Generic AI Voice
When LLM output replaces your clinical observations with polished but interchangeable language that could describe any client.
You’ll recognize this instantly: the LLM takes your specific, idiosyncratic session notes and returns a paragraph that sounds professional but could describe literally any client with a similar diagnosis.
The Bad Example
Your raw notes:
kid was really engaged today, kept requesting the red car during play, used 2-word combos like “more car” and “my turn” (hasn’t done “my turn” before!) got frustrated when I changed the activity and threw the car once but recovered with a deep breath after I modeled it
The LLM output:
“The client demonstrated increased engagement during the session. Expressive language skills were observed, including two-word combinations. The client exhibited some difficulty with transitions, which was addressed through co-regulation strategies. Overall, the session reflected positive progress toward communication goals.”
Everything specific (the red car, “my turn” as a new utterance, the thrown car, the modeled deep breath) is gone. What’s left is a progress note that could be copy-pasted into any chart.
Why This Happens
LLMs are trained on vast amounts of professional writing. When you give them clinical notes, they pattern-match to the most common documentation style in their training data: generic, hedge-heavy, professional-sounding text. The model doesn’t know that “my turn” appearing for the first time is clinically significant; it just sees two words and maps them to “two-word combinations.”
The Fix
Add an explicit instruction to your prompts:
Preserve the specific details from my notes — exact utterances, specific materials, specific behaviors. Do not generalize "red car" into "toys" or "my turn" into "two-word combinations." The specifics ARE the clinical data.
The Fixed Version
“Student demonstrated increased engagement during a play-based activity with a red car. Produced novel two-word combination ‘my turn’ (first documented use of possessive + noun for turn-taking). Also produced ‘more car’ (recurrence + object). Difficulty with activity transition observed: threw car once. Recovered independently after SLP modeled deep breathing strategy. Co-regulation strategy effective within 30 seconds.”
Every detail from the original notes is preserved. The note is still organized and professional, but it’s this child’s note, not anyone else’s.
The Takeaway
If you can swap a client’s name into the note and it still makes sense, the note has lost its clinical value. Specificity is not just good writing. It’s the evidence that you were actually in the room.
Pair With
- Documentation tasks – how to preserve voice in structured notes
- Pattern: Family-Centered Prompting – a philosophy that naturally produces specific notes
- Antipattern: The Polished but Unmeasurable Goal – the goal-writing version of the same problem