An automatic transcript can help you find content and prepare a summary. The text requires verification when used for decisions or communications. Names, numbers, and attribution of sentences may be relevant; a readable form does not prove that the conversation was correctly represented.
I propose to distinguish transcription, correction and synthesis. Each step changes the relationship with the initial material. Content recording and processing must follow appropriate conditions and procedures; not all conversations can be processed in the same way or by any means.
Let us imagine a discussion in which a date is proposed and then another is agreed upon. An incomplete synthesis could retain only the first one. Comparison with available material and relevant participants helps to verify what will be used as an outcome.
The review must focus on the information that supports the next work. NIST offers a voluntary framework to manage AI risks; I consider outcome verification to be an essential operational step. Transcription remains useful when its degree of reliability is consistent with intended use.