Care in the Machine No. 5: The Believing Gap

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. 5 · Monday, August 31, 2026

Three hundred people put on headphones and listened to two recordings. Same therapist. Same session. Same words, in the same order. The only difference was a label on the screen before each one played: human, or AI generated. People rated the "human" recording warmer, more credible, more likely to actually help, every time (Blake et al., 2026). Nothing about the care changed. Only what they were told about it did.

I have been carrying that finding around all week. Not because it is shocking, we already suspected something like it, but because of what it does to the word "disclosure." We teach it as a box on a form. This says it might be closer to a clinical moment, one that can cost you something the instant you say it out loud, and cost the client something worse if you never do. Let's get into it.

This week's teaching ideas

For the therapist. Your own feelings about AI are data, not a distraction from the work. A survey of 159 professional counselors this year found real excitement sitting right next to real worry, about privacy, about bias, about losing the human thread. It also found something worth sitting with: counselors from racial and ethnic minority groups, and school counselors, reported more optimism about AI than their peers (Fulmer et al., 2026). If a colleague of color on your team sounds more hopeful about what a tool could take off her plate, that is not naivety. Ask her what she is seeing that you are not.

For the supervisor. If you or a supervisee uses any AI tool to track sessions or flag cues, do not let the printout replace the conversation. A review of AI in clinical supervision names the risk plainly: feedback delivered by an algorithm, with no trusted relationship carrying it, tends to land as more punitive and less like growth (Brinck et al., 2026). The fix is not throwing out the tool. It is keeping a human voice between whatever the tool flags and the supervisee who has to hear it, and giving that supervisee a real way to opt out.

For the counselor educator. Thirty-one counseling professionals, educators, licensed counselors, supervisors, and trainees among them, mapped out where they expect AI to help the field and where they expect it to strain it (Crofford et al., 2026). Six regions came out of that map, from scholarship help to what the researchers called "growing pains." Borrow the exercise before your program writes a policy from the top down. Ask your own students and faculty to sort their hopes and worries the same way. You will learn where the real anxiety actually lives, and it is rarely where the policy memo assumes.

For the counselor in training. A small study had fifty counseling students rehearse hard client conversations with a ChatGPT partner before trying them with people. Self-efficacy for working with challenging clients jumped, especially around building the relationship, and the students named real friction too, moments the bot felt off, ways it flattened what should have stayed complicated (Han et al., 2026). Use it the way the study did, as a rehearsal room before the real one, never a replacement for the person across from you. Notice what the tool gets wrong. That noticing is training too.

This week's free resource

Naming the Machine: A Disclosure Script for AI-Assisted Care. One page, printable. Three moments to say out loud that a tool touched your paperwork, a script for each, and a short checklist for what your written consent language should cover. It grew straight out of the field items below, the finding that belief changes care even when the care itself does not change at all.

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

This week I had a stack of new studies to get through for a talk I'm building, and before any of them earned a spot in my slides, I ran each one through scite. I do this with all of it now, my teaching, my writing, the research I hand back to a doctoral student. Search the claim, pull up the actual record, and read who has cited the paper and how, whether they were backing it up, pushing against it, or just mentioning it in passing.

Here is where I kept my hands on the wheel. scite can tell me a study has been cited two hundred times. It cannot tell me whether those two hundred were agreeing with it or quietly taking it apart; that part I still read myself. And it flags retractions, which matters more than it sounds. A retracted study resolves online just as cleanly as a sound one, so "the link works" was never proof of anything. The tool sorts the pile. I decide what I am willing to stand behind in front of a room of clinicians.

Try this: before you cite the next study, in a case note, a lecture, a policy you are writing, paste its DOI into scite or doi.org and look for an editorial notice. Thirty seconds. It is the cheapest protection your credibility will ever buy.

This week in the field

Taking race out of a model does not take the racism out of the data. A governance piece on psychiatric and clinical algorithms makes a sharp claim: these tools perform worse for Black patients largely because they fail to account for the cumulative weight of racism-related stress, the uneven variance a lifetime of differential exposure leaves behind (Fields et al., 2025). The authors lay out four principles for anti-racist AI, starting with putting Black researchers and community members in the room where the tool gets built. So what for us: before anyone adopts an AI-assisted assessment or risk tool, this is the question to ask up front: whose lived experience did this model actually learn from?

