Loading request...
AI scribes need to significantly improve accuracy, especially for critical data like SI/HI cues, medication dosages, and medical history, to reduce the need for extensive manual verification and ensure patient safety.
I tried an AI scribe to cut after hours charting.... I now double check half the notes. The tool misses SI and HI cues, flips doses like 5 mg to 50 mg, and invents history. I spend another 10 to 15 minutes per patient fixing errors, so the time savings disappear. Vendors (i dont wanna name them here) show 90 to 95% accuracy in demos. My psych sessions land closer to 85 to 90%. Fast speech, tangents, and interruptions break it. I see large omission rates and some fabrications like made up MSE details. I also see rare hallucinations that add risks with no clear reason. Automation bias worries me. It pushes you to sign bad risk assessments. Emotional outbursts and collateral history push errors even higher. Scripted benchmarks do not match real intakes. I audit risks and meds every visit. I want tools tuned for psych. I plan a 20 visit trial to track my error rate. I could get manual time down to 5 to 10 minutes if I stay alert. Does this match your experience with psych scribes that handle MSEs and therapy notes without constant babysitting? EDIT: I gave freed ai a try (free tier only), mostly out of curiosity. It handled rapid speech and interruptions better than other tools I've tested, but it doesntt change the CORE issue here: I still have to VERIFY everything. SI/HI, meds, and MSEs all require manual review because the liability doesnt move. It may help as a rough draft or checklist, but it doesnt eliminate the time or responsibility that makes scribes questionable in my use case.