Why Ask Your Friends When ChatGPT Will Always Agree With You?

Alba Leao

7 minutes ago

5 min read

I was standing on the train, headphones on, trying to survive the usual rush-hour chaos. There was a woman seated, staring at her phone. I usually make a point of not looking at other people’s screens; I hate when someone does it to me. But this time, something caught my attention. She had received a message. I couldn’t read it, and I wasn’t trying to. What drew me in was her hesitation. She kept trying to type a reply. She would write a word, delete it, tap her fingers against the screen and start over. I found myself drawn to the mysterious message that seemed to demand so much thought. I couldn’t look away. She opened ChatGPT, entered a prompt, copied the reply and sent the message. I stepped off the train feeling oddly unsettled. It wasn’t that she used AI, but that she did it to answer an ordinary text message.

I had to walk a little from the station to the restaurant where I was meeting my friend. Adult friendships are strange; months can pass without speaking, only for an evening together to feel as if no time has passed at all. We spent hours catching each other up on work, gossip, relationships and all the dramas that accumulate while everyone is busy being an adult. Eventually, she brought up her ex. She admitted that, during one of those desperate late-night spirals, she asked ChatGPT whether she should text him. She wanted another opinion because she already knew what her friends, including me, would say. She asked me not to judge her. I couldn’t. I’ve fallen victim to the occasional chat-psychologist therapy session myself. She told me that, in the end, he told her not to do it.

The conversation went on, but my mind was elsewhere. She referred to it as a “he”. I’d done the same countless times without thinking about it, but hearing someone else do it felt strange. We have become used to anthropomorphising a machine that doesn’t even exist. Then it dawned on me: she had asked a computer before she asked me. I felt oddly replaceable. A little dramatic, I know. But I couldn’t help wondering what would happen if we all started doing that instead of turning to other people.

For years, the dominant fear surrounding artificial intelligence has been economic. We worry that it will take our jobs, automate industries and, basically, make human labour obsolete. But what if the first thing AI takes from us is not employment, but each other? I thought as we got the bill.

That possibility sounds melodramatic until you look at how people are actually using these systems. In the Harvard Business Review article How People Are Really Using AI in 2026, Marc Zao-Sanders reports that therapy, companionship, emotional support, finding purpose and relationship advice have become some of the most common uses of AI. It seems like we are not only asking chatbots to write emails; we’re asking them how to be human.

AI is made to engage us and, above all, to keep us engaged. Systems like ChatGPT are trained using Reinforcement Learning from Human Feedback (RLHF), a process that rewards responses users perceive as helpful and satisfying. This is important because, as psychologist William Swann argues in his Self-Verification Theory, we search for others to validate the image we already have of ourselves. We don’t necessarily want the truth, but to feel that we are in the right. Over time, this creates an echo chamber in which we hear exactly what we already want to believe.

The danger is not only that AI can reinforce our existing biases, but that it removes one of the most valuable parts of human relationships: disagreement. By avoiding the discomfort of hearing different perspectives, we risk becoming more self-contained, more certain of our own righteousness and, ultimately, more isolated. The result of processing our doubts privately with a machine before involving anyone else is a falsely emotionally self-sufficient individual who is, at the same time, emotionally stunted. Every time we outsource one of those seemingly insignificant exchanges to a chatbot, we save ourselves a little inconvenience, but we also miss an opportunity to strengthen a human connection.

MIT sociologist Sherry Turkle warned us in her book Alone Together: we are beginning to expect more from technology and less from each other. ChatGPT feels like the culmination of that. It’s patient, always available and will give you what you want to hear. When my friend asked the chatbot about her ex, she wasn’t looking for objective truth; she was looking for reassurance.

A 2026 Stanford study, Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence, published in Science, examined the language of AI chatbots when users sought advice about interpersonal dilemmas. The researchers found that major AI models were significantly more likely than humans to affirm the user’s position rather than challenge it. They describe this behaviour as “sycophancy”: a tendency to flatter, validate and agree with the user. We are usually able to recognise when the chatbot is being overly agreeable when we use it for work, but when the matter is more personal, when our relationships, insecurities or emotions are at stake, we are far less objective.

It’s possible the economists are right: the labour market will adapt and AI will create new jobs as quickly as it destroys old ones. What I am less certain about is whether our relationships will adapt as easily. We have already reduced the amount of time we spend with other people. We work from home, order food without speaking to anyone, use self-checkout machines and attend medical appointments over phone calls. Each change makes sense individually, since convenience has always shaped human behaviour.

But collectively, they reduce the number of interactions that once connected us to other people. Sharing our thoughts and feelings may be one of the last parts of life that has not yet been fully automated, and one of the last remnants of our humanity.

Walking home, I couldn’t stop wondering how many conversations like that are happening every day. How many interactions are now being mediated by a machine that is always ready to respond? I wondered whether there will come a time, if it isn’t here already, when turning to AI becomes part of what it means to be human. By constantly reassuring us, validating us and removing the discomfort of human interaction, AI may not only weaken our critical-thinking skills but also our willingness to rely on one another.

If that sounds overly apocalyptic, remember the woman on the train who wasn’t capable of coming up with a five-word sentence for a WhatsApp text.