AI text reads as more empathic, until you find out it is AI. A synthesis of recent studies lays out the paradox plainly: people often rate AI-written language as more empathic than human-written language, right up until they learn it came from AI, at which point the same words drop in their eyes (Ong et al., 2026). And even when AI empathy tests higher on paper, people still choose to wait longer for a human's. So what for us: a warmly worded AI summary of a session is not standing in for the relationship it describes, and your clients already sense the difference, even when they cannot name it.

Thirteen organizations agree on more than you would guess, and are silent on more than they should be. A content analysis of AI ethics guidance from thirteen national and international mental health organizations found real consensus on client welfare, competence, and justice, alongside real gaps the guidance leaves conceptually thin (Xiong et al., 2026). So what for us: this doubles as a checklist for anyone writing or updating a practice AI policy this fall. Start where the field already agrees. Build slower, and more carefully, where it does not.

An idea I'm floating

Every ethics code I teach says tell the truth about the tools in the room. I still believe that. But sit with the finding above for a minute: the same care, rated worse, for no reason but a label. If naming the machine costs trust, and staying quiet is its own kind of harm, informed consent is not the clean fix it gets treated as in a syllabus. I do not think the answer is softer wording or a gentler script. I think the answer might be that disclosure has to come with something alongside it, a person willing to sit with whatever the client feels once they know, rather than a form that assumes telling them is the whole job. I have not landed this one. So let me put it to you instead: when you name the machine, who is in the chair when the client hears it? Write back. I read every one.

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.

More of my work at PESI. My trainings there run from AI and the ethics of digital afterlives, with Megan Devine, to advanced grief counseling. Browse my PESI catalogue

References

Blake, J. R., Berman, J. S., & Solomon, E. M. (2026). Perceptions of therapists identified as AI generated or human. Counselling and Psychotherapy Research, 26(3), Article e70211. https://doi.org/10.1002/capr.70211

Brinck, E. A., Soldner, J. L., Kuo, H. J., Sabella, S. A., Landon, T. J., Bernacchio, C. P., & Boland, E. A. (2026). An AI perspective on counseling supervision. Behavioral Sciences, 16(6), Article 1038. https://doi.org/10.3390/bs16061038

Crofford, H., Bor, E., & Kemer, G. (2026). Counseling professionals' perspectives on AI integration in education and supervision: A concept mapping study. Counselor Education and Supervision, 65(1), 22-32. https://doi.org/10.1002/ceas.70016

Fields, C. T., Black, C., Thind, J. K., Jegede, O., Aksen, D., Rosenblatt, M., Assari, S., Bellamy, C., Anderson, E., Holmes, A., & Scheinost, D. (2025). Governance for anti-racist AI in healthcare: Integrating racism-related stress in psychiatric algorithms for Black Americans. Frontiers in Digital Health, 7, Article 1492736. https://doi.org/10.3389/fdgth.2025.1492736

Fulmer, R., Zhai, Y., & Beeson, E. T. (2026). Counsellors' attitudes, emotions, and ethical concerns regarding artificial intelligence: Results from a professional survey. Counselling and Psychotherapy Research, 26(1), Article e70086. https://doi.org/10.1002/capr.70086

Han, E., Hsu, P.-L., & Shin, J. (2026). Enhancing counselor education with artificial intelligence chatbots: Attitudes toward artificial intelligence and counseling skill self-efficacy. Journal of Counseling & Development. Advance online publication. https://doi.org/10.1002/jcad.70058

Ong, D. C., Goldenberg, A., Inzlicht, M., & Perry, A. (2026). AI-generated empathy: Opportunities, limits, and future directions. Current Directions in Psychological Science. Advance online publication. https://doi.org/10.1177/09637214261444274

Xiong, Y., Zhai, Y., & Fulmer, R. (2026). Regulating artificial intelligence in mental health: A content analysis of global professional ethical and policy guidelines. Journal of Counseling & Development, 104(3), 339-357. https://doi.org/10.1002/jcad.70044

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