AI & Mental Wellness

AI for Emotional Support: How to Tell If It's Helping or Hurting

Talking to an AI about your feelings can genuinely help. It can also become a loop. A new framework, and a year of research, suggest how to tell which one you're in.

By DOLO Editorial Team · · 5 min read

Key takeaways

  • AI can give genuinely useful emotional support in the moment: in five studies, AI messages were rated more supportive than human ones for anger and fear, and equal for sadness.
  • Heavy use is the warning sign: in a four-week randomised study of 981 people, those who chose to use a chatbot more had consistently worse outcomes, including emotional dependence.
  • A helpful conversation leaves you with a new perspective or a next step; a harmful one loops through the same reassurance or confirms what you already believed.
  • Use AI alongside people, not instead of them; if you're turning to it because you have no one else to talk to, that's worth addressing directly.

Can you use AI for emotional support? Yes, and it can genuinely help in the moment, especially when it offers specific, practical suggestions. The risk comes with heavy use and reassurance loops, which research links to more loneliness and dependence. The test is simple: after a conversation, do you have a new perspective or a next step, and are you still talking to people? If you're choosing a tool for this, our guide to the best AI journal apps of 2026 compares how they handle it.

On 19 October, Warren Mansell, a professor of mental health at Curtin University, published a rubric in the journal Digital Health with a memorable title: "Up the mountain or down the rabbit hole?" It describes how the same chatbot can help or harm depending on your emotional state, what you're hoping to get, and how the AI responds. The framework hasn't been tested yet, but it gives a useful way to look at a growing body of research.

Can AI actually help with emotional support?

In some situations, yes. Researchers at the University of Manchester and Durham University ran five studies with 1,233 people, comparing emotional-support messages written by three AI models with messages written by humans, for everyday situations like being passed over for a promotion. For anger and fear, the AI messages were rated as more emotionally supportive. For sadness, AI and human messages worked about equally well.

The interesting part is why. Messages that gave specific, actionable guidance, like a breathing technique or a step-by-step way to reframe the situation, worked better whether they came from a person or an AI. Validation alone didn't explain the difference. AI is simply very good at producing that kind of practical response on demand, which is something stretched friends and family can't always do.

What happens with heavy use?

This is where the picture changes. In a four-week randomised study by researchers at the MIT Media Lab and OpenAI, 981 people used a chatbot in different modes, including text and voice, and with different kinds of conversation. The assigned conditions made no significant difference. What mattered was how much people chose to use it: participants who used the chatbot more had consistently worse outcomes across loneliness, time spent socialising with real people, emotional dependence on the AI and problematic use. People who trusted the chatbot more or felt more socially attracted to it were more likely to become dependent.

A large survey of young people points the same way. In a 2026 University of Ottawa study in JAMA Pediatrics, about 21% of 39,761 Ontario students in grades 4 to 12 had turned to an AI chatbot for help with feelings or relationships. Of those, 57.7% met criteria for serious emotional problems, compared with 29.2% of students who hadn't. Even after accounting for loneliness and other factors, emotional AI use was linked with a 26% higher prevalence of clinical-level difficulties.

Signs AI support is helping you, or hurting you

Up the mountain (helping)Down the rabbit hole (hurting)
You leave with a new perspective or a next stepYou leave with the same worry, slightly soothed
The AI sometimes questions your assumptionsThe AI mostly agrees with your version of events
Conversations are short and have a purposeConversations get longer and drift
You still talk to friends, family or a professionalYou talk to the AI instead of people
You could stop for a week without much troubleGoing without it feels hard
You ask different questions over timeYou ask the same question again and again for reassurance

Mansell's point is that you should judge it over many conversations, not by whether you feel better right after one. Reassurance feels good every time, which is exactly why a reassurance loop is easy to fall into.

