Written by Wani Sharma, Owl Practice’s Product Manager
More clinicians are using AI to help with documentation now, and I think that’s a good thing. Used well, it takes the weight of admin work off their plate. Notes get done faster, workflows get simpler, and clinicians get some of their day back.
But the more I talk to customers about it, the more I notice that the questions I get aren’t really about what the AI can do.
They’re about what happens to the conversation after the session is over.
- How long are transcripts kept?
- Who owns them?
- Is any of this being used to train a model?
- Can a transcript be deleted, and what happens to it a year from now?
I’ve heard versions of these in customer calls, in industry threads, in offhand comments at conferences. Honestly, they’re the same questions I’d be asking. None of them are about whether the AI can write a decent note. They’re about trust, and in mental healthcare that distinction is everything.
The Conversation We Had Before We Built Anything
When we started building Smart Notes, Owl Practice’s AI notetaker, the first real debate wasn’t about model accuracy or which features to build. It was about what we were actually comfortable doing with a therapy session once our technology touched it.
That’s the part of building healthcare AI most people never see. The decisions about retention, ownership, and data use happen behind the scenes, early, and they’re easy to defer until later. We decided to treat them as product requirements instead of compliance footnotes to sort out after launch.
One idea anchored the whole thing: the customer’s data belongs to the customer. It sounds obvious. When building an AI product, upholding this value means turning down things that would otherwise be convenient.
Why Ownership Isn’t an Abstract Question
A therapy transcript is not like other data. People say things in session they may never say anywhere else, about trauma, addiction, grief, their marriage, the worst week of their life. When AI enters that room, even just to help with the note afterward, the clinician needs to trust that the tool respects what was said.
So we made a few deliberate calls.
Do We Train Models on Customer Sessions? No.
We don’t use customer session data to train AI models. That started as an ethical line for us, not a legal one, though it happens to satisfy both.
A Smart Notes transcript exists to do exactly one job: help a clinician write a clinical note. That’s the entire purpose of it. A session between a client and their therapist should stay what it was meant to be, a private conversation, not a data set that is later used to improve someone’s model.
What Happens to the Transcript?
Smart Notes deletes transcripts automatically after seven days.
Seven was a judgement call. Clinicians sometimes need a window to review a note, regenerate it, or come back and finalize it after a busy day, and a same-day cutoff felt too tight for how people actually work. But I couldn’t find a good reason to sit on something this sensitive indefinitely either. Seven days covers the real workflow without turning our servers into a long-term archive of people’s therapy sessions.
For clinicians who don’t need the seven days, there’s an option to delete the transcript the moment the note is generated.
The finished clinical note is a different thing entirely. The clinician reviews it, edits it, signs it, and it becomes part of the client’s record. The transcript is just the scaffolding. Once the note is built, the scaffolding comes down.
Privacy Is Also a Question of Time
It’s easy to think about privacy only as who can reach your data. The part that gets less attention is how long the data exists at all.
Anything you retain, you’re on the hook to protect for as long as you keep it. More retention means more surface area and more time for something to go wrong. You can’t engineer away every risk, but you can decide which risks are worth carrying. Cutting transcript retention was one we decided we didn’t need to hold.
Keep what’s useful, drop what isn’t. Simple to say, but this concept shaped a lot of how we built Smart Notes.
Compliance Is the Bare Minimum
Smart Notes is built to support PIPEDA, PHIPA, and HIPAA. But when you’re evaluating any AI tool for a clinical setting, I’d push past the compliance checklist.
Compliance tells you what a company is required to do. It doesn’t tell you what a company chose to do when no rule forced its hand. That second answer is the one I’d want.
So if you’re weighing an AI note-taker, ask the vendor directly. Do you train on our data? How long do you keep transcripts? Can we delete them? Who actually controls the information once it’s in your system? How they answer tells you how they think about the problem and whether they care about you, your practice and clients at the end of the day.
What Trust Is Built On
When a client agrees to AI-assisted notes, they’re not just trusting their clinician. They’re trusting the tool behind the clinician, and every company standing behind that tool. That’s a lot of trust to hand over, often without realizing it’s being handed over at all.
The least we can do is explain, plainly, what we collect, why it exists, how long it lasts, and who controls it.
AI is going to keep making documentation easier, and I’m happy about that. But I don’t think lasting adoption comes from notes being generated a little faster than last year. It comes from clinicians feeling safe using these tools in the moments where privacy matters most.
And that safety isn’t something you bolt on at launch. It’s decided long before anyone launches a tool as game-changing as Smart Notes.
Owl Practice’s Smart Notes automate your note taking using privacy compliant AI, so you can spend more time with clients and less time on paperwork. Learn more about Smart Notes.
