I corrected the product model in Descript’s script; why does the narrator still say the old model?

Identify whether the error is spoken or transcribed

A merchant reviewing a product demonstration notices that the script displays the wrong model number. Before editing, compare the actual spoken words with the approved product information.

There are two fundamentally different situations. Automatic transcription may have misheard a correct recording, or the narrator may genuinely have spoken the wrong model.

Correct addresses the first situation. It changes the written transcript without altering the audio or video. Correct