Don’t let convenience outrun clinical judgment
Believe it or not, there was a time before GPS (yes, I am that old). Back then, people needed to unfold one of those paper maps and figure out how to get from one place to another (and then how to fold the map back up). Nowadays, thanks to technology, I don’t even need to know where I am going. Google Maps knows, and that is enough for me. Likewise, in ancient history (before smartphones), I used to remember at least ten phone numbers: my home number, my grandparents’, and those of my best friends. Now I can hardly remember my wife’s.
Since the beginning of time, we have used technology to make our lives simpler. We could even argue that this is the purpose of technology. From the abacus and the calculator to Claude, we humans love to figure out how to make things easier and more efficient. This is often called cognitive offloading: using external tools to reduce the amount of mental processing required to do something. In the past, this might have meant mostly taking notes. Nowadays, it includes consulting Google or asking ChatGPT to summarize something, interpret information, or even write something for us (or as us).
The advent of LLMs and AI scribes
General-purpose AI systems powered by Large Language Models (LLMs), such as Claude or ChatGPT, have begun to revolutionize how we work. Months of research can be done in days; piles of books can be summarized in hours. They help us do things faster and often (but not always) better.
Just like clients, therapists and health care professionals are enthusiastically jumping on the AI bandwagon. The use of AI in health care, including AI scribes (note-takers), has grown rapidly1. It is undeniable that AI can reduce the administrative burden and make life easier for clinicians. As discussed in my last article, AI can also be useful for psychoeducation, providing additional resources, and as a tool for reflection.
As harmless as scribes may appear, they present important challenges for clinicians. Should clients give informed consent before an AI scribe is used? (probably yes2). Does sending client-identifiable information to a general-purpose LLM constitute a breach of confidentiality? (most likely yes, since information shared with an LLM is not confidential, at least not in the psychotherapy sense). Would it be a HIPAA violation3? (depending on the information, but potentially also yes). Things are even more delicate when considering that even purpose-built scribes can hallucinate, make inferences, or make mistakes4.
The deep end of the pool
However, note-taking is only the shallow end of the pool. Some AI scribes generate treatment language, conceptualizations, diagnoses, and suggested interventions, blurring the boundary between scribe and clinician. What is more, some therapists are using ChatGPT during sessions asking for insights or interventions5. Products are being marketed that listen during therapy sessions and provide therapists with real-time prompts, insights, and guidance.
When myth mirrors reality
In the Dialogues, Socrates tells Phaedrus a story about the Egyptian god Theuth. Theuth presents King Thamus with the gift of writing, claiming that it will make people wiser and improve their memories. The king is unimpressed. He argues that writing will produce forgetfulness because people will stop exercising their own memory and rely instead on external sources. They will think themselves intelligent because they have access to enormous amounts of information, while lacking real wisdom. It is uncanny how well this myth mirrors our current predicament.
What can go wrong?
As I have expressed, I am convinced that AI is an incredibly valuable tool when we know how to use it. However, outsourcing clinical thinking to LLMs presents potential risks that need to be acknowledged:
- Relational disruption: Therapeutic success relies heavily on the relationship. Introducing AI into the equation changes the dyad, creating an artificial third: an opinionated, know-it-all chatbot.
- Diluted presence: Real-time prompts, like an overactive inner critic, can get in the way of authentic presence. It may be difficult, if not impossible, to create a true bond when the therapist is simultaneously engaging with an LLM.
- Automation bias and generic interventions: Because AI responses sound polished and authoritative, therapists may give them more weight than they deserve. Likewise, a chatbot may suggest elegant interventions that are technically plausible but misattuned or disconnected from the client’s reality. When that happens, besides potential risks6, the client may not feel understood, compromising the relationship.
- Clinical deskilling: There is enough research to suggest that when we find a tool that makes something faster and easier, we are inclined to keep using it. As this process is repeated, we may stop practicing the underlying skill, become less competent, and rely even more on the tool (think of me and phone numbers). One study showed how by regularly using AI for diagnosis, specialists became less accurate when they did not have access to the tool7.
Offloading formulation, diagnosis, and treatment planning may gradually reduce clinicians’ skills. It may also prevent newer clinicians, who began practicing after the arrival of these tools, from developing those inner capacities in the first place. Therapists may slowly lose their clinical acumen.
The way forward
We have always used tools to extend our minds: dictionaries, calculators, GPS, Google… In some ways, AI is simply the latest iteration of this pattern, but it goes further because it can take over functions such as formulation, diagnosis, and treatment planning that have traditionally belonged to the clinician.
Psychotherapy has always been (and hopefully will continue to be) a profound encounter in which two human beings come together to work toward healing. This requires a human therapist who can create a safe environment and an authentic relationship where clients feel seen, understood, and safe enough to explore the deepest and darkest corners of their psyche.
Let’s use AI for what it is good at, but let’s not offload human connection. That is where true healing occurs.
AI can reduce administrative burden, but clinicians and organizations need thoughtful boundaries around privacy, consent, clinical judgment, and therapeutic presence. Reach out for lectures, professional trainings, and consultation on the responsible use of AI in psychotherapy.
- See https://www.ama-assn.org/practice-management/digital-health/more-80-physicians-use-ai-professionally-ama-survey and https://www.theguardian.com/australia-news/2026/mar/29/doctors-using-ai-for-notes-australia ↩︎
- https://www.nature.com/articles/s41591-026-04506-3 ↩︎
- HIPAA stands for the Health Insurance Portability and Accountability Act, is a federal law that aims to protect sensitive patient health information. It includes provisions for their confidentiality, privacy and security. ↩︎
- https://www.nature.com/articles/s41746-025-01895-6 ↩︎
- https://futurism.com/patients-furious-therapists-using-ai ↩︎
- https://spirals.stanford.edu/assets/pdf/moore_expressing_2025.pdf ↩︎
- https://www.thelancet.com/journals/langas/article/PIIS2468-1253%2825%2900133-5/abstract? ↩︎
