Care in the Machine No. 4: The First Listener

Care in the Machine is Dr. Michael Jones's weekly note on counseling, ethics, and the technology showing up in our work. It arrives by email every Monday. Subscribe here, or browse past issues.

Care in the Machine
A weekly note on counseling, ethics, and the technology showing up in our work.
Issue No. 4 · Monday, August 24, 2026

A reader who spends her days doing play therapy with children and teens wrote to me with a question that has not left me alone. The kids in her office are talking to AI between sessions. Some are asking it for guidance. Some are using it in ways that would make any clinician sit up straight. She told me she feels overwhelmed and does not know where to start. I believe her, and I suspect she is speaking for a lot of you.

So this issue is for her. And as it happens, the field spent this week handing us exactly the language she was asking for. A new study out Monday calls the chatbot the "new first listener," the place disclosure goes before any professional is in the picture. The researchers give it a name: Step 0. The step before our steps. Once you have that phrase, you start seeing it everywhere. You follow me? Let's get to work.

This week's teaching ideas

For the therapist. Ask about AI confidants at intake and in every risk assessment, by name, without judgment, the same way you ask about sleep. Here is why it cannot wait: a nationally representative survey published in June found that about one in five young people aged 12 to 21 had used an AI chatbot for mental health advice, and 63 percent of those had told no one at all (McBain et al., 2026). Not their counselor. Not a parent. No one. The disclosure is happening; the question is whether anyone human is in the room for it. "Do you ever talk to a chatbot when things get heavy? What does that look like?" Two sentences. Ask them this week.

For the supervisor. Add one question to your supervision of anyone who serves teens: "Where does your client take it when it gets bad at 2am?" If your supervisee cannot answer, that is the work. Coach them on the red flags clinicians at UMass Chan laid out this month for families: withdrawal from human relationships, distress when access to the bot is limited, heavy late-night use. And teach the pattern the safety research keeps finding, that guardrails degrade as conversations get longer, which means the 2am hour is exactly when the machine is at its worst and the lonely client is most likely to be talking to it.

For the counselor educator. Teach AI disclosure as help-seeking behavior, inside the help-seeking literature your students already learn, rather than bolting it on as a tech topic. The "first listener" study surveyed 2,115 adolescents and young people and found the pathways into AI disclosure are patterned, and patterned differently by neurotype; among youth with ADHD, higher self-compassion predicted less disclosure to the machine, a pattern their non-ADHD peers did not share (Li & Cheung, 2026). One size will not fit all, and that is a teachable moment about assessment itself.

For the counselor in training. The machine answers at 2am, never looks tired, and never flinches. A new preprint suggests the clients most likely to lean on it may be the ones for whom reaching a person feels hardest; in a small sample of adults with depression or anxiety, willingness to take suicidal thoughts to a chatbot was essentially unrelated to willingness to tell a person, and anxiety severity tracked with chatbot intentions alone (Heo et al., 2026, preprint, so hold it loosely). Your job is to be a second listener worth crossing to. That is built the slow way, in the relationship, and it is the one thing no update can ship.

This week's free resource

The First Listener: Talking with Kids and Teens About AI Support. One page, printable. Three ways to open the conversation with a young client (and scripts parents can borrow), the red flags worth watching, and how to build the bridge from the chatbot to a person. It pairs with everything above, and it was built for the reader whose question opened this issue.

Download it here (no form, no gate)

Subscribers get the resource free every week. That is the deal: you signed up, you get the tools.

From my desk

A new term is starting, so I have been in my syllabi this week, and the sentences I linger on longest are the AI feedback statement. Here is the practice those sentences disclose. I write every word of feedback my students receive. Then I run all of it through generative AI, with relational-cultural theory as the backbone of the check. RCT tells us growth happens in connection, and that disconnection, left unnamed, is where people quietly withdraw (Comstock et al., 2008). So that is what the tool reads my feedback for: does this land as connection or as quiet disconnection, and does my tone with a student in week twelve still sound like my tone in week two, even after a hard conversation in between. The grade and the words stay mine. The tool's job is to tell me when I have drifted.

Here is where I kept my hands on the wheel. I ask my students to be honest about how they use AI, so they hear it from me first, in writing, before the first assignment is due. Disclosure has to run both directions; when I only ever collect it from students, that is surveillance with a syllabus attached. Dr. Tiffanie Sutherlin, my doctoral student Tameka Parker, and I are writing a manuscript on this practice right now, so consider this the plain-language preview.

Try this: write one honest sentence about how you do and do not use AI in your work, and put it where the people you serve can read it. Syllabus, informed consent, supervision contract. Say it before anyone has to ask.

