I Asked a Real Therapist to Review My AI Therapy System—Here's What He Said

Is using AI for Self-Discovery actually useful?

When I first wrote my AI Therapy Workflow, I honestly thought I was alone.

Maybe a few people would find it interesting, but I figured most would think using AI for self-discovery was too weird or too personal.

I was completely wrong.

That post resonated with more readers than anything I'd written. People shared their own stories of using AI for reflection. They asked deeper questions about the psychological implications. Most surprisingly, it caught the attention of

When Freud Meets AI

, a practicing therapist who's generous to share his professional perspective.

Marc brings something I can't: clinical expertise in how humans actually process emotions and change behavior. While he's optimistic about AI's potential in therapy and self-discovery, he also sees limitations that only trained human therapists can address.

If you've been curious about mixing therapy concepts with AI—or if you tried my workflow and wondered "is this actually psychologically sound?"—you're going to love Marc's insights:

Marc, take it away.

---

Hello, Marc here!

What if somebody offered you something to reduce anxiety that costs nothing? It has few side effects and will boost your self-efficacy. It might not be as effective as therapy or a drug, but it won't make you dizzy or your stomach hurt either. Do you want to try it?

Here is a pen and paper.

Journaling is a powerful tool for mental health. You can use it for self-reflection, to process what you feel and think, and it can also be integrated in therapy. Journaling leads to an average score reduction of 5% in common mental health questionnaires, and greater benefits are seen for those with anxiety or PTSD (Post-Traumatic Stress Disorder). As a therapist, I encourage my patients to write and use journaling as a tool for self-reflection.

Obviously, it does not have to be paper, as many people journal digitally.

I was therefore intrigued when I read Wyndo’s article My AI Therapy Workflow: Turn Claude/ChatGPT and NotebookLM Into Your Self-Discovery Tool.” He describes in detail how he used NotebookLM to process over three years of personal journals—over three years! It's unbelievable as a therapist to see a patient and take the time to skim through those journals yourself. You need somebody to summarize this, spot patterns, and hint at what is left out in those pages. In comes AI.

Not quite therapy, but close: When AI helps you reflect smarter

What I found fascinating is how Wyndo describes using NotebookLM first to process journals, uncovering recurring themes and patterns. He then uses Claude to transform this data on self-recovery into a tool that can nudge him in specific situations to change his actions. From a therapy perspective, I appreciate how this empowers him to self-reflect. Self-reflection differs from therapy, but it can undoubtedly serve as a powerful complement.

Besides its empowering effect, he uses his time efficiently to uncover insights from his journaling. This could again support mental health. In therapy, you have only 50 minutes once a week to work with a client. If you spend roughly an hour in therapy, what else are you doing with the remaining 167 hours of the week besides eating, sleeping, and working? As a therapist, you always strive to integrate as much therapy as possible into your client's everyday life. Why? Real life happens outside the therapy office. In my opinion, Wyndo shows an interesting workflow that exactly incorporates the patterns he uncovers through self-reflection into his daily life.

Yet, a therapist will not necessarily agree with your opinions and reflections. They might gently challenge you. They will support you in questioning your own thinking because our thinking is often biased and affected by cognitive errors. That is not what AI will necessarily do. It is a real issue that AI can act overly agreeable or flattering - OpenAI removed an update because of sycophancy in GPT-4o in April 25. Keeping AI in check to serve as a critical, reflective voice rather than a sycophantic ally that naively praises dysfunctional thoughts may be a genuine challenge.

Ultimately, it will encounter limitations because AI lacks the rich life experience of a real human being. While it can summarize and reflect back based on probabilities, it may overlook the deeper meanings of life that only a genuine human therapeutic interaction can reveal.

I wish I could share a story now on how AI has miraculously supported self-reflection in a client but the truth is that I cannot because so far I have not met a client who uses AI in a way comparable to Wyndo. When I ask them they explain that they use it for work or in their private life but that they genuinely enjoy the fact that they can come to a place and talk with a real person about their thoughts, feelings and actions. Therapy for them is a safe space where it is perfectly fine to cry and talk about distressing thoughts that keep them up at night.

Recently a young client told me how helpful it is for him to reflect on what we talked about when he drives home. What he perceives as helpful is that there is a certain space and time reserved for a personal exchange and then he can back to his daily life. I do not think he would want an instantly available AI mental health support. For him, putting some distance between his day-to-day life and the therapy space is exactly what he perceives as useful and I do not think I am to judge.

Journaling becomes data—and things get complicated

While I remain optimistic about this, I can't deny that some aspects of this workflow give me a slight headache. When you journal, you generate patient-generated health data (PGHD). This data can be produced passively. For example, consider the step counts on a smartphone or a smartwatch that tracks your sleep patterns. However, it can also be generated actively, meaning you fill out a questionnaire or journal. Generally, physicians view passive data as more objective, whereas active data generated by patients is considered more reflective and subjective.

The subjective aspect of the data is precisely where AI can assist. What Wyndo describes is that AI does not replace his reflection, but rather enhances it. If AI aids you in improving your self-reflection, it could also foster better communication between patients and therapists. Imagine having a meaningful discussion when you start therapy because you’ve already begun to self-reflect on patterns and triggers you identified in your journal with AI's help. That could be a real benefit.

But I told you I had headaches, which is the major peril of data accuracy, security, and privacy. Maybe you journaled as a teenager. Just try to remember what would have happened if that journal had gotten into the hands of your parents or a friend in school who liked to make fun. Now, imagine what happens when you use generative AI to uncover patterns in your journal. You basically hand over sensitive information to NotebookLM. And that's a major concern for clinicians regarding patient-generated health data. Protecting patient privacy is of great importance. GenAI tools should be transparent about what happens with the data whenever data is shared.

Thinking of using AI for self-reflection? Here’s what to know first

Even without AI, journaling is a proven and accessible mental health tool. If you’re interested in taking it a step further by using tools like NotebookLM, Claude, or ChatGPT, here are a few things to consider from my perspective as both a therapist and someone who believes in the power of reflection:

• Privacy is everything: You’re uploading thoughts, patterns, and perhaps even memories of distressing moments. Ensure that the tool you’re using has a transparent privacy policy and be clear about how you want your data to be handled. Where is your data stored? Will it be used for further training? Can you delete it permanently?

• Local vs. cloud-based tools: If you’re technically inclined, consider using local models (like AnythingLLM or running a HuggingFace model locally). This keeps your journal entirely off the cloud and solely in your hands.

• Treat AI like a reflective partner, not a therapist. AI can help you identify patterns and even provide interesting questions to consider. However, it won’t challenge your blind spots or dysfunctional patterns the way a trained human can. It’s a mirror, not a guide. Understand its limitations.

• Curate what you share: You don’t need to feed your AI tool every personal detail. Avoid names, places, and personal specifics. Be intentional. Summarize a few key thoughts and prompt it with “Here are my last 5 journal entries—what themes do you see?”.

• Always bring it back to you: The output from AI is a starting point, not the final word. Let it ignite ideas, but trust your own interpretation and emotional intelligence to create meaning. You decide what to do with it, whether it fits or not. Do not blindly believe in something just because it comes from something called “intelligent.”.

Ultimately, self-reflection is a deeply personal process. If AI assists you along the way, that’s great. Just make sure you’re using tools that honor your privacy and support, rather than replace, your inner work. With this wonderful community of AI enthusiasts that Wyndo brings together, I am happy to discuss solutions that make his workflow even safer, more private, and secure.

And if you are in doubt, you can still stick to pen and paper.