A Companion Guide to “Can AI Be Your Therapist?"
This article is a companion guide for Episode 20 of The Work Within, “Can AI Be Your Therapist?”
The volume and frequency of AI conversations are exhausting. At the same time the technology is genuinely changing how people take care of themselves. Those two things are true at once, and pretending otherwise has not helped anyone.
In a study published in JMIR Mental Health in March 2026, researchers surveyed 1,805 U.S. adults aged 18 to 49. More than a third of them, 35.2%, said they use an AI tool for mental health support at least once in a typical week. Among the subset who had ever seen a human mental health professional, 28.4% reported seeing that professional less often since they started using AI for the same purpose.
Two other findings in that study deserve more attention than they have received. People who screened positive for moderate to severe depressive or anxiety symptoms were significantly more likely to use AI for mental health support than people without those symptoms. And people reporting suicidal ideation were more than twice as likely to be classified as heavy users, meaning they regularly spend hours discussing their mental health through AI.
The takeaway? The people leaning hardest on these tools are, disproportionately, the people in the most distress.
There is a counterweight in the same data. Nearly 60% of respondents said they would still turn first to family or friends when facing a mental health concern, and participants overall held more favorable views of human professionals than of AI tools. The exception was heavy users, who rated the two about equally.
If you assume the therapists reading these numbers are alarmed, you might be surprised. Laura Harrison, LPC-S and Ly Tran, LPC-S are both licensed professional counselors, both early adopters, and both use AI daily in their own lives. When Ly informally surveyed his own caseload, somewhere between a third and half of his clients said they were using AI as part of their mental health journey.
So the question is no longer whether people are doing this. They are. The useful question is what AI actually does well, knowing its limits, and how to use it in a way that helps you rather than hollowing you out.
Where AI genuinely helps
It is available at two in the morning.
Panic does not schedule itself for business hours. Ly described clients who are awake at 2 a.m., struggling to regulate, in the middle of an anxiety spike. No therapist is available then. Something that responds, helps slow the spiral, and walks through a grounding exercise is better than nothing. And in many cases, he’s seen it add real value to his clients.
It extends the work between sessions.
This one surprised us. One of the hardest parts of therapy is getting people to do anything with it during the other 167 hours of the week. Insight fades. Homework goes undone. Ly has watched clients take something raw from a session, keep working on it with an AI assistant during the week, and come back with substantial development on it. From a clinician’s chair, that is a meaningful gain.
It lowers the cost of getting oriented.
Laura described using AI to map out the certification path her son should take for his nursing assistant program. Ly uses it to help plan his kids’ homeschool curriculum, where there are a lot of unknowns and variables. Call it a faster, more personal search engine. Applied to mental health, that means understanding a diagnosis, learning what a treatment approach involves, or figuring out what questions to bring to a first appointment.
Where AI has limits
It may not recognize a crisis.
This is the one that matters most, and the survey data makes it urgent. If people in the most distress are the ones using these tools most heavily, then the tool’s ability to recognize distress is not an edge case. It is the central safety question.
Can an AI tool tell the difference between a hard night and an emergency? Laura has spent a year and a half deliberately testing models with emotionally loaded scenarios. They have improved, and they are now more likely to say “you should talk to a professional” or “I am not a therapist.” But that improvement is recent, and it followed real harm. Early on there were stories of chatbots encouraging deeply vulnerable people toward awful things.
It only knows what you type.
AI requires self-disclosure. You have to already be aware of something, and then be willing to write it down. In a room, a therapist watches you hesitate before answering a question, and reads the shame or fear in that pause. Then they make a clinical decision: give this person space, or gently press in. A chatbot has no idea why you stopped typing. You could be crying. You could be getting a glass of water.
It cannot see your body.
Laura would rather do therapy in person than over Zoom, and her reason is instructive. In the room you can tell when someone’s words and their body disagree. You catch the small sigh, the shifted posture, the energy someone carries in with them. Those tells open up doors for much deeper work. As she put it: maybe someday it reads body language. It will never have a heartbeat.
It gives you the average, not the discernment.
An AI response is drawn from a distribution of plausible replies. A therapist also has a range of possible responses. What Ly described is what happens next: a prompting he believes is from the Holy Spirit that tells him which true thing this particular person needs to hear right now. That is not a technical capability. It comes from having lived inside the same human condition and being attentive to something beyond the data.
The foundational premise
Ly said something that reframes the entire conversation:
People’s wounds often occur within the context of human connection. That means healing has to happen in the context of human connection.
If that is true, then the limits of AI in mental health are not a temporary engineering problem. They are structural. A tool can help you prepare for connection, process after connection, and hold you steady between moments of connection. It cannot be the connection itself.
Four guardrails for using AI wisely
1. Tell your therapist you are using it.
Ly asks clients which model they are using, and often asks them to have it write a short summary of what they have been working through. Most clients are glad to. They would rather their therapist be informed. If you are in therapy and using AI on the side, bring it into the room. It stops being a secret coping mechanism and becomes material for deeper processing.
