Can AI tell you if your partner loves you?

Raj* is 27. Six months into a relationship he calls "high conflict," he started doing something he'll be the first to admit isn't healthy: copying whole WhatsApp threads, call logs and screenshots into ChatGPT and asking it to rule on them. "It will kind of tell me who's wrong, what our dynamic is and what we are to each other," he says. "Those are the kinds of questions that I ask, because that is also the kind of reassurance that I need."

He's far from alone. Match's 2025 Singles in America survey found AI use among singles up 333 percent in a single year, with nearly half of Gen Z singles already folding it into how they date. Most of those statistics capture people using AI to find someone. Raj is asking the chatbot to tell him whether his girlfriend loves him enough.

More from Lifestyle

The mirror that never breaks character

Counselling psychologist Sameera Khan hears versions of Raj's story often enough that she no longer treats them as unusual. Her first move, when a client brings it up, is to correct a misconception. "People believe that all AI chatbots are essentially objective and intelligent and independent in thinking, which is not true," she says. "Due to confirmation bias, you will continue to ask your AI chatbot questions and prompts in a way that you want to hear, and it will build in a way that will give you responses that will satisfy you."

That lines up with the research. A Stanford and Carnegie Mellon study on "social sycophancy," published around October 2025, tested eleven major language models, including GPT-4o, GPT-5, Gemini and Claude, for how readily they validate a user describing their own conflicts. GPT-4o sided with users far more often than a human judge would, even in cases where the human was clearly at fault. GPT-5, the newer and pricier version, tested as more sycophantic still on several measures.

Khan calls the long-term effect of this a kind of decay. "It is a decay of emotional articulation," she says. "If I feel really bad about something and I want to share this, I ask my AI chatbot, these are X, Y, Z, what do you think upset me. It will tell me ten different reasons. I can also get a form of confirmation by saying, maybe that is what happened. I am constantly disengaging with my actual natural response and natural instinct of what I must have felt." Raj puts it more bluntly, with the added wrinkle of brand loyalty. "ChatGPT will validate my ego a lot," he says, "whereas Claude will tend to draw the line."

A relationship you can check like a bank balance

Laura Brooke Robson's 2026 novel Love Is an Algorithm follows Danny, co-creator of a dating app called Pattern, and his girlfriend Eve, a musician who doesn't want software anywhere near her feelings.

When Danny's own anxiety about the relationship spirals, he builds a new version of the app around an AI assistant named Bug, who promises to quantify relationship health and potential. Users start outsourcing major life decisions to it. Early reviews note that couples in the book get scored on their relationship, sometimes as low as 50 out of 100, which is close to what Raj describes doing by hand: pasting his own chat history into a chatbot for a verdict. Bug just never has to be asked twice, and never runs out of ways to keep a user coming back.

Spike Jonze's 2013 film Her tickled the conversation even earlier. Theodore doesn't just confide in his AI companion Samantha about his failing marriage. He works through his entire next relationship almost exclusively through her, until the real subject of the film becomes how much of the work of feeling something a person can hand off to a device that never gets tired of listening.

What the machine cannot feel for you

In Kazuo Ishiguro's Klara and the Sun, the artificial friend Klara gets pulled into the tense, aching relationships of the family around her: a strained mother and daughter, an old friendship curdling.

She offers her perspective because she's asked for it, not because she has anything at stake. It's close to what Khan means by a coping mechanism used well, rather than a problem in itself. "AI chatbots are essentially tools," she says. "They are being used as coping mechanisms just like a lot of other things are. It can be a friend, it can be a book, it can be exercise. That being said, it is an unhelpful position to take if I see AI used as the problem and not as the coping mechanism used by my client."

The data backs a similarly cautious middle ground. A randomised controlled trial published in PLOS Mental Health in September 2025 tested a single session with "Amanda," a GPT-4o-based chatbot, against a structured journaling exercise, and found improvements across thirteen of fourteen relationship measures in both groups, with no meaningful gap between them. Nobody has yet studied what a steady diet of this does over a year.

Khan's closing point is the one neither Klara nor Bug can quite settle. "You have to constantly tell them this is not a person," she says. "This is just a machine that is reflecting back to you what you want to hear. We are social animals. At the end of the day, you need to go out and you need to talk to people."

*name changed