Rolling out AI documentation across a clinic

The AI use policy and consent wording are written before the first consult is recorded. The policy names who may use the scribe, which consult types are excluded, how long audio is held, and who signs.

One form of consent wording is agreed for the whole practice, with what happens when a patient declines.

The first group runs with three to six clinicians over four to six weeks, including one who documents in an unusual pattern and one who argued against the purchase.

Record the group, the dates and the consult types in scope. Documentation time per clinician is measured in the week before recording starts.

Every clinician is trained on the review step before their first recorded consult, against a fixed checklist: medications and doses, laterality, numbers spoken aloud, the plan, and anything said in the room that does not belong in the record.

The first review session runs as a group, on drafts from consented consults.

Four numbers per clinician carry the decision to continue.

• Time from the end of a consult to a signed entry.
• The proportion of drafts that needed more than a light edit.
• Errors found at review, counted as omission, addition or transcription.
• Consults where the scribe was switched off, with the reason.

A sample of signed notes is audited against the policy before access widens beyond the first group, at a fixed sample size, against a written standard, with a record of who reviewed each entry and when.

The pass condition is in the policy before the sample is pulled.

Access widens one group at a time on the conditions the first group met: policy read, consent wording in use, training done, first audit booked.

Use stays optional, and a clinician who documents without it works to the same standard. The audit repeats across the clinic at the interval the policy sets.