AI scribes should be free

Doctors should get basic AI scribing for free. Vendors should charge only for optional EHR integration and the work to keep that connection reliable. Local transcription, drafting, clinician review and a usable manual copy or export route belong in the free utility.

A basic scribe turns an encounter recording or dictated account into a transcript and a draft note. Its free software should include the screen where the clinician checks the transcript, edits the draft and decides what belongs in the chart.

A free draft already exists

Heidi's current free tier includes unlimited transcription and notes using standard templates, with limits on advanced features. Doximity offers its scribe free to U.S. physicians, nurse practitioners, physician assistants and medical students with verified accounts. These are vendor offers, with eligibility rules and terms that can change. They establish that a zero software price exists, not where processing happens. A free offer can be subsidized and does not establish the economics or quality of an on-device scribe.

We read those offers as evidence of price pressure on basic drafting. Vendors should explain what a practice buys beyond the first draft. We would draw the line at the EHR connection. The basic utility remains free whether or not the practice buys it.

There is also a route to running the components on practice hardware. Google's developer documentation describes running medical speech recognition locally and deploying its health AI building blocks on local servers or devices. This establishes a technical route, not a finished scribe for every practice. A team still has to test the actual hardware, languages, room noise and encounter lengths. On-device processing does not promise the same quality as every cloud service.

Keep the conversation at the practice

A recording contains more than the final note. The transcript and unfinished draft are patient information too. Each external scribe processor adds another destination for that material. Our proposed business model starts by keeping it at the practice.

The core scribe should run only on practice hardware. It must not send patient audio, transcripts or draft notes to an external processor, including through a silent cloud fallback. Only the clinician-reviewed note is exported to the approved EHR, manually or through an optional connector. That EHR may have its own cloud infrastructure and contracts. Paying for a connector does not change the boundary around the raw encounter.

HIPAA does not require this architecture. HHS permits cloud processing of electronic protected health information with an appropriate business associate agreement and compliance with the applicable HIPAA requirements. A well-run cloud service can be a defensible choice. Practice hardware does not create compliance automatically. We prefer it for basic scribing because it removes a routine external processing route when that route is unnecessary.

The practice still needs security controls and maintenance. Encrypt device storage and transfers on the practice network, restrict access, install security updates and set retention rules for recordings and drafts. Keep a protected review record that identifies the clinician, the draft version and the export. Keep diagnostic logs free of patient content. Check that crash reports, support tools and automatic backups do not quietly upload it. Test recovery. A local drafting screen tells you little about these other routes.

Charge for the EHR connection

Free software still has operating costs. Hardware, electricity, installation and internal maintenance cost the practice money. Those costs belong in its budget. They should not become separate vendor charges for basic deployment, general support or access to the copy button. The vendor charge should cover only the optional EHR integration and its continuing upkeep.

A paid connector should confirm the correct patient and encounter, show the approved write destination and respect the permissions granted by the practice. It should preserve an audit trail linking the clinician-reviewed note to the attempted write and its result. A rejected write should be visible to the responsible person. The vendor earns the integration charge by making those steps inspectable and keeping them reliable.

EHR capabilities differ. The FHIR specification describes capabilities for each implementation, while SMART's authorization rules make clear that available patient and encounter context can vary and write requests remain subject to underlying permissions. A standards label does not prove that a particular note can be written into a practice's chosen destination. Verify that destination before selling the connection.

The federal SAFER guidance treats patient identification and interface maintenance as operating work. The integration charge should include monitoring failures, adapting to interface changes, retesting permissions and support specific to that connector. Name the owner who responds when it breaks. State unsupported EHRs and missing capabilities plainly. Do not sell a connection that has not been verified.

Keep a free route into the chart

The free utility must expose the reviewed note in a usable format for the practice's approved manual copy or export workflow. The clinician still confirms the patient, encounter and destination. Some practices may choose that route indefinitely. The vendor should compete on the value of the connector, without making manual transfer deliberately difficult.

The clinician's part remains substantial. The practice explains recording and obtains patient consent as required for its jurisdiction and workflow, with another documentation method when recording is declined. The AMA's ethics toolkit treats disclosure and consent as decisions tied to the use and risk. Its AI principles call for physician consent and final review of generated records. Every draft needs a clinician to check omissions, additions and meaning before it becomes the signed record. A subscription or a device cannot take that responsibility.

This model depends on enough practices finding the maintained connection worth paying for. It does not erase development costs. Test the choice against the complete job, including consent, setup, review, correction and transfer. Compare the manual route with the proposed integration. Record meaningful omissions, unsupported additions and failed transfers. Neither free drafting nor a paid connector guarantees time savings.

Keep the local scribe useful without a subscription. Charge for the optional EHR connection only when the practice can verify what it does and who keeps it working.