AI Is Already in Your Practice. Your Patients Know It. Do You?
- Jul 13
- 12 min read
Updated: Jul 27
By Dr. Ernest Wayde, PhD, MIS
Why AI literacy is becoming a clinical competency for psychologists, therapists, and healthcare professionals
Imagine a therapist has just ended a session and her patient mentions they talked through their anxiety with ChatGPT three times this week before their appointment. The therapist nods and smiles. But internally, alarm bells are going off. She has no idea what the chatbot told her patient, whether it affirmed unhealthy patterns, whether it responded appropriately when the patient mentioned feeling hopeless, or whether it made the patient easier or harder to reach in that session. She has no idea how the chatbot has impacted her patient and their session. She does not know because she never asked. And she never asked because it did not occur to her that she needed to.
Nowadays, this is not an unusual scenario. It is a Tuesday.
Whether we like it or not, AI has entered the clinical relationship. In many practices, it entered quietly, through the habits of patients who found it accessible, available at 2 a.m., and non-judgmental in ways that felt easier than calling a crisis line or waiting three weeks for an appointment. Clinicians did not invite it in, and many are not even aware of how frequently it is showing up in their sessions. Most have not yet developed the knowledge to evaluate how it is impacting their clients and clinical sessions.
This does not mean clinicians need to become AI experts. It means they need to know enough to ask better questions, recognize when AI is influencing care, and respond when patients bring AI into the therapeutic conversation. That is what AI literacy looks like in practice. And developing it is becoming part of what it means to practice responsibly.
In June, we examined what happens when AI crosses the line from administrative tool into clinical territory, shaping patient behavior, influencing therapeutic decisions, and creating accountability gaps that organizations are not prepared for. The next question is unavoidable: are clinicians prepared to recognize when that line has been crossed? For most, the answer is not yet. And that gap is no longer just a professional development concern. It is a gap with real consequences for patients and clinicians.
Only 13.8% of clinicians feel their training adequately prepared them for AI integration (Chatzichristos et al., 2025, JMIR). At the same time, FDA’s list included 1,247 AI-enabled devices in the United States as of August 2025 (U.S. Food and Drug Administration). The tools are proliferating. The training is not keeping pace.
That mismatch has consequences, not just for how organizations adopt technology, but for what happens in individual clinical encounters every day.
For mental health professionals, the situation is more immediate. Research published in Frontiers in Psychology in early 2026 found that most psychologists have limited understanding of how AI tools work, with their knowledge coming primarily from mainstream media and commercial advertising rather than systematic professional training. Many report using AI tools in their practice without fully understanding the risks those tools carry (van Zyl, 2026).
A 2026 survey of healthcare leaders across 90 countries conducted by the Digital Medicine Society and Google for Health found a consistent pattern: AI adoption stalls and fails when people lack the skills, confidence, and clarity to use AI in their actual workflows. Acquiring AI tools is not the same as being ready to use them responsibly. Clinicians cannot assume that gap will close on its own.
Your Patients Are Already There
The most urgent argument for AI literacy among mental health professionals is not what AI is doing inside clinical systems. It is what AI is doing in the space between sessions.
According to the American Psychological Association's 2026 Chatbots and Mental Health Survey, 35% of psychologists report having patients who turn to AI to act as an additional mental health professional (American Psychological Association). Thirty-six percent (36%) have noticed patients developing some level of dependency on a chatbot (Becker’s Behavioral Health). Patients are using these tools to manage anxiety and depression symptoms, process relationship conflict, explore ways to improve their wellbeing, and in some cases, work through what they cannot yet bring into the therapy room.
The appeal is understandable. General-purpose AI tools are free or low-cost, available at any hour, and tend to be affirming in ways that can feel validating to someone in distress. What they often are not is clinically supervised, clinically accountable, or designed to meet the safety expectations of professional care.
Research presented at the ACM Conference on Fairness, Accountability, and Transparency in 2025 found that AI chatbots models responded inappropriately twenty or more percent of the time (Moore et al., 2025, FAccT Conference). In an analysis of six popular chatbots, none met clinical standards for crisis response (American Psychological Association). The APA's own survey found that 94% of psychologists said chatbots cannot treat mental health conditions with the appropriate level of nuance.
