As the world of healthcare increasingly embraces AI, an area that seems to be impacted the most is clinical documentation. Specifically, in the world of therapy, there is an increasing use of AI scribes that record sessions and automatically generate notes. Even though this technology has the potential to save clinicians time, there is also a growing concern about what we are losing in the process. As a therapist-in-training, I've never used an AI scribe to record my sessions. Instead, I've developed my own system that I feel might be even more effective. Today, I'm going to share my entire workflow for taking notes after every session I have. I'll share the thinking behind each step, along with the specific tools I'm using and why. # Session debrief In a previous post, I already shared the difference between progress notes and personal notes and how I view them as two distinct forms of documentation. What I didn't mention was how I actually go about writing all of these notes. My goal is to keep things as simple as possible, without duplicating effort or wasting time. With this in mind, the linchpin of my entire workflow is a session debrief I perform immediately after every session. Usually within a few moments of a client walking out the door, I begin doing a quick debrief to capture my thoughts while they are still fresh. I've realized that it is incredibly important to do this as soon as possible. The longer I wait, the less I remember. The quality of my notes is substantially better when I do an immediate debrief compared to waiting even a few hours. I'll share the specific tools I use later in this post, but it is important to mention here that I don't physically write or type these session debriefs. The most effective way for me to capture my thoughts is via voice notes. Voice notes, in my opinion, are one of the most underutilized techniques when it comes to interacting with technology. We are able to capture our thoughts much faster by speaking them out loud rather than typing or writing them. Yes, it does feel a bit awkward at first, sitting in a room talking to yourself, but after doing this for several months now, it's become second nature. When it comes to privacy, I'll also mention that I never say any identifying information in my session debriefs. Even though I make it a point to only use secure tools that are HIPAA-compliant, I take the extra step of not saying anything that could potentially identify a client and keep things as general as possible. The structure of my debriefs is usually the following: - A brief summary of the session focus and content discussed - Interventions used and case conceptualization - Therapist reflections # Why are session debriefs better than AI scribes? At first glance, having an AI scribe record your sessions might seem like a great idea. They can capture every word that was said and provide you with a session recap with high accuracy. This, in turn, can even be used to create automatic progress notes. Sounds like a dream, right? But here's the problem: relying on a word-by-word session transcript alone is a missed opportunity. Even though I've never personally used an AI scribe, I imagine they can do a lot. Other than progress notes, I'm sure they can also provide you with some supervision, feedback, and conceptualization. However, to me, this feels like the equivalent of giving a chef your credit card to buy a bunch of ingredients from a grocery store and make something for you. As therapists-in-training, or even as licensed clinicians, we must be mindful about how much of our work we're letting AI do for us. If we keep letting the chef do everything, we'll never learn how to cook for ourselves. With my approach of doing a session debrief and talking everything out into a voice note, I'm forcing myself to prepare all the ingredients (recalling relevant session details) and cook the meal (explaining my own conceptualization). It will be imperfect. It will be messy. But it will also contain more depth. Additionally, by including my therapist reflections, there's a certain flavor I'm adding that an AI scribe would never be able to replicate. I answer reflection questions such as: - Where did I feel stuck? What did I do with that? - What went well? Were there any standout moments? - What was the emotional weight of the room? How was the nonverbal experience? These are the types of things that an AI scribe will never be able to capture with just a transcript. # What do I do with session debriefs? At a high level, once I record a voice note with my session debrief, the transcript becomes my personal case note. These transcripts are already written in my own words and don't require any editing. I simply paste the entire debrief into my notes verbatim. It's easy enough to transform a session debrief into a formal progress note. It's just a matter of rewording a few sentences and removing the personal reflections. Over time, I end up building a log of these session debriefs for each client, which I call a longitudinal client file. The term longitudinal is important here. These client files aren't static; they change every week. With each session that passes, the file becomes richer with more context about the work being done. There is a profound difference between having one individual session transcript versus a dozen session debriefs all saved in the same place. Unlike my progress notes, these client files are filled with personal reflections and clinical conceptualizations that have evolved over time. This is where the magic happens. I can find patterns in my own work and build a true "second brain" for each client I work with, allowing me to reference a level of detail that would never have been possible otherwise. It's worth reiterating here that these client files don't contain any identifying information. The primary focus of the files is the clinical work itself, along with my own reflections, not so much the content of the client’s life beyond the bare minimum needed to understand the context of the work. I’m also mindful that even non-identifying details can become identifiable in combination, so I keep content intentionally broad. Because of this, I feel comfortable having these files in the first place. Even though they don't contain any identifying information, I still treat my client files with the utmost care and security, which I'll discuss in the next section about specific tools. In a future post, I'll share how I use AI as a layer above my client files to make all of this post-processing work as quick as possible. To me, this is a much more effective way of using AI without losing the important work we should be doing ourselves. # Specific tools As you can probably tell, it's very important to me that the tools I use are secure. I try to keep things simple, without any unnecessary exposure to my data. The first tool I use is Obsidian. At the end of the day, Obsidian is just a text editor. It allows you to write notes and organize them cleanly. What I appreciate most about Obsidian is how all your notes are stored locally on your computer. There's no cloud involved unless you choose to set up a syncing service. I keep all my client files in a separate folder from all my other notes. They only exist on my computer, which is already encrypted at an OS level using Apple's FileVault. While no system is 100% secure, I feel better knowing everything is on my computer and not in the cloud. The next set of tools I use are for the voice notes I record. I've experimented with two different options here. The first is the built-in Voice Memos app on my iPhone. This is a great option that gets the job done. After recording, the app automatically transcribes the audio into text, which you can copy and paste into your notes. Once I save the transcript, I delete the voice note itself. I like this option a lot because it's free and you can talk in an open-ended way like you would naturally. The downside is your transcript will read more like a conversation you're having with yourself than a note. The second option I’ve been experimenting with is using an AI app called Wispr Flow. The concept is simple: you open up your text editor and hold the "Fn" key on your keyboard while you speak. The app will use AI to turn your voice into clean sentences. This is different than using the normal dictation feature because the AI gets smarter each time you use it. It begins to build a dictionary of the words you use and will adapt your writing style depending on what you're writing. I've been using this not only for clinical work, but also for basic things like writing emails and messages. What I love most about this app is all the privacy features. Wispr Flow offers a HIPAA-compliant mode with a BAA and zero data retention, which is why I’ve been comfortable experimenting with it in my workflow. This means the app doesn't store any of your audio or transcripts on their servers. My only complaint about Wispr Flow is that you have to record a few sentences at a time. It doesn't work well if you try to record a full debrief in one go. However, this also means your notes will probably sound more like a note than a voice transcript, which is nice. # Recap To recap, I strongly believe in the power of recording session debriefs rather than relying on AI scribes. Doing a quick debrief via voice note after a session is a great way to capture what just happened and what you felt as a therapist. Building a log of these debriefs for each client will unlock a wealth of insights over time that will make you a better therapist. How do you normally take session notes? Do you use AI scribes? I'd love to hear from you. --- Related Notes: - [[Therapy With Maneet]] This note was originally created on **April 19, 2026**.