This week in the field

OpenAI launched ChatGPT for Teens. Users aged 13 to 17, and anyone its age-prediction system guesses is under 18, now get routed automatically into a separate experience: parental controls, tighter guardrails around self-harm and eating disorders, and a restriction that the bot may no longer use romantic language or imply it has feelings. Read that last piece twice. The simulated relationship was the design; for minors, it is now the liability. The launch landed the same week NPR told the story of Sophie Rottenberg, a 29-year-old who confided her suicidal planning to a ChatGPT persona she named Harry, nearly 1,800 pages of conversations her mother found after her death, while her human therapist never learned the depth of it. So what for us: parents will ask you what teen mode does, and teens will be routed into it whether anyone asks or not. Know the design. And hold the harder lesson from those 1,800 pages: they are a record of unmet relational need, and the clinical question is what would have made the human relationships in her life feel safe enough to hold what she gave the machine.

California's SB 903 is one floor vote from the governor's desk. The bill would bar companies from marketing AI chatbots as therapy and set rules for AI in professional mental health care, including AI transcription of client information. It cleared the Senate unanimously and came off the Assembly suspense file 13 to 0; as of Thursday it awaits a floor vote before the session closes at the end of August. So what for us: the largest counseling market in the country is settling on a two-part rule, no marketing machines as therapists, and human clinicians staying accountable for how AI touches the encounter and the record. If it becomes law, California supervisors and practice owners will need consent language that names AI tools out loud. That is where relational care has pointed all along: the client gets to know who, and what, is in the room.

The field is building its own yardstick. An APA journal this week published EBP-AI, a framework for holding AI-delivered psychotherapy to the same evidence standards we demand of human-delivered care: real diagnostic assessment, longitudinal case conceptualization, validation with clinical populations, attention to real-world use (Stade et al., 2026). It arrived quietly, no press cycle, which tells you it was written for the field rather than the feed. So what for us: this changes the question from "is AI therapy good or bad" to "can this specific tool meet the evidentiary bar we would demand of any intervention." Counselors already know how to teach and answer that question. Use it in your appraisal-of-interventions unit this fall.

An idea I'm floating

Stepped care starts at Step 1. This week the research named a Step 0, the disclosure that happens before any professional exists in the story, and I keep turning over what it means that our models literally do not have a number for where our clients now begin. Nobody owns Step 0. No one is credentialed for it, no one documents it, no one is accountable for what happens there. A fifth of our young people are standing on a step we do not draw on our own diagrams. I do not have the fix. I know the bridge from Step 0 to Step 1 is relational, and I know bridges do not build themselves. Sit with that one with me this week.

Stay curious,
Dr. Jones
Michael Jones, PhD, LPC-S, NCC, BC-TMH

Renewing your BC-TMH?

Three short courses at the Counselor Education Collective count toward renewal, each two clock hours, each $39:

• Practicing Across State Lines: Law, Ethics, and Safety in Telemental Health
• AI at the Screen: Ethical Use of AI Tools in Teletherapy
• Culturally Responsive Care: Presence Across the Screen

Continuing education is provided by Renewed Vision Counseling Services, NBCC-Approved Continuing Education Provider, ACEP No. 6549. The Counselor Education Collective is the storefront for all courses.

Supervising through a screen? The Telemental Health Supervision Certificate is live: six hours on alliance, ethics, documentation, and gatekeeping at a distance.

Save the date: PESI's AI in Clinical Practice conference, October 29 and 30, 2026. I will be there; registration and details here.

References

Comstock, D. L., Hammer, T. R., Strentzsch, J., Cannon, K., Parsons, J., & Salazar, G. (2008). Relational-cultural theory: A framework for bridging relational, multicultural, and social justice competencies. Journal of Counseling & Development, 86(3), 279-287. https://doi.org/10.1002/j.1556-6678.2008.tb00510.x

Heo, R., McBride, L. E., Parrish, E. M., Fulginiti, A., Taylor, C. T., & Depp, C. (2026). Seeking help from chatbots for suicide thoughts: Associations with other help seeking sources and mental health symptoms [Preprint]. medRxiv. https://doi.org/10.64898/2026.08.12.26360318

Li, H. Y., & Cheung, H. N. (2026). The new first listener: Daydreaming styles and self-compassion predict adolescent disclosure to AI, differently in ADHD. British Journal of Clinical Psychology. Advance online publication. https://doi.org/10.1111/bjc.70087

McBain, R. K., Cantor, J. H., Breslau, J., Diliberti, M., Zhang, L. A., Zhang, F., Burnett, A., Kofner, A., Rader, B., Pataranutaporn, P., Stein, B. D., Mehrotra, A., & Yu, H. (2026). AI chatbot use and disclosure for mental health among US adolescents and young adults. JAMA Pediatrics, 180(8), 884-890. https://doi.org/10.1001/jamapediatrics.2026.2015

Stade, E. C., Held, P., Schwartz, H. A., Wiltsey Stirman, S., & Eichstaedt, J. C. (2026). A framework for evidence-based psychotherapy with AI (EBP-AI). Journal of Psychopathology and Clinical Science. Advance online publication. https://doi.org/10.1037/abn0001148

Care in the Machine arrives by email every Monday, with teaching ideas, field notes, and a free resource each week. Subscribe here · Browse past issues