These tools also express bias. Research has documented cases in which AI chatbots expressed stigma toward mental health conditions, agreeing, for example, that people might reasonably avoid socializing with someone who has a mental illness. The model did not intend to be harmful. It did not intend anything. It produced statistically likely output based on training data that encoded those attitudes.
Clinicians are also beginning to report cases in which extended chatbot interaction may reinforce delusional beliefs, reality distortion, or psychological dependency in vulnerable patients. These are not hypothetical risks. They are showing up in practice.
The questions this raises are not abstract. What did the chatbot tell your patient this week? Did it affirm avoidance when the therapeutic goal is engagement? Did it reinforce hopelessness when the patient mentioned feeling stuck? Did the patient disclose something to the chatbot that they have not yet brought into the room and did the chatbot respond in a way that made disclosure more or less likely going forward? Did AI change your patient's expectations of what therapy should feel like, or what progress should look like?
A clinician who does not know how these tools work cannot answer those questions. And a clinician who cannot answer those questions is operating with incomplete information about a significant influence on their patient's experience between sessions.
The Psychology Problem Hiding Inside the Technology Problem
Most conversations about AI risk in healthcare focus on model accuracy, data privacy, regulatory compliance, and governance structures. These are real concerns, and they matter. However, for clinicians working with patients, the risks are more immediate and personal than that. They show up in the consulting room, in the session, in the moment a clinician reads a note, reviews a recommendation, or sits across from a patient who has spent the past week taking advice from a chatbot.
One of the most significant risks sits at the intersection of AI and human judgement. The problem is not only that AI may be wrong. The problem is that AI may be wrong in ways that make humans less likely to notice. This is automation bias, the well-documented tendency to over-rely on automated output, to defer to AI recommendations even when something about those recommendations does not look right, and to reduce active verification when a system presents its output with confidence. It is not a character flaw. It is a predictable feature of human cognition under real-world conditions.
The evidence from clinical settings is concerning. A review examining AI-assisted diagnostic imaging found that in musculoskeletal imaging, 45.5% of the total mistakes made by clinicians in AI-assisted rounds were due to following incorrect AI recommendations (Wang et al., 2023). That pattern held across all levels of expertise. Experience did not protect clinicians from deferring to a confident-sounding wrong answer. In mammography, when AI provided incorrect suggestions, radiologist accuracy dropped markedly regardless of years of practice. In medication management research involving UK general practitioners, clinicians switched from a correct to an incorrect prescription after receiving erroneous AI advice in 5.2% of all cases examined (Goddard et al., 2014).
What makes this particularly relevant for mental health professionals is that the conditions that amplify automation bias are the conditions of everyday practice. Time pressure. Cognitive load. High patient volume. Fatigue. When a clinician is at the end of a full day of sessions and an AI documentation tool generates a note that sounds plausible, the probability that it will be read closely drops. When a decision support tool offers a recommendation with apparent confidence, the probability that it will be questioned diminishes.
This is not a technology problem. It is a human factors problem. And it sits directly at the intersection of what AI does and what psychology has studied for decades, how humans make decisions under uncertainty, how authority and confidence shape judgment, how cognitive shortcuts operate under load, and how context determines what people notice and what they miss.
What AI Literacy Actually Means
AI literacy tends to generate a particular kind of anxiety in clinical audiences. It may sound like coding or statistics. It often sounds like something that requires going back to school. However, it does not necessarily mean any of those things.
AI literacy is the ability to critically evaluate, effectively interact with, and meaningfully use AI technologies (Markus et al., 2025). For clinicians, that translates into a practical set of competencies that build directly on skills they already have.
It means being able to recognize when AI is involved in a workflow, which is not always obvious, since AI is increasingly embedded in systems without prominent disclosure.
It means knowing enough about how a tool works to ask the right questions. Not how to build the model. How to evaluate it. What was it trained on? How does it perform on patients like mine? What does it do when it encounters something outside its training data? Who is accountable when it produces something wrong?
It means understanding the basic risk profile of different types of AI. A documentation assistant that drafts notes for clinician review carries a different risk profile than a triage tool that influences which patients are flagged for follow-up. Treating them the same is not sound oversight, it is a failure to evaluate.
