Medical SLPs operate in high-stakes, fast-paced environments where documentation accuracy carries immediate clinical consequences — your note is read by people who will act on it within hours. The physician skims your bedside swallow eval before writing the diet order. Nursing reads your recommendations before the next meal. An error in a school progress note surfaces at the next review; an error here surfaces on a lunch tray.

Regulatory Context

• HIPAA governs all patient information; no PHI in public tools, ever
• BAA (Business Associate Agreement) required for any tool handling patient data — see what’s actually covered by provider
• Documentation supports medical decision-making, billing, and continuity of care
• Instrumental findings (MBSS, FEES) require clinician interpretation, never AI

The EMR Reality

You already work inside a documentation machine — smart phrases, copy-forward, dictation, templated flowsheets. An LLM is a new instrument in an old orchestra, and it inherits the EMR’s existing failure mode: text that propagates without being re-read. Copy-forward errors are a documented patient-safety problem; AI-generated text that gets pasted in and signed adds a second source of unexamined language. The discipline is the same one you already know from copy-forward: if you didn’t verify it this time, for this patient, it doesn’t go in the chart.

Appropriate Uses

• Structuring dysphagia progress notes and session documentation
• Organizing cognitive-linguistic assessment findings into narrative format
• Drafting patient/family education materials in plain language
• Structuring prior authorization and medical necessity letters
• Summarizing discharge recommendations into readable format
• Brainstorming functional goal wording for rehab settings

Worked Example: Organizing a Bedside Swallow Evaluation

“I am a medical SLP organizing a bedside swallowing evaluation note. Here are my findings: [de-identified observations — presentations trialed, textures, signs/symptoms observed per trial, patient alertness and positioning, what I recommended]. Organize into: history available at bedside, presentation, trial-by-trial observations, clinical impressions AS I STATED THEM, and my recommendations AS I STATED THEM. Do not add signs I did not observe, do not interpret aspiration risk, and do not generate or upgrade diet recommendations. Keep my clinical voice.”

What to check: The two danger zones are the impression and the recommendation — exactly where the reader acts. Confirm no observation got upgraded (“no overt signs of aspiration” is not “swallowing within functional limits”), and that the diet recommendation in the output is your recommendation, verbatim in substance. The dysphagia case study shows this exact failure caught in the wild — the plausibility of the error made it harder to detect.

Red Flags

• Entering patient names, MRNs, or PHI into public models
• AI interpreting instrumental findings (MBSS, FEES)
• Generating diet level recommendations or aspiration risk judgments
• Copying AI language directly into medical records unread — the copy-forward failure with a new engine
• Using AI output as basis for swallowing safety decisions
• Substituting AI summaries for clinical reasoning in discharge planning

Related

• Workflow: Discharge Planning (Medical)
• Prompts: Dysphagia Bedside Note · SOAP Note Formatter · Prior Authorization
• Domain: Dysphagia — the hard boundaries in full

See it in practice: Case study — MBSS documentation and the diet-recommendation error.