2. Let it be directed, not freelance.
The most productive version of this is a therapist saying, “go work on this specific thing with your AI this week, and bring back what comes up.” That is different from unsupervised late-night processing. Ly’s read on where things are heading: some clients may come in every other week instead of weekly, using AI more in between, but under clinical direction.
3. Do not extend it a hundred percent trust.
It makes mistakes. It misunderstands you. Laura’s framing is the right posture: treat what it offers as “that is an interesting thought, let me see how that fits.” Not as “this is the truth God gave me through AI.” Some models also skew sycophantic, agreeing with you and reflecting your framing back rather than challenging it. That feels wonderful and is often the opposite of what growth requires.
4. Watch for the loop.
Ly, who is candid about being ADHD, called always-on AI an ADHD person’s dream come true. Any thought, any hour, instantly engaged. The same quality that makes it useful makes it sticky. If you notice yourself cycling, going back again and again for reassurance at two in the morning, that is a pattern worth naming. It looks a lot like how social media worked on us, complete with the dopamine hit.
The anxiety underneath the AI anxiety
Ask people what worries them about AI and you will hear about jobs. Listen longer and you hear something else.
“Will my job be replaced” is rarely only about income. It is about purpose and identity, about what a person is for. Ly named it directly: if parts of what we do can be stripped away and automated, the question of what gives our lives value gets very loud. That is a spiritual question wearing financial dressing, and it deserves a spiritual answer.
The other layer is generational. Laura hears it constantly, including from people asking whether it is even responsible to have children now. Her response is bracing and, we think, correct: it has never been safe to have kids. Look back through history and find the era when the future was assured. There isn’t one. The reason to invest in family and relationships and community was never a guarantee of safety and happiness. It was the value of the relationship itself, and the values you want carried forward into the world.
A faith-grounded posture
Two ideas from the conversation are worth carrying with you.
Nothing new under the sun. Ecclesiastes 1:9 has outlasted every technology that was supposed to end the world. Y2K. The atomic bomb. The printing press, for that matter. This does not mean AI is trivial. Ly is clear that the speed is genuinely new and that Christians who put their heads in the sand will pay for it later. It means the fear is not new, and the antidote is not new either.
Trouble is promised, and so is the ending. John 16:33 does not promise a life free of difficulty. It promises trouble, followed by “take heart.” Ly went further, naming a version of Western Christianity that has absorbed prosperity-gospel assumptions about safety, immunity, and comfort. We are not promised freedom from sorrow. We are promised that none of it is out of God’s hands, and that in the end he makes all things new.
When we are afraid, we tend to do one of two things: over-control, or retreat. Neither is the posture of faith. The posture of faith is to stay present and engaged.
C.S. Lewis, writing to people who expected to be vaporized at any moment by the atomic bomb, put it better than anyone since. Let the bomb, when it comes, find us “doing sensible and human things.” Praying, working, teaching, reading, bathing the children, playing tennis, talking with our friends. Not huddled together like frightened sheep, thinking about bombs.
That is a workable instruction for 2026.
Frequently asked questions
Can AI replace a therapist?
No. It can supplement care between sessions, but it cannot read your body, hold the nuance of your story, reliably recognize a crisis, or provide the human connection that healing requires.
Is it bad to use ChatGPT or Claude for emotional support?
Not inherently. Many therapists see real benefits, especially for after-hours support and processing between sessions. The risks come from using it without a clinician involved, extending it too much trust, or letting it become compulsive.
Should I tell my therapist I use AI?
Yes. Most therapists would rather know. It gives them useful information and lets them direct how you use it.
What should I do if I am in crisis?
Contact a human. In the U.S., call or text 988 to reach the Suicide and Crisis Lifeline, or go to your nearest emergency room. Do not rely on an AI tool to assess the severity of what you are experiencing.
Is using AI for mental health compatible with Christian faith?
Most Christian counselors would say yes, with discernment. The concerns are about what it displaces: human community, spiritual direction, and the discernment that comes from prayer and relationship rather than from a model’s output.
Where to go from here
If this conversation surfaced something for you, that is worth paying attention to.
If you are looking for a clinically trained Christian therapist, we would be glad to help you find one. And if you lead a church and are thinking about how to support your people’s mental health with real clinical care, that is exactly what the Watershed Initiative was built for.
Listen to the full conversation on Episode 20 of The Work Within, wherever you get your podcasts.
Source
Ueda, M., Birnbaum, M. L., Liu, Y., Yu, Q., Tian, X., Mirer, A., Ramanathan, S., & Sinyor, M. (2026). Help-seeking in the age of AI: Cross-sectional survey of the use and perceptions of AI-based mental health support among US adults. JMIR Mental Health, 13, e88196. https://doi.org/10.2196/88196
Ready to Deepen Your own Inner Work?
The team at Watershed Initiative would be honored to come alongside you. We provide clinically excellent, faith-integrated care for individuals, couples, and leaders seeking Christian counseling.