AI Essentials · a two-minute exercise
Sign or send back.
An AI drafted five clinical notes from what was recorded. Some of what you are about to read was never recorded at all.
Every draft reads well. That is the point. Fluent prose is what these tools are built to produce, and fluency carries no signal about whether the content is true. Your job is the second read.
How it works
- The left panel holds the source: what was recorded, said, or already in the chart.
- The right panel holds the AI draft. Click any statement the source does not support.
- Sign the note or send it back. You will see what you caught and what you signed.
Nothing you do here is stored or sent anywhere. Every patient, encounter, and document is invented for teaching and carries no clinical guidance.
Click a statement to flag it. Click again to clear it.
End of shift
0 of 10 caught
0
Caught
Statements the source did not support, flagged before signing.
0
Signed anyway
Fabrications that went into the record with your name on them.
0
Over-flagged
Supported statements sent back. Real cost in a real workflow.
The part nobody watches
How long you spent on each note
Note three was faithful all the way through. Note four hid a single word. Watch what a quiet stretch does to the time you give the note after it.
Where this sits
This is one habit, from one module, of six.
AI Essentials covers what these tools are, how they fail, where patient information can and cannot go, and the verification habits that hold up under a real workload.
Every patient, encounter, and document in this exercise is invented. The drafts were written to demonstrate known failure patterns in AI-generated clinical documentation and carry no clinical guidance.