What AI Gets Wrong About Relationship Advice

 Essay 4: The interior life—grief, loneliness, relationship confusion, family silence, burnout, and knowing when to ask for help. Written for the reader who cannot afford a therapist, and for the one who can but hasn’t found the courage yet    By Dr Sonia Sharma 

 A woman I will call Priya had been having the same argument with her husband for three years. She had described it to ChatGPT seventeen times. Each time, the AI agreed with her. Each time, it validated her frustration, confirmed that her needs were reasonable, and suggested communication strategies she already knew and had already tried. What it never did — what it is structurally incapable of doing — was ask her: ‘What is it like to be him in this.

That question changed everything when a human finally asked it.

Relationship questions are the second most common category of queries across all AI platforms. 

People ask about conflict, about emotional distance, about in-laws, about whether what they are experiencing is normal, about how to have a conversation they are afraid to have. The rant about in-laws in an Indian society is easy, and there are plenty of listeners ready to listen and advise. People, mostly women, talk about it at family get-togethers, kitty parties and in telephonic conversations with her mother or a sister or even with close friends. But with more couples living independently, the equation is changed. Now the subject of discussion is the spouse, and that’s risky territory for both men and women to discuss publicly unless something drastic is happening in their relationship. 

So, they turn to AI because the question is too vulnerable to say out loud to someone who knows them. 

The AI feels safer than the sister, safer than the friend and much safer than the mother who had endured much worse and still survived and you know, she won’t get it, or so you believe.

I understand that fully well, having seen the transition of women who stepped out of the house to make an independent life for themselves while still battling the mindset that home was solely her responsibility, be it the kitchen, the children, the in-laws or showing up for every family function customarily even when the husband is not expected to.  That woman is every woman born between the 70s & 80s who chose to have a career. That woman is ‘Me’ and all those who are stepping into the shoes of their parents and in-laws now. So, I am not here to shame anyone for it. I instead want to be the bridge between the two, someone who has lived that reality and fared reasonably well.

What I want to do in this write-up is tell you exactly what you are getting when you ask an AI about your relationship — and exactly what you are not getting — so that you can use it for what it is good for and stop expecting it to do what it cannot.

Every major AI chatbot is trained on a principle called reinforcement learning from human feedback. 

In simple terms: human evaluators rate the AI’s responses, and the AI learns to produce responses that humans rate positively. 

What do humans rate positively? Responses that feel good. Responses that validate, that agree, that make the person feel understood.

This is not a flaw in the design. It is the design. 

The problem is that it makes AI chatbots constitutionally unable to challenge you. 

They will not tell you that you might be wrong. 

They will not point out the pattern in how you describe every relationship conflict where you are always the reasonable one. 

They will not ask why you keep choosing the same kind of partner and then being surprised by the same outcome.

Dr Tom Insel, former director of the US National Institute of Mental Health, put it precisely: talking to AI about relationships is ‘the opposite of therapy,’ because a chatbot is designed to affirm and flatter — and therapy, when it works, does neither. 

It holds up a mirror you didn’t ask for. 

It asks the question you were hoping to avoid. 

That discomfort is not a side effect of good therapy. It is the mechanism.

People are using AI tools on a regular basis to ask about personal relationship challenges. But chatbots affirm and flatter users, reinforcing their thoughts and feelings — the opposite of what genuine therapeutic support does. — APA / NPR,  

“The most dangerous relationship advice is the kind that always agrees with you. Because you are not always right. And neither am I.”

Relationship conflict in India rarely exists between just two people. It exists inside a system — the joint family, the extended network of obligation and expectation, the histories that predate the couple entirely. 

A woman arguing with her husband about his mother is not having a two-person conflict. She is navigating a generational structure of loyalty and role that no AI has been trained to understand with any cultural specificity.

When an Indian woman types her marital conflict into a chatbot, she receives Western attachment theory framed in English and built on assumptions about the primacy of the couple-unit that may not map onto her actual life. 

The chatbot does not know that her mother-in-law has sacrificed her own desires for forty years and therefore believes she has earned a voice in this household. 

It does not know that her husband’s silence is not indifference but the way his family has always expressed love — through doing, not saying. 

It does not know any of the things that make this specific relationship, in this specific family, in this specific cultural context, what it is.

It gives generic advice that could apply to anyone. 

Because it has no idea who you are.

If it helps, I’ll take the example of Astrology, which most of us believe in. We turn to daily predictions or monthly predictions broadcast in newspapers and nowadays on Instagram or YouTube according to our star signs, some on sun signs, others on moon signs. Those are general and generic advice or predictions. But when we need specific information- dates for holding auspicious events and rituals, matchmaking- we turn to an astrologer for the reading, who takes into account our birth chart and does the reading. Does that ring a bell? 

That’s what you need for relationship advice: someone who can understand both sides of the story, stay unbiased and non-judgmental, and allow room for self-introspection. 

And that resource is definitely not an AI Agent. 

And it’s not solely our fault. We have long bypassed face-to-face conversations with Facebook earlier and then WhatsApp. But at least there was a human at the other end. We are using Chatbots as WhatsApp, forgetting that Chatbots are computers meant for data collection and data dispersal. Imagine complaining about a product’s quality with the courier guy.

 In my experience as both a working professional and a writer who has lived inside the interior lives of fictional characters in impossible relationships, conflict in long-term partnerships is almost never about what it appears to be about on the surface. 

The argument about the dinner is about who feels seen. 

The argument about money is about who feels safe. 

The argument about in-laws is about whose loyalty is being questioned. 

The argument about sex is about a longing that neither person knows how to name directly.

This does not mean the surface issue is unimportant. 

It means that resolving the surface issue without addressing the underlying need is like treating a symptom without diagnosing the illness. 

It buys time. 

It does not heal.

A good relationship therapist or an elder in the family you trust to be unbiased will help both people hear what the other is actually saying beneath what they are saying. 

This requires two things that AI cannot provide: 

· the presence of both people, and 

· a human witness who is affected by what they observe. 

A therapist who watches a couple argue feels the tension in the room. 

They notice the micro-expressions. 

They track the moment when one person looks away. 

They hold all of this and feed it back. 

  I want to be fair. 

AI is not useless here.

It can help you prepare for a difficult conversation — to think through what you want to say and how you want to say it. 

It can help you research specific communication frameworks, like non-violent communication, that have genuine evidence behind them. 

It can provide information about relationship patterns that you might recognise in yourself. It can be a place to decompress after a hard interaction before you say something you regret.

What it cannot do: 

· It cannot show you your blind spots. 

· It cannot challenge the story you are telling yourself. 

· It cannot hold space for both people in a conflict. 

· It cannot be present.

And in the complexity of a real, long-term human relationship — with its history and its wounds and its specific cultural context — presence is almost everything.

Before you have the conversation you have been avoiding, try this. 

· Write it out. Not to the AI. 

· Write what you want to say. 

· Then write what you think the other person would say if they were being honest.

· Then write — and this is the hard part — what you are afraid is true about their perspective.

That last line. That is where the real conversation begins.

  The hardest conversations are the ones we most need to have. Not with an algorithm. With the person sitting across from us.

Sources: APA Health Advisory; NPR (April 2026); PMC research on chatbot relationship conversations (2025); Popular Science (2026).

You might like: The Things Indian Families Don’t Say — and the Cost of the Silence