It means being able to guide patients who are using AI outside of formal care. This is perhaps the most immediately relevant competency for mental health professionals. A clinician does not need to be an AI expert to ask a patient what tools they are using, explore how those tools are shaping their thinking, or bring AI-generated content into the therapeutic conversation the same way they would bring in a journal entry or a message from a family member.
And it means maintaining the kind of critical stance toward AI output that good professional judgment has always required, treating what AI produces as information to be evaluated, not conclusions to be accepted.
Where to Start Right Now
Building a full picture of AI literacy takes time. But there are steps worth taking now, regardless of where you work or what tools you currently use.
Start by asking your patients. Not just as a formal intake question, as a professional habit. Are they using AI for health information, emotional support, or anything that feels like a therapeutic conversation? What are they using it for? What has it told them? This single practice closes an information gap that most clinicians do not yet know exists.
Do not enter protected health information into public AI systems. This applies to ChatGPT, Claude, Gemini, and any general-purpose AI tool not covered by a Business Associate Agreement with your organization. The convenience is real. So is the liability.
Treat AI-generated output as unverified until you have checked it. This applies to documentation drafts, decision support suggestions, risk flags, and any other AI output that enters your workflow. The note sounds right. The recommendation seems reasonable. That is exactly the condition under which automation bias operates. Read it. Verify it. Approve it because you believe it is accurate, not because it was generated.
Know what is running in your organization. If AI tools are being used in scheduling, patient messaging, documentation, triage, or decision support, you have a professional interest in understanding what those tools are doing. You do not need to become the system administrator. You do need to know enough to ask whether the tool has been evaluated, who is monitoring it, and what the protocol is when it produces something wrong.
AI Literacy Is Professional Responsibility
Professional practice has always required clinicians to understand the tools they rely on, the instruments they use, the assessments they administer, the technologies they bring into care. That standard did not emerge because clinicians were expected to be engineers or pharmacologists. It emerged because the tools clinicians use affect patients, and clinicians are accountable for that effect. AI extends that responsibility into new territory.
Patients are bringing AI into the therapeutic relationship whether or not clinicians are prepared for it. AI tools are being deployed into healthcare workflows whether or not the people using them understand how they work. The question is not whether AI will be part of practice. It already is. The question is whether clinicians will engage with it on their own terms, with the informed judgment their training equips them to apply, or whether they will encounter it reactively, after something has gone wrong.
AI literacy is not about becoming more technical (Markus et al., 2025). It is about remaining professionally competent in a world where AI is already shaping what patients believe, what clinicians see, and how care decisions are made.
In the meantime, the most important thing any clinician can do is start. Ask the patient on your schedule tomorrow whether they have been using AI. Read the documentation draft your system generated before you sign it. Ask your administrator what AI tools are running in your practice and whether a governance structure exists.
That is where AI literacy begins. Not in a training module. In the next clinical encounter.
Staying current on AI in healthcare does not have to mean hours of reading. The Wayde AI Brief is a short weekly intelligence brief for healthcare and behavioral health leaders navigating real-world AI adoption, governance, and risk. Subscribe for free.
If your organization is working through how to build AI literacy across clinical and administrative roles, that is exactly the work we do at Wayde AI.
Frequently Asked Questions
What is AI literacy and why does it matter for clinicians?
AI literacy is the ability to critically evaluate, effectively interact with, and meaningfully use AI technologies. For clinicians it is not about coding or data science. It is about knowing enough to recognize when AI is influencing care, ask the right questions about tools being used in your organization, and guide patients who are using AI outside of formal treatment.
My patients haven't mentioned using AI. Does this still apply to me?
Probably yes. More than a third of psychologists report patients using AI as an additional mental health resource, according to the APA's 2026 survey. Many patients do not volunteer this unprompted. The most important first step is simply asking.
I don't use AI in my practice. Why does this affect me?
Your patients likely do. They may be processing anxiety, grief, or suicidal ideation with a general-purpose chatbot between sessions, one with no clinical oversight and no accountability. What happens in those interactions can directly affect what happens in your sessions.
What is automation bias and how does it affect clinicians?
Automation bias is the tendency to over-rely on automated output and reduce active verification when a system sounds confident. Research has found it affects clinicians at all experience levels. The conditions that amplify it, time pressure, fatigue, high patient volume are the conditions of everyday clinical practice.
Are AI chatbots dangerous for patients with mental health conditions?
