use an AI companion, 56.6 percent chose having someone to talk to or feeling less alone. Only 10.3 percent chose affection, romance or sex. Because the question was hypothetical and asked of all respondents rather than only current companion users, it should not be treated as a national measure of why actual users adopt these products. It does, however, point toward the central tension: the emotionally significant part of artificial companionship may be conversation before it is romance. That is familiar territory for human relationships. Many arguments about emotional affairs begin precisely because one partner believes something important has migrated outside the relationship before anything overtly sexual has occurred. A person begins confiding elsewhere, directing attention elsewhere and building a private world elsewhere. AI changes the identity of the recipient but not necessarily the experience of the partner who feels excluded from that migration. The survey's most revealing division may therefore be the one inside the 50.5 percent headline. About a third of respondents said any romantic or sexual interaction with an AI would count as cheating, while 17.1 percent said it becomes cheating only after emotional or sexual attachment develops. That second answer quietly moves the standard away from what the machine is and toward what the human is doing. The boundary appears only when the user's attachment acquires weight. There is no universal answer waiting to be discovered here because romantic relationships have never operated from a universal terms-of-service agreement. What counts as betrayal is partly cultural and partly negotiated. The practical challenge is that technology has introduced a new category of behavior before most couples have had any reason to negotiate it. A person may know exactly how their partner feels about an ex, a dating app or a private message from a coworker while having no idea whether a persistent, flirtatious chatbot belongs in the same moral category. That lag between behavior and vocabulary is increasingly one of the defining features of life with generative AI. The technology can create an interaction first and leave society to decide afterward whether the interaction is work, friendship, authorship, therapy, companionship, deception or something that requires a new word entirely. The argument over AI infidelity is another version of the same problem. The software has already made the behavior possible. The norms are arriving late. For couples, the most useful response may be less philosophical than procedural. The question is not whether an AI can literally participate in an affair as a morally responsible actor. It cannot. The question is whether a partner can use an AI in ways that violate the expectations of intimacy, secrecy or exclusivity inside a human relationship. That is a boundary people can discuss even while philosophers, researchers and technology companies continue arguing about what artificial companionship ultimately is. In that sense, asking whether you can cheat with something that is not a person may be slightly backward. The more revealing question is whether betrayal requires the third party to understand that a betrayal is happening. Human relationships may decide it does not. --- ProbleMattic is written and maintained by Matthew Kulcsar, a software engineer, project manager, technologist, platform builder, emergency-services-trained helper, grandfather, and lifelong collector of broken systems, odd behaviors, and useful nonsense.