The pitch for AI medical scribes is straightforward: the tool listens to your patient consultation, writes the notes, and gives you your time back. A JAMA study found ambient scribing cut clinician burnout from 51.9% to 38.8% in 30 days and saved roughly 26% of documentation time. On paper, it's one of AI's clearest wins.
Dr. Benn Gooch, a UK general practitioner and member of the Royal College of GPs' Informatics Group, bought in completely. He used an AI scribe (MHRA-registered and GDPR-compliant, though he doesn't name it) for 18 months. Then he stopped.
His reasons go deeper than "it didn't work." The scribe worked fine. That was the problem.
Appointments Got Longer, Not Shorter
Gooch found that once patients knew everything was being recorded comprehensively, the implicit time contract of a GP appointment dissolved. Patients talked longer. He documented more. Surgeries ran 30 minutes over schedule. The tool that was supposed to save time ended up consuming it.
The Notes Were Accurate but Unrecognizable
When Gooch reviewed notes from a patient he'd seen six weeks earlier, the scribe's output was factually correct but "clinically unrecognizable." The AI organized information by structure rather than clinical significance. His priorities, his reasoning, his professional judgment about what mattered most - none of that came through. "The voice in the note was not my voice," he writes.
This connects to a subtler problem: too many issues got documented per visit. The scribe captured everything, creating what Gooch calls "a creeping sense of clinical overextension" with loose threads he felt obligated to follow up on.
The Real Cost: Outsourcing Your Thinking
Gooch's sharpest argument is that writing clinical notes isn't just paperwork. It's a cognitive process that drives clinical thinking. When the scribe took over, his clinical memory degraded. He became, in his words, "a passive observer in encounters where I had previously been an active architect."
He's particularly worried about medical trainees. If new doctors never develop the habit of synthesizing patient encounters into structured notes, what happens to their diagnostic reasoning?
Gooch is careful to say he's not calling for AI scribes to be pulled from the market. He acknowledges the burnout data. But he points out that the studies use 30- and 90-day measurement windows, too short to catch the slow erosion he experienced. A separate rapid review screening over 1,400 studies found that burnout measured on standardized scales was "essentially unaffected."
This isn't an anti-AI argument. It's a reminder that efficiency gains in one metric can mask losses in others. For knowledge workers outside medicine - consultants, lawyers, analysts - the same question applies: when you hand off the writing, are you also handing off the thinking?