Reading LLM-Generated Clinical Writing
Most clinicians are developing this skill informally and privately. This section makes it explicit, shared, and craft-based.
Why This Matters
LLM-generated clinical writing is a new kind of artifact. It isn't your writing and it isn't the machine's writing — it's a collaboration that inherits the strengths and blind spots of both your prompt and the model's training data.
Most clinicians are already developing the skill of reading this writing critically. They notice when a goal sounds right but isn't performable. They catch when a progress note adds clinical interpretations that weren't in the raw data. They recognize when the phrasing is too smooth for the complexity of the case.
This section names that skill and makes it shareable. The goal isn't suspicion — it's the same clinical precision you'd bring to any professional document that carries your name.
Common Patterns in LLM Clinical Writing
This section is in development. We are collecting and categorizing patterns in LLM-generated clinical text — the recurring shapes that show up across models, settings, and clinical domains.
More patterns coming. Contributions welcome. Unfamiliar terms along the way? The glossary defines each one from both the clinical and technical side.
Antipatterns: The Worked Examples
The patterns above are the theory. These are the practice: concrete before-and-after examples of LLM output that looks plausible and fails clinically — what went wrong, why the model produced it, and exactly how to fix it.
The Polished but Unmeasurable Goal
When AI-generated goals sound professional but fail the most basic test: can you actually measure this?
Generic AI Voice
When LLM output replaces your clinical observations with polished but interchangeable language that could describe any client.
The Clinic-Only Goal
When a goal measures performance in therapy but says nothing about the real world where the skill actually matters.
Copy-Paste PHI
When a clinician pastes real patient or student names, dates of birth, and diagnoses directly into a public LLM.
Hallucinated Test Scores
When the LLM invents standardized test scores, percentile ranks, or normative data that you never provided.
The Overcorrection
When repeated AI-assisted revisions strip a note of its original clinical observations until it is technically polished but clinically empty.
One-Size-Fits-All Goals
When the same prompt produces nearly identical goals for every student, differing only in the name at the top.
The Scope Creep
When the LLM generates recommendations outside the SLP's scope of practice, including medication suggestions, educational placements, or medical diagnoses.
Submit a Specimen
Have an LLM-generated clinical artifact you've been thinking about? Share it here. Your noticings help build the field's shared vocabulary for reading AI-assisted clinical writing.
No client names, dates of birth, locations, medical record numbers, or any details that could identify a real person. De-identify everything before submitting. This form is not HIPAA-compliant.
All submissions are reviewed by a clinician. Nothing is published without explicit permission.