How AI works across a visit: recording, transcription, and the note
The end-to-end picture — where the transcript is generated, how it powers the note draft and the code/medication suggestions, and where you stay in control.
- Updated
- Updated
- Reading time
- 6 min read
- Tags
- aitranscriptionscribenotesbilling
One conversation A finished note
See where the transcript is generated and how it powers the note, codes, and prescriptions - with you in control at every step
EnWella uses AI at three points in a visit: it captures the conversation, it turns that audio into a transcript once the visit ends, and it reuses that transcript to draft the note and suggest codes, diagnoses, and medications. This article walks the whole pipeline so you know exactly what happens when — and where your sign-off is required.
The pipeline at a glance
Record
During the visit
Transcribe
When the visit ends
Generate note
In Sign & Bill
Review & accept
Codes, meds, dx
Sign & bill
You commit it
There are five stages: record during the visit, transcribe once the visit ends, generate the note from that transcript, review and accept the suggested codes and medications, and sign and bill. Each stage hands off to the next, and the AI never skips ahead — the transcript is not generated until the visit is over, and nothing reaches the chart until you sign. If a visit was never recorded, the pipeline simply starts at the note: you write it yourself or draft from the structured chart context alone.
Stage 1 · Capturing the conversation
For telehealth, audio is captured in the visit room from the moment recording is on; for in-person, the phone app captures it. Both require explicit patient consent, and either side can turn recording off mid-visit. Audio is captured per speaker — the provider track and the patient track stay separate — which is what lets the transcript label who said what without a guessing step. If recording is declined or turned off, no transcript is produced and the downstream AI features fall back to the structured chart context.
Stage 2 · Where the transcript is generated
Visit transcript
Diarized by speaker · provider + patient
- Queued
- Preparing
- Ready
The transcript is not produced live — it is generated in the background once the visit is marked complete. The speech model runs over the captured audio and returns a diarized transcript: an ordered list of turns, each tagged with the speaker, a timestamp, and the spoken text. While it runs, any AI feature that needs it shows a clear status — Queued, then Preparing with a rough ETA, then Ready. A visit with no usable audio resolves to "no transcript available" rather than spinning forever, so you always know whether a transcript is coming.
Stage 3 · Turning the transcript into a note
Claim submission note
HPI
Follow-up for generalized anxiety. Reports improved sleep and steadier mood since the last visit; mild morning dry mouth on the current dose…
Two paths consume the prepared transcript. The AI scribe drafts a full structured note into your queue automatically after the visit. And inside Sign & Bill, a Generate with AI button compiles a clean note on demand — it reads the patient’s chart context, the prepared transcript, and any free-text instructions you add to the prompt, then writes the body right there for you to edit. If the transcript is still preparing, the button shows the status and ETA instead of running on incomplete input.
Stage 4 · Codes, diagnoses, and medications you accept
AI suggestions · accept to add
99214
Office visit, established patient
F41.1
Generalized anxiety disorder
Sertraline 50 mg
Take one tablet by mouth daily - draft Rx
Alongside the note body, the model proposes CPT procedure codes, ICD-10 diagnoses, and medications — each clearly marked as Suggested. Nothing is auto-coded, auto-diagnosed, or auto-prescribed: you accept each suggestion, and accepted items are written to the chart only when you sign. Accepted medications become draft prescriptions you complete and send through the normal e-prescribing flow. Every chart change carries provenance back to the note that produced it.
One transcript, reused — never re-recorded
Because the transcript is stored with the visit, both the auto-draft and the on-demand Generate with AI read the same source. Regenerating a note, or generating it weeks later, does not re-run the speech model or ask the patient to repeat anything — it reuses the transcript already on file. That keeps results consistent and avoids burning time (and audio you no longer have) on a second pass.
Privacy, consent, and retention
Audio and transcript are encrypted and stored alongside the visit, governed by the same consent and release rules as the rest of the chart. The models are contractually prohibited from training on your data. Recording is consent-gated and reversible mid-visit, and a declined visit never affects the next one. The AI assists documentation and coding; it never signs, never bills, and never commits a diagnosis or prescription on its own.
Frequently asked questions
Why is the transcript not ready immediately after the visit?
Can I generate the note if there was no recording?
Do accepted codes and medications save immediately?
Keep reading
AI scribe: auto-drafted clinical notes
How the AI scribe listens during in-person and telehealth visits, drafts a structured note, and waits for you to sign — never bypasses you.
AI Frontdesk: an AI receptionist that answers, books, and triages calls
How the AI receptionist answers your phone 24/7, what it can and cannot do, how warm-transfer to live staff works, and how to read the call log.
Signing and co-signing notes
Who can primary-sign, when co-sign is required, and how to delegate signing.