What They Studied

Miner and colleagues examined whether AI systems could generate clinical documentation that meets the standards expected in real healthcare settings. They evaluated AI-generated clinical notes for accuracy, completeness, and clinical appropriateness, comparing them against notes written by clinicians to understand where AI adds value and where it falls short in the documentation workflow.

What They Found

• AI-generated notes captured the core clinical information accurately in most cases: diagnoses, key findings, and treatment plans were generally correct.
• However, the AI-generated notes lacked the individualized clinical reasoning and patient-specific nuance that characterized human-written documentation.
• Templated and formulaic language was a consistent pattern in AI outputs, making notes feel generic even when factually correct.
• The notes sometimes missed contextual details that a clinician would naturally include, such as social determinants, patient preferences, or clinical judgment calls that inform the plan of care.
• Clinician review and editing of AI-generated drafts produced notes that were both efficient and appropriately individualized.

Methodology

The study evaluated AI-generated notes across clinical encounter types, with physician reviewers assessing accuracy, completeness, and clinical utility. The evaluation framework considered both factual correctness and the subtler dimensions of clinical documentation quality. The sample was drawn from general internal medicine encounters, limiting generalizability to other specialties.

What This Means for SLPs

• This directly validates the “copilot not autopilot” approach: let AI create the structured draft, then add your clinical voice and patient-specific details.
• SLP documentation is particularly nuanced. Describing communication behaviors, family dynamics, and functional outcomes requires the kind of individualization AI currently lacks.
• The finding about generic language is especially relevant: SLP notes that read as templated may not meet payer requirements for medical necessity or individualized treatment planning.
• A practical workflow emerges: use AI to handle the repetitive structural elements (session format, standardized descriptions, billing language) while the clinician focuses on what makes this patient’s story unique.
• This supports investing time in learning to edit AI drafts effectively rather than trying to get AI to produce final-quality notes independently.

Limitations to Keep in Mind

• The study evaluated general medical documentation, not rehabilitation or communication disorder notes, which have distinct documentation requirements and conventions.
• The comparison focused on note quality at a single point in time, not on how AI-assisted documentation affects longitudinal care or clinical decision-making.
• How clinicians interact with AI drafts in real-time workflow, including the risk of automation bias where clinicians accept AI text without adequate review, was not directly studied.

The Bottom Line

AI can accurately structure clinical documentation, but the individualized reasoning and patient-specific nuance that define quality notes still require a clinician’s voice and judgment.