Why clinicians stop using AI tools, and what to do about it
1. One bad answer early on
What goes wrong: A clinician’s first answer is wrong, and they never come back.
What to do: Start with a small group and a narrow task, test it hard first, and expand once it’s reliable.
2. No way to check the answer
What goes wrong: Verifying takes longer than doing the task without the AI.
What to do: Show the source passage next to every claim so a check takes seconds.
3. Extra clicks and extra screens
What goes wrong: The tool lives outside the workflow, so using it feels like more work.
What to do: Put it inside the EHR screens people already use, and measure time saved honestly.
4. Confident answers with no limits
What goes wrong: The AI sounds sure when it isn’t, and clinicians learn to ignore it.
What to do: Let it say “I don’t know,” and make the decline useful by saying what’s missing.
5. Feedback goes nowhere
What goes wrong: Clinicians report a problem, see no change, and stop reporting.
What to do: Make flagging one click, name a reviewer, and tell reporters what was fixed.
6. Unclear responsibility
What goes wrong: Clinicians don’t know what they’re accountable for when they use an AI answer.
What to do: State in plain language what the AI does, what a clinician must still check, and who owns the tool.
7. Skipping the clinicians who’ll use it
What goes wrong: The tool is designed without the people who use it daily.
What to do: Name clinical champions early, and let them help choose the test cases and grade the results.