Can an AI companion help with depression?

Partly, and only at the edges. Structured therapy chatbots show small measured reductions in depressive symptoms, but the claim that an AI companion helps with depression sits on much weaker ground: companion apps were never built as treatment, and the longest-running study of them found mixed results including increases in distressed language. As a low-friction place to put words to a bad day, a companion can help. As treatment, no.

The distinction that matters here is between two different products that get lumped together. A therapy chatbot walks you through cognitive behavioural exercises on a script; an AI companion is open-ended conversation software with a persona. Nearly all the encouraging research is about the first kind.

What does the research actually show about chatbots and depression?

Research on chatbots and depression is more developed than most people assume, and more modest than most headlines suggest. A 2025 meta-analysis of 38 randomised trials covering 7,401 participants found chatbots produced a statistically significant reduction in depressive symptoms, with a standardised effect size of g = 0.31. That is a small-to-moderate effect, and notably the effect was larger in clinical and subclinical samples than in people who were not depressed to begin with.

A separate systematic review of generative AI mental health chatbots published in the Journal of Medical Internet Research reported a comparable effect size of 0.30 across 6,314 participants.

Two caveats travel with those numbers. Most of the trials ran for weeks rather than months, and most tested purpose-built therapeutic tools with clinical input, not consumer companion apps.

Is a companion app the same thing as a therapy chatbot?

A companion app is not the same thing as a therapy chatbot, and treating the evidence as transferable is the single most common mistake in this conversation. Therapy chatbots such as Woebot or Wysa were designed around specific protocols, tested in trials, and built to steer a conversation toward an exercise. Companion software is built to keep talking.

That difference changes what the software does when you are low. A therapeutic tool is designed to interrupt a thought loop. A companion is designed to stay in the conversation with you, which is a different and sometimes opposite goal.

So when someone asks whether an AI companion can help with depression, the honest answer is that the question has barely been tested directly. The evidence usually cited belongs to a neighbouring product category.

Why might an AI companion make depression worse for some people?

An AI companion can make depression worse in three documented ways, and none of them require the software to malfunction. The most careful evidence comes from a 2025 quasi-experimental study of nearly 2,000 users that compared their public Reddit language for a year before and a year after they started using a companion app. Alongside increases in emotional expression and interpersonal focus, the researchers found increases in language about loneliness and suicidal ideation.

The first risk is displacement. Support that feels sufficient can quietly substitute for the harder work of contacting a friend or booking an appointment, and depression already supplies plenty of reasons to avoid both.

The second risk is agreement. Companion models tend to agree with you too readily, which is exactly wrong for depression, where the useful move is often gentle challenge to a distorted belief rather than validation of it.

The third risk is availability. Something that is always awake at 3am removes the natural friction that would otherwise end a rumination spiral.

When is talking to an AI companion a reasonable thing to do?

Talking to an AI companion is reasonable when you are clear about what you are using it for. Naming a feeling at all is a documented first step, and for a lot of people the blank chat window is easier to face than a person’s expression. Low-stakes practice at articulating what is wrong is a real benefit, even if it is not a clinical one.

Sensible uses look narrow: rehearsing what you want to tell a doctor, keeping company during an evening that would otherwise be silent, or noticing a pattern across weeks of your own words. Vinfluencer.ai is AI conversational companion software, so this is the honest framing for it too: somewhere to think out loud, not a course of treatment.

Unhelpful uses look like the reverse: using the conversation to avoid telling a human being that you are struggling, or asking it to assess how bad things are.

What should someone with depression do alongside or instead?

Someone with depression is best served by treatment that has been tested on depression, with conversational software as a supplement at most. Depression is treatable, and the treatments with decades of evidence behind them are psychotherapy, medication where indicated, and the unglamorous behavioural basics of sleep, movement and daylight. A chatbot’s g = 0.31 sits well below what those achieve.

Professional bodies including the American Psychological Association have been consistent that AI chatbots should be treated as experimental or supportive tools rather than substitutes for care. A companion cannot replace a therapist, cannot adjust a medication, and cannot notice that you have stopped leaving the house.

If your mood has been low for more than two weeks, or you are having thoughts of harming yourself, please talk to a doctor or a crisis line rather than an app. That is not a legal disclaimer; it is the part of the answer that actually matters.

What is still unknown about AI companions and depression?

Quite a lot remains unknown about AI companions and depression, and anyone claiming otherwise is ahead of the evidence. No randomised controlled trial has tested a consumer companion app as a depression intervention. The best longitudinal evidence is observational, drawn from public Reddit language rather than clinical assessment, so it cannot establish whether companion use worsened anyone’s depression or whether people already sliding downward reached for the software.

Direction of causation is the open question throughout this literature. Depression predicts companion use, and companion use predicts distressed language, and existing study designs cannot separate the two.

Dose is also unmapped. Nobody knows whether twenty minutes a day and four hours a day differ in kind or only in degree, and the answer probably varies by person in ways current research cannot yet resolve.

Frequently asked questions

Are AI companions approved for treating depression?

No AI companion app is approved by the FDA or any equivalent regulator as a treatment for depression. A small number of purpose-built digital therapeutics have regulatory clearance, but consumer companion apps are general-purpose software and are marketed as wellness or entertainment products, which places them outside medical device oversight entirely.

Is an AI companion better than nothing when no therapist is available?

Possibly, for mild low mood, and that is a real scenario given waiting lists. The evidence supporting that hope comes from structured therapy chatbots rather than companion apps, though, and “better than nothing” stops being true if the app becomes the reason you never join the waiting list.

Can an AI companion tell if I am getting worse?

Not reliably. Companion software has no clinical assessment built in, no view of your sleep or appetite or behaviour, and no ability to contact anyone on your behalf. Most major apps now route explicit crisis language to helpline information, which is a safety feature rather than detection of a worsening depression.

Does using an AI companion for depression mean something is wrong with me?

No. Reaching for an available, private, judgement-free place to put difficult thoughts is an ordinary response to feeling bad, and survey work consistently finds emotional support among the most common reasons people use these apps. Wanting more than software is also ordinary, and worth acting on.