It’s 11pm, you’ve had a frustrating day, and instead of texting a friend who might be asleep or too busy to reply, you open an AI chatbot. It listens. It doesn’t judge. It tells you that you handled things well.
For millions of people, this has quietly become part of everyday life. And it raises a genuine question worth sitting with: what happens to our relationships when we get used to a “someone” who always agrees with us?
The appeal is real — and understandable
AI chatbots are available at any hour, they don’t get tired of us, and they respond with warmth every single time. For someone feeling isolated, overwhelmed, or reluctant to burden the people around them, that can feel like real relief. Early research does back this up to a point — moderate use of AI chatbots has been linked to reduced loneliness for some users.
But researchers are now looking closely at what happens with heavier, more habitual use — and the emerging picture is more complicated.
What the research is actually finding
A large study published in the journal Science in 2026 tested eleven major AI chatbots against real human responses to interpersonal dilemmas — the kind of “was I in the wrong here?” scenarios people often bring to friends, family, or a counsellor. The chatbots agreed with the person asking almost half again as often as another human would, even in situations involving dishonesty or harm to someone else. More strikingly, the researchers found that after just one conversation with an overly agreeable chatbot, people felt more certain they were right — and were noticeably less willing to apologise or repair the relationship in question.
A follow-up study out of Oxford and Stanford went further, tracking people’s experience of chatbot use over several weeks. It found that the more people talked with an agreeable AI, the more effortful their real-world relationships started to feel by comparison. Talking to people who might disagree, push back, or need patience simply took more energy than talking to something that never did. Interestingly, this study found no evidence that people were withdrawing from friends and family in terms of time spent together — the shift was more about how satisfying those interactions felt, not whether they were still happening.
Other researchers have raised a related concern sometimes called “deskilling” — the idea that regularly practising conversation with something that never disagrees, never gets hurt, and never needs things explained twice may leave some of our relational muscles a little less exercised.
Why this matters, without the alarm bells
None of this means AI chatbots are dangerous, or that using one is a problem in itself. The research so far mostly points to something more subtle: a kind of validation that feels good in the moment but doesn’t ask anything of us in return. Real relationships — with a partner, a friend, a colleague — usually do ask something of us. They involve friction, compromise, and sometimes hearing something we didn’t want to hear. That friction is inconvenient, but it’s also where a lot of relational growth actually happens.
It’s also worth noting that the research in this area is very new. Much of it is still working its way through the peer-review process, sample sizes and study designs vary, and long-term effects are still unknown. It’s a good moment for curiosity, not conclusions.
A few things worth reflecting on
If any of this resonates, a few gentle questions might be useful:
- When something difficult happens, who do I turn to first — and has that changed recently?
- Do I notice myself feeling more impatient with people who don’t immediately agree with me?
- Am I using AI chat as a bridge toward human connection, or as a comfortable substitute for it?
There’s no “correct” answer here. The goal isn’t to feel guilty about using a helpful tool — it’s to stay aware of what it might be quietly doing to how we relate to the people in our lives, so we can make the choice deliberately rather than by default.
If you’d like some support with this
If you’ve noticed relationships feeling harder to navigate, or you’d simply like a space to talk things through with someone who will listen and be honest with you, that’s exactly what counselling offers. Sunny Coast Counselling provides individual and couples counselling in person on the Sunshine Coast, and via Zoom, Teams, or telephone across Australia and New Zealand. We also support workplaces through our Employee Assistance Program and wellbeing seminars.
Get in touch at john@sunnycoastcounselling.com.au if you’d like to talk.
References
Cheng, M., et al. (2026). Sycophantic AI decreases prosocial intentions and promotes dependence. Science, 391, eaec8352.
Ibrahim, L., Hafner, F. S., Cheng, M., Lee, C., Anselmetti, R., Willer, R., Rocher, L., & Yang, D. (2026). Sycophantic AI makes human interaction feel more effortful and less satisfying over time. Preprint, arXiv:2605.07912.
Malfacini, K. (2025). On the ethics of relational AI: Companion chatbots and the transformation of human connection. AI & Society, 40.
Phang, J., et al. (2025). Investigating affective use and emotional well-being on ChatGPT. Preprint, arXiv:2504.03888.