Key takeaways:
- AI scribes were most often used for initial patient visits and telehealth consultations.
- Unedited AI-drafted instructions received the worst readability scores.
NEW ORLEANS — AI-drafted patient instructions in nephrology practices may be longer and harder to read, according to data presented at the National Kidney Foundation Spring Clinical Meetings.
AI scribes are being used more frequently in clinical practices for documentation purposes, according to Kevin Shi, MD, a nephrology fellow at University of California, San Francisco (UCSF), and colleagues. However, the way patients may comprehend AI-drafted instructions is unknown.
AI scribes are being used more frequently in clinical practices, but the way patients may comprehend AI-drafted instructions is unknown. Image: Adobe Stock
Kevin Shi, MD
“Although Al scribe use may increase documentation generation, the effect of the increasingly widespread utilization of such tools on patient communication is unclear,” Shi told Healio.
At a UCSF medical center, Shi and colleagues evaluated AI-drafted patient instructions across 19,691 outpatient nephrology encounters (761 using AI scribes, 18,930 without AI) from 2023 to 2025. The researchers assessed quality measures, including readability scores, character counts and the number of edits made to the AI drafts, including additions, removals and substitutions.
Results showed that AI scribes were most often used during initial patient visits and telehealth appointments. AI scribe use was associated with significantly higher odds of patient instruction documentation (adjusted OR = 2.18; 95% CI, 1.41-3.38).
Patient instructions drafted with AI were longer than human-written instructions (median character count, 1,451 vs. 398) and written at an advanced reading level compared with human written instructions, according to the researchers.
Furthermore, clinicians edited a median of 37% of AI-drafted instructions, with the most common changes being deletions (79%). Unedited AI-drafted instructions received the worst reading scores across four reading indices, the researchers wrote.
Overall, the findings suggest AI-drafted patient instructions may not be ready to distribute without human edits, Shi said.
“Although this is a small, single-center study, our findings suggest AI scribes do not produce ready-to-send patients instructions,” Shi said. “Human oversight seems necessary to generate acceptable patient-facing documentation.”
More research is needed to inform AI scribe implementation efforts, according to Shi.
“Future research should focus not only on AI scribe capabilities, but also realistic implementation barriers like patient and provider attitudes toward AI, cost, etc,” Shi said. “This will be critical to delivering high quality care.”
For more information:
Kevin Shi, MD, can be reached at nephrology@healio.com.
Sources/Disclosures
Source:
Shi K, et al. Abstract G-589. Presented at: NKF Spring Clinical Meetings; May 6-10, 2026; New Orleans.
Disclosures:
Shi reports no relevant financial disclosures.
Ask a clinical question and tap into Healio AI’s knowledge base.
- PubMed, enrolling/recruiting trials, guidelines
- Clinical Guidance, Healio CME, FDA news
- Healio’s exclusive daily news coverage of clinical data
<
Leave a Reply