What should a dictation software disclaimer say?

A dictation disclaimer should tell readers three things: that the text may contain errors and must be reviewed, that accuracy depends on the microphone and the speaker, and where the audio and transcripts go. Each point rests on evidence below. This is an explanation of what the sources support, not legal advice. If a disclaimer will be published, have a lawyer in your jurisdiction review the wording. The overview of dictation software and its guide to dictation software types explain how each kind of product handles audio.

Say that the text can contain errors

Recognition produces drafts with errors. In a 2018 study of clinical notes, 96.3 percent of the speech recognition drafts contained at least one error ¹. The same study found an error rate of 7.4 errors per 100 words in those drafts ². A disclaimer should say that the output is a draft produced by speech recognition, and that errors are possible. The question on what voice recognition software is used for shows which uses depend on this accuracy.

Say that review reduces errors

Review is the step that makes dictated text usable. Review and editing cut the error rate of dictated clinical notes from 7.4 to 0.3 errors per 100 words, about 25 times fewer ³. A disclaimer can tell the reader that the text should be checked by a person before it is relied on, and name who is responsible for that check.

Say that accuracy depends on the microphone and the speaker

Conditions change the error rate. In 2017 figures from Ravanelli and colleagues, a single distant microphone gave a word error rate roughly 3.2 times the close-talking rate ⁴. The two conditions were not identical, so treat this as a rough ratio. Accuracy also varied by speaker: across five commercial systems in a 2020 study, the average word error rate was 0.35 for Black speakers and 0.19 for white speakers ⁵. A disclaimer should say that accuracy varies with recording conditions and speakers, and that the software was not tested for every case.

Say where the audio goes

Privacy terms differ between products, so the disclaimer should describe the actual setting. Apple says dictation sent to its servers is not stored unless you opt in to Improve Siri and Dictation ⁶. Wispr Flow says it deletes your audio as soon as it has been transcribed, unless you turn on its cloud storage setting ⁷. Aqua Voice says it may store transcript data on its servers when Privacy Mode is off ⁸. A disclaimer should name the setting that applies to the product in use, and not generalize from another product. If a product also builds a voice profile, that is a separate question, covered in the question on voice recognition and speaker recognition.

A sample disclaimer

The wording below is a starting point built from the points above, and it needs the review described in the previous section:

This text was produced by speech recognition software and may contain errors, including words that were misheard or missing. It was reviewed by [name or role] on [date] before use. Accuracy depends on the microphone, the recording conditions and the speaker. Audio was processed [on the device / on the provider's servers] and stored [for how long, or not at all].

Where dictated text informs medical or legal decisions, the disclaimer should say that the text is not a substitute for the judgment of the qualified professional who reviews it.