Last updated 18 September 2026
AI Scribe listens to a consultation and drafts the clinical note. This page explains, in plain terms, what it records, where the audio and words go, who processes them, what Zavy keeps, and what your practice and your patients can control. It is written so a clinician can show it to a patient.
AI Scribe has two transcription options. Your practice chooses which one clinicians use. The steps below are the same for both except for how the speech is turned into text.
The audio is streamed to a speech-to-text provider in the United States. This option gives the most accurate transcript for clinical vocabulary and multiple speakers.
Consent, then record
Consulting room
The clinician confirms everyone present has agreed. The microphone captures the consultation.
Audio streamed for transcription
Speech-to-text provider
The audio is sent live, through Zavy's servers, to a contracted speech-to-text provider in the United States, which returns the words.
Transcript sent for drafting
Language model providers
The transcript, as spoken, plus basic context from the record (not the patient's name) goes to one or more contracted language model providers in the United States. They return draft notes.
Clinician reviews the draft
Zavy360 in the browser
The clinician reads the draft, picks between alternatives, edits, and decides whether to save. Nothing is saved without this step.
Saved to the record
Zavy-controlled storage, Sydney
The approved note is saved to the patient's record. The transcript and edits are kept by Zavy in Sydney. Audio is kept only if your practice has turned that on.
Speech is turned into text on the clinician's own computer. The audio is not uploaded for transcription. The transcript still goes to the US language model providers to write the draft, so this option is not entirely local.
Consent, then record
Consulting room
The clinician confirms everyone present has agreed. The microphone captures the consultation.
Transcribed on the computer
Clinician's own machine
A speech recognition model runs in the browser and turns the audio into text without uploading it. The model itself is downloaded once from a public model repository.
Transcript sent for drafting
Language model providers
The transcript, as spoken, plus basic context from the record (not the patient's name) goes to one or more contracted language model providers in the United States. They return draft notes.
Clinician reviews the draft
Zavy360 in the browser
The clinician reads the draft, picks between alternatives, edits, and decides whether to save. Nothing is saved without this step.
Saved to the record
Zavy-controlled storage, Sydney
The approved note is saved to the patient's record. The transcript and edits are kept by Zavy in Sydney. Audio is kept only if your practice has turned that on.
| Information | What it is | Where it goes | What Zavy keeps |
|---|---|---|---|
| Consultation audio | The recording of the appointment. | Cloud option: streamed to a speech-to-text provider in the US. On-device option: stays on the clinician's computer for transcription. | Stored in Sydney and attached to the patient's record, as part of the health record. |
| Transcript | The words spoken, as recognised by the speech-to-text step, with speaker labels. | Sent to language model providers in the US to write the draft. | Kept by Zavy in Sydney with the session, as part of the health record, so the note can be regenerated, reviewed or audited. |
| Draft and saved notes | The AI drafts, the clinician's edits, and the note the clinician saved. | Drafts come back from the US providers. The saved note lives in Zavy360. | The saved note is part of the clinical record and follows your practice's record-keeping obligations. Drafts and edits are kept in Sydney for quality review. |
| Consent clip | An optional recording of up to ten seconds of the patient agreeing to be recorded. | Uploaded directly to Zavy storage in Sydney. Never transcribed, never sent to an AI provider. | Kept as evidence of consent until your practice asks for it to be removed. |
| Consent log | Who confirmed consent, for which patient, when, and what they confirmed. No transcript content. | Zavy storage in Sydney. | Kept while your account is active as a record of consent decisions. |
| Patient context | Age, alerts, existing conditions and treatment plan items from the record. Not the patient's name. | Sent with the transcript to the language model providers so the draft fits the patient. | Not stored separately; it is already part of the patient's record. |
Zavy uses contracted providers to deliver AI Scribe. We describe them by role and country rather than by name, because the mix of providers is part of how the product works and changes as we improve it. Your practice can request the current list of providers from info@zavy.com.
| Who | What they receive | Country | Why |
|---|---|---|---|
| Speech-to-text provider (cloud option only) | Consultation audio, live | United States | Turns speech into text with dental vocabulary and speaker labels. |
| Language model providers (one or more) | Transcript as spoken, patient context without the name, the note template, and reference examples | United States | Write the draft note. Zavy may ask more than one and show alternatives. |
| Zavy-controlled database and storage | Transcripts, drafts, edits, saved notes, consent clips, audio and logs | Australia (Sydney) | Runs the feature, keeps your records, supports quality review. |
| Zavy processing servers | Audio and transcripts in transit | Australia and Singapore | Relay audio to the speech-to-text provider and run the drafting pipeline. Being moved to Sydney only. |
Our providers keep audio and text only for as long as they need to process it and to monitor for misuse of their service, under their contracts with Zavy. They are not permitted to use it to train their models. Zavy does not control when a provider clears its short-term copy.
Before recording a patient whose consent is not already on file, AI Scribe asks the clinician to confirm that consent was obtained, and logs the answer. The clinician can also capture a short clip of the patient agreeing. If anyone says no, the clinician takes notes the usual way.
A patient can ask your practice not to use AI Scribe for them. Staff record that on the patient's profile in Zavy360 and AI Scribe checks it before recording. Opting out stops future use; it does not remove notes already in the clinical record.
Your practice can use the on-device option so audio never leaves the computer for transcription, or the cloud option for higher accuracy. The transcript goes to US language model providers in both cases.
AI Scribe records are part of the patient's health record, so Zavy does not delete them on request from a practice or a patient. They may be deleted only after the practice's account is terminated and the data export period has ended. Patients who want access to their record, or a correction, make that request to the practice, which holds the record.
No draft becomes part of the record until a clinician has read it and chosen to save. Drafts can be edited freely or thrown away.
Questions about how AI Scribe handles data? Contact info@zavy.com.
The binding rules are in the AI Scribe Notes Supplemental Terms, the Terms of Service and the Privacy Policy.
