What Is a Medical Transcription Service?

A clinician dictates - into a phone, a recorder, or a browser - and the audio goes to the service. Automated recognition produces a first draft. Depending on what you bought, a human transcriptionist then checks that draft against the audio, correcting what the model got wrong, before the document is returned. The output is a document rather than a wall of text: structured into the sections the document type calls for, with the speaker attributed where the recording has more than one voice.

Turnaround is normally tiered. Same-day costs more than a four-day turnaround because it reserves capacity, and the practice chooses per job. Pricing is almost always per minute of recording rather than per page, because a minute of audio is a measurable thing and a page is not.

It is not a scribe. A medical scribe is present during the encounter, works in the chart in real time, and is making documentation decisions as the visit happens. A transcription service works from a recording after the fact and documents what was said. The two are often compared as alternatives, and for some practices they are, but they are different jobs with different costs and different failure modes.

It is also not a dictation app. Speech recognition software on a phone produces text immediately and nobody checks it. That is a perfectly reasonable tool for a note you will read and edit yourself; it is a different proposition from documentation that enters a patient record.

Ask what happens to the document after it comes back. If it is a reference for your own use, unreviewed output is probably fine and paying for review is paying for nothing. If it is going into a chart, onto a referral, or in front of a payer, then whether a person checked it is the whole question - and so is whether the service can tell you, afterwards, that one did.

That second part is more important than it sounds. CMS audit guidance treats a note signed seconds after an encounter closed as evidence that no review took place, and the only affirmative defence against a negligent-documentation claim involving AI-generated content is a contemporaneous log proving the clinician reviewed the specific content before signing.

Verification is a per-job choice with a stated rate difference, decided at upload rather than per account. A clinic can send a clinical dictation through human review and a staff meeting through AI-only on the same account, and the finished document records which mode produced it.

We also fill forms rather than only producing transcripts. Upload a blank form - an intake sheet, a referral, a state document - dictate the answers, and we write them into that same document in its own layout. For the large amount of clinical administration that is "complete this form", a transcript is an intermediate step and the filled form is the thing you actually needed.

Every approval is recorded over a SHA-256 digest of exactly the content the signer saw, with the elapsed time between the transcript becoming available and the signature. That record is exportable and independently verifiable.

Medical transcription is priced per minute of recording almost everywhere, and that is the right unit because it is the only one both sides can measure before the work starts. Per-page pricing depends on formatting decisions the provider makes after the fact, and per-word pricing quietly rewards a provider for verbosity. A minute of audio is a minute of audio.

Published rates for human-verified medical work generally run from about $1.10 to $3.25 per minute, with specialist work quoted higher. Unreviewed automated output is an order of magnitude cheaper, because nobody is paying for a reviewer. Turnaround is the other variable: same-day costs more than a four-day turnaround because it reserves capacity that might otherwise go unused.

Two things to check in any quote. First, whether there is an order minimum, because a five-minute dictation billed against a $25 floor is not costing you the per-minute rate. Second, how difficult audio is handled - most providers charge more for heavy background noise, crosstalk or strong accents, and a provider that does not say so in advance will say so on the invoice.

The delivery mechanism matters more than it sounds, because a document you have to re-key into your record system has only moved the typing rather than removed it. Most services deliver a file by download or email. Some offer a webhook or an API so the document arrives in your own system without anyone handling it.

The useful question is what happens at the boundary with your record system. A provider that emits a structured document, rather than a block of prose, is a provider whose output can be filed or imported rather than retyped. That is also where FHIR matters: a signed document represented as a FHIR DocumentReference carries its own metadata, which a PDF does not.