Loading...
AI scribes in pediatric therapy
Give it the benefit of the doubt: fast, accurate, better every year. In a pediatric therapy room it still has to:
There’s a shorter path. Not convinced? Good.
Argue with us ↓First, the fair part
They can genuinely save time.[1] We’re not going to pretend otherwise, and our case doesn’t depend on AI being bad at its job. It gets better every year.
So assume it keeps getting better. Better at what, exactly?
You might be thinking
“If it saves time, why not take the win?”
Fair. But
Making a step faster doesn’t prove you need the step. If the facts can go straight into the record, writing prose about them faster is still extra work.
01
Pediatric therapy is loud, crowded and mostly not talking. Children don’t adapt themselves to the documentation technology.
| What happened | What the microphone heard |
|---|---|
| He pointed to the right picture. | [no speech detected] |
| She signed “more.” | [no speech detected] |
| Hand-over-hand on the first loop, then on her own. | [no speech detected] |
| A visual model and a tactile cue. | [no speech detected] |
| Another child answered for him. | “It’s a CAT!” |
| They moved to the gym halfway through. | [inaudible] |
You might be thinking
“Video AI will fix that.”
Fair. But
Maybe. Then it’s a camera in a room full of children, with its own consent and privacy questions. And it still ends the same way: turning what it saw into prose.
02
The workaround is to say the facts out loud. That just moves the data entry into the session, while you’re treating a child.
“Okay, for the notes: seven of ten correct, moderate verbal cues, with a visual model.”
You might be thinking
“Dictating is fast and natural for me.”
Fair. But
If it works for you, keep doing it. It just doesn’t prove you need AI to write prose, when the same facts can be entered directly in under a minute.
03 + 04
Even the best scribe turns what you observed into a paragraph. Then, when the progress report is due, something has to read the paragraphs and pull the numbers back out.
AI scribe
4 AI guesses. 4 rounds of your checking. 1 leap of faith.
Parlance
That’s the record. Nothing to reconstruct.
0 guesses. 0 reconstruction.
Stop generating prose just to extract the data back out of it.
Download the round trip as slides (PDF)You might be thinking
“People need readable notes, not database rows.”
Fair. But
Agreed. Parlance writes a clean, readable note from your data. The difference is that the sentence is written from the numbers, not the other way around.
The shorter path
When you pick the target, the score and the cue level, the meaning is fixed right there. Nothing has to be inferred later. Progress becomes arithmetic, not interpretation: the same data in, the same report out.
Don’t use AI to infer tomorrow what you already knew today.
You might be thinking
“AI can summarize thirty notes in seconds.”
Fair. But
It has to decide which phrases mean the same thing, which number goes with which goal, and what a missing detail means. Then you check its work. Data kept as data needs no second read.
You might be thinking
“Bad data in is still bad data out.”
Fair. But
True of any system. The difference is that you can see exactly where a number came from, with no guess in between.
This isn’t anti-AI
Some problems really do need interpretation and prediction, and AI is good at them. Turning a handful of known facts into a note isn’t one of those problems.
You might be thinking
“AI will be flawless one day. Then this argument goes away.”
Fair. But
Only the accuracy part goes away. The facts could still be stored directly, the note could still be written from them, and no one would need to dig them back out.
You might be thinking
“Some sessions don’t fit in a box.”
Fair. But
Then write that session however it needs writing. Free text is always there. Structure handles the routine sessions, which is most of them.
Structured firstReadable secondNo reconstruction required
Built for pediatric SLPs and OTs. Free text whenever a session needs it.