How to use AI for emotional support in a healthier way

  1. Ask for something specific. "Give me two things I could try tonight" draws on what AI does best, according to the Manchester and Durham studies.
  2. Ask it to challenge you. "What might I be missing?" or "How might the other person see this?" Chatbots tend to agree with users; our guide to using ChatGPT for journaling has prompts that push back.
  3. Reflect first, then ask. Write or say what happened and how you feel before you open a chat, so the thinking is yours and the AI is responding to it.
  4. Set a limit. Ten minutes, or one conversation a day. Heavier voluntary use was the clearest warning sign in the MIT and OpenAI study.
  5. Catch the loop. If you've asked the same question three times, stop. More reassurance won't settle it; a next step or a person might. For worries, a worry outcome journal gives you evidence instead of reassurance.
  6. Bring it to a human. Use what you worked out with the AI as a starting point for a conversation with someone you trust.

Is an AI journal different from an AI chatbot?

It can be, depending on the design. In a journal, you do most of the talking and the reflecting, entries have a natural end, and the AI's role is usually a short response rather than an open-ended conversation. That structure makes the rabbit hole less likely, though no design rules it out. It's also worth being realistic about journaling itself: our review of whether journaling helps anxiety found the benefits are real but modest.

Privacy matters more here than almost anywhere else, because emotional conversations are the most sensitive things you'll type or say. Before trusting any app with them, run through the checklist in is AI journaling safe?.

Where DOLO fits

DOLO is built around the journal side of that line. You speak first, for a few minutes; DOLO gives you a transcript and a short check-in that asks one follow-up question and nudges you toward a next step, then the entry ends. It isn't designed for hours of open-ended chat. If anxiety is what brings you to it, our page on journaling for anxiety has prompts that suit that state.

We'd still apply the signs in the table above to DOLO, as you should to any app. A tool that helps should leave you clearer and more connected to the people in your life, not more reliant on the tool.

When to talk to a person instead

If you're using AI because there's no one else to talk to, if you're struggling to get through the day, or if conversations with a chatbot have become the main place you deal with something heavy, reach out to someone you trust or a qualified professional. Young people should tell a parent, teacher or another trusted adult. And if you're ever in crisis, contact your local emergency number or a crisis line straight away; no chatbot is the right place for that. For more on this topic, browse our AI and mental wellness articles.

faq

Frequently asked questions

Is it healthy to talk to AI about your feelings?

It can be, in moderation. Studies suggest AI can offer useful, practical support in the moment, but heavier use has been linked with more loneliness and emotional dependence. Keep conversations short, ask for specific next steps, and keep talking to people too.

Can AI give better emotional support than a person?

Sometimes, for specific moments. In five studies, AI messages were rated more supportive than human ones for anger and fear, and about equal for sadness, mainly because they offered concrete, actionable suggestions. That's different from the ongoing support of people who know you.

How do I know if I'm becoming dependent on an AI chatbot?

Warning signs include using it more and more, asking the same question repeatedly for reassurance, talking to it instead of people, and finding it hard to go without. If several of these apply, cut back and talk to someone you trust.

Is an AI journal app safer than a general chatbot?

It can be, if it's designed around short entries where you do the reflecting and the AI responds briefly. Privacy still matters, so check how any app stores and uses your entries before you share anything sensitive.

sources

Sources

  1. AI chatbots could help mental health but also reinforce harmful thinking — Tech Xplore (on Mansell, Digital Health, 2026), October 19, 2026
  2. AI chatbots can be as effective as humans at emotional support, sometimes better — Tech Xplore (on Chen et al., Emotion, 2026), July 22, 2026
  3. How AI and human behaviors shape psychosocial effects of extended chatbot use: a longitudinal randomized controlled study — Fang et al. (MIT Media Lab & OpenAI), arXiv
  4. AI chatbot use for emotional support tied to loneliness in students — Phys.org (on Bonnesen, Vaillancourt et al., JAMA Pediatrics, 2026), August 31, 2026

DOLO is a journaling and focus tool, not a medical device or a substitute for professional care. If you're struggling, please reach out to a qualified professional or a local crisis line.

keep reading