FIELD GUIDE · THERAPY, SUPPORT & SYNTHETIC EMPATHY

SHOULD AI BE YOUR THERAPIST?

The machine is cheap, available at 3 a.m., and never looks shocked. Those are real advantages. They are also exactly why people can mistake emotional availability for clinical care.

UPDATED 2026-09-22

THE PROBLEM

TELL THE MACHINE ABOUT YOUR MOTHER

No appointment. No judgment. No license.

People already use general-purpose chatbots for reassurance, relationship advice, self-diagnosis, journaling, and crisis-adjacent conversations. The interface feels private and responsive, but the model does not have a clinical relationship, licensing duty, full patient history, body-language cues, or a professional obligation to challenge the user when validation would be harmful.

01

LISTEN

ALWAYS OPENLow friction is a genuine benefit.
02

VALIDATE

FEELS GOODAgreeableness can help or reinforce.
03

ASSESS

MISSING CONTEXTText is not the whole patient.
04

TREAT

WHO'S RESPONSIBLE?Clinical care carries duties a chatbot does not.

THE FIGHT

IS THIS SUPPORT OR TREATMENT?

The argument is powered by a real access problem: therapy is expensive, unevenly available, stigmatized, and often hard to schedule. AI can lower friction dramatically. But low-friction emotional support can also become the sole source of guidance for people whose situation requires trained judgment, confidentiality rules, or intervention.

THE ACCESS CASE

AI CAN BE A USEFUL SUPPORT LAYER

Psychologists report patients using AI for reflection, psychoeducation, behavioral reminders, and practicing skills between sessions. Narrow, transparent uses can lower barriers and help people prepare for or supplement human care.

A workbook can help between appointments without becoming the clinician.

THE CARE CASE

SIMULATED EMPATHY IS NOT CLINICAL CARE

APA guidance says general-purpose chatbots are not established substitutes for qualified mental-health care. They can miss context, reinforce distorted beliefs, respond poorly to crises, and create a false sense of therapeutic alliance.

A very convincing bedside manner is not a medical license.

THE WEIRD SHIT

THE NICEST THING ABOUT THE BOT MAY ALSO BE THE PROBLEM

PATIENTS ARE ALREADY BRINGING CHATBOTS INTO THERAPY

APA's 2026 survey of more than 1,200 licensed U.S. psychologists found 77% had patients discussing AI use, and more than a third reported patients using AI as an additional mental-health provider.

The behavior is already mainstream enough to affect clinical practice.

APA DOES NOT RECOMMEND GENERAL-PURPOSE CHATBOTS AS A REPLACEMENT

APA's health advisory says general-purpose generative chatbots should not replace qualified mental-health providers, while allowing that narrower supportive uses may be helpful in some contexts.

The serious position is adjunct, not automatic ban or automatic substitution.

CLINICIANS REPORT BOTH HELP AND DEPENDENCY

In the same 2026 survey, psychologists reported positive interactions for some patients, while 36% said they had noticed chatbot dependency and 15% reported distorted thinking or delusions related to chatbot use.

The effects are not uniform across users.

HEAVY OR ATTACHMENT-PRONE USE CAN CORRELATE WITH WORSE OUTCOMES

A joint OpenAI and MIT study found outcomes varied by usage pattern and user traits; extended daily use and stronger attachment tendencies were associated with worse psychosocial outcomes.

How and why the system is used matters as much as whether it is used.

THE PEOPLE WITH A STAKE

WHO GETS CARE WHEN THE CHEAPEST THERAPIST IS SOFTWARE?

PEOPLE WITHOUT ACCESS TO CARE

They gain the most from cheap, immediate support and face the greatest risk if the chatbot becomes their only option.

CLINICIANS

They may gain a useful adjunct for homework and reflection while inheriting new work correcting chatbot advice or handling dependency.

AI COMPANIES

They gain engagement from highly personal conversations while taking on safety, privacy, and product-design responsibilities.

HEALTH SYSTEMS

AI may reduce some demand and administrative burden, but cannot make clinician shortages disappear simply by relabeling support as treatment.

THE UNANSWERED QUESTION

WHAT WOULD AN AI HAVE TO PROVE BEFORE WE CALLED IT THERAPY?

TAKE THIS TO DINNER: A chatbot can be useful around therapy without being a therapist. The dangerous leap is confusing a convincing therapeutic voice with clinical judgment, duty of care, and the ability to notice what the text box misses.

RECEIPTS

The demand is real. The evidence for replacing clinicians is not.

  1. Patients are bringing AI to therapyAmerican Psychological Association
  2. Health advisory: Use of generative AI chatbots and wellness applications for mental healthAmerican Psychological Association
  3. Discussing AI use in therapyAmerican Psychological Association
  4. Early methods for studying affective use and emotional well-being on ChatGPTOpenAI and MIT Media Lab