Current evidence warrants significant caution. Popular chatbots respond inappropriately to mental health symptoms at least 20% of the time, and none of six commonly analyzed chatbots met clinical standards for crisis response. These tools carry no clinical accountability and are not regulated as care.
What should I actually do right now?
Four steps worth taking immediately: ask patients routinely whether they are using AI for emotional support or health information; avoid entering protected health information into public AI systems without a Business Associate Agreement; treat AI-generated output in your workflow as unverified until you have reviewed it yourself; and find out what AI tools are running in your organization and whether governance structures exist.
Does AI literacy mean I need technical training?
No. It means developing enough working knowledge to evaluate tools, recognize their limitations, and maintain the critical professional judgment that good clinical practice has always required. The goal is not technical mastery. It is professional competence in a changed environment.
How is this different from general concerns about technology in healthcare?
Most technology discussions focus on organizational risk, privacy, compliance, governance. AI literacy addresses something more immediate: the direct effects of AI on individual clinical encounters, the therapeutic relationship, and the judgment of the clinician in the room. That dimension is personal, not organizational, and it requires a personal response.
About the Author
Dr. Ernest Wayde is the Founder and Principal of Wayde AI, a healthcare AI ethics consulting firm. He works with healthcare and behavioral health organizations on responsible AI adoption, governance, risk management, and implementation strategy.
References
American Psychological Association. (2026). Chatbots and mental health survey: Findings on psychologists' experiences with AI chatbots in mental healthcare. https://www.apa.org/pubs/reports/chatbots-mental-health-2026
American Psychological Association. (2026, March). AI in the therapist's office: Uptake increases, caution persists. Monitor on Psychology. https://www.apa.org/monitor/2026/03/ai-reshaping-therapy
Chatzichristos, C., Chatzichristos, G., Borremans, I., Gruyaert, S., De Vos, I., De Vos, M., & De Backere, F. (2025). Bridging the AI-literacy gap in health care: Qualitative analysis of the Flanders case study. Journal of Medical Internet Research, 27, e76709. https://doi.org/10.2196/76709
Ruder, E. (2026, June 16). 36% of psychologists report patient chatbot dependency. Becker's Behavioral Health. https://www.beckersbehavioralhealth.com/ai-2/36-of-psychologists-report-patient-chatbot-dependency/
Digital Medicine Society, & Google for Health. (2026). 2026 Health AI Horizon: Three key insights for healthcare leaders. https://dimesociety.org/newsroom/blog/3-key-insights-for-the-2026-health-ai-horizon/
Goddard, K., Roudsari, A., & Wyatt, J. C. (2012). Automation bias: a systematic review of frequency, effect mediators, and mitigators. Journal of the American Medical Informatics Association : JAMIA, 19(1), 121–127. https://doi.org/10.1136/amiajnl-2011-000089
Moore, J., Grabb, D., Agnew, W., Klyman, K., Chancellor, S., Ong, D. C., & Haber, N. (2025, June). Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (pp. 599-627). https://arxiv.org/abs/2504.18412
Rousmaniere, T., Goldberg, S. B., & Torous, J. (2026). Large language models as mental health providers. The Lancet Psychiatry, 13(1), 7–9. https://doi.org/10.1016/S2215-0366(25)00269-X
U.S. Food and Drug Administration. (2026). Artificial intelligence-enabled medical devices. https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices
Wang, D. Y., Ding, J., Sun, A. L., Liu, S. G., Jiang, D., Li, N., & Yu, J. K. (2023). Artificial intelligence suppression as a strategy to mitigate artificial intelligence automation bias. Journal of the American Medical Informatics Association, 30(10), 1684-1692.
Wolters Kluwer Health. (2026, June). AI literacy: The missing link in digital health technology. https://www.wolterskluwer.com/en/expert-insights/ai-literacy-the-missing-link-in-digital-health-tech
Markus, A., Carolus, A., & Wienrich, C. (2025). Objective measurement of AI literacy: development and validation of the AI competency objective scale (AICOS). Computers and Education: Artificial Intelligence, 100485. https://www.sciencedirect.com/science/article/pii/S2666920X25001250
van Zyl, L. E. (2026). The unintended negative consequences of artificial intelligence use for psychologists. Frontiers in Psychology, 17, 1729050. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2026.1729050/full
Comments