How to Know When You Need a Therapist vs When AI Is Enough

The number of people seeking support for their mental health has grown significantly over the past decade, while the supply of therapists has not kept pace. Wait times for therapy appointments extend for months in many areas. Costs remain high. The gap between people who could benefit from professional support and people who can actually access it is not closing quickly. Into that gap, AI tools have moved — apps, chatbots, and conversational AI that offer responsiveness, availability, and some of the processing functions that make conversation useful for mental health. This has created a genuine question that deserves a serious answer: when is AI support appropriate, and when does the situation require a licensed therapist?

What AI Does Well

AI is available around the clock, costs nothing or very little, and carries none of the logistical friction of scheduling, travel, and insurance navigation. It does not judge, does not get tired, and does not need you to manage its emotional response to what you share. For certain kinds of support, these features are not just convenient — they are clinically relevant. For processing everyday stress, getting a sounding board for a difficult decision, naming emotions that feel murky, practicing how to have a hard conversation, or working through the minor disappointments and frustrations of ordinary life, AI can be genuinely useful. It can ask questions that promote reflection, offer reframes you might not have considered, and provide a form of engagement that helps many people feel less alone with whatever they are carrying.

What AI Cannot Do

Therapy involves a trained professional applying clinical knowledge to a therapeutic relationship that develops over time. The relationship itself — the experience of being consistently seen, the ruptures and repairs that mirror relational patterns, the attunement that a skilled therapist develops — is not incidental to outcomes. Research from the American Psychological Association consistently identifies the therapeutic alliance, the quality of the relationship between therapist and client, as one of the strongest predictors of therapeutic outcomes, independent of the specific modality used. AI does not form a therapeutic relationship. It does not have clinical training. It cannot diagnose, prescribe, or apply evidence-based protocols with the kind of individualization a therapist brings. It cannot read the subtle shifts in your presentation across sessions that indicate whether you are actually improving or simply habituating to distress. And critically, it cannot respond appropriately to a crisis situation — an escalating suicidal ideation, a break from reality, an acute trauma response — in the way a trained clinician can.

The Clinical Threshold Questions

The clearest indicators that professional support is warranted rather than optional include: symptoms that have persisted for more than two weeks and are affecting daily functioning, substance use that is increasing as a coping mechanism, thoughts of harming yourself or others, significant changes in sleep, appetite, or concentration, trauma that has not been processed and continues to intrude into daily life, and any symptom pattern that is escalating rather than stable. These are not exhaustive, but they share a common thread: they describe conditions where the person's own resources and low-intensity support are insufficient to manage what is happening. Continuing to rely on AI in these situations is not harmful in most cases, but it risks becoming a reason to delay care that the situation genuinely requires.

A Detour on the Peer Support Literature

Long before AI became part of this conversation, the field of mental health developed robust evidence for peer support — trained individuals with lived experience of mental health challenges providing support to others. Studies from Dartmouth-Hitchcock Medical Center and other institutions have found that peer support produces meaningful improvements in social functioning, engagement in care, and hope among people with serious mental illness, effects that sometimes exceed what professional-only approaches achieve. This literature is relevant because it establishes that trained non-clinicians providing support that is relational, consistent, and grounded in shared experience can have real clinical value. It does not validate AI as equivalent to peer support — AI has none of the lived experience component — but it does suggest that the clinical question is not a simple binary between licensed therapist and nothing.

The Right Frame

Rather than asking whether AI or therapy is better, the more useful question is what you are actually dealing with and what level of support it calls for. AI is probably sufficient for navigating the normal difficulty of being human — stress, uncertainty, relational friction, minor mood fluctuation, the ordinary work of self-reflection. A therapist is probably necessary when the clinical threshold described above is crossed, or when you have tried lower-intensity support and are not making progress. The two are not mutually exclusive. Many people in therapy find AI useful between sessions for processing, practicing skills, or simply having somewhere to put what is happening. Used this way, AI is a supplement that extends access between appointments — not a replacement for the clinical relationship that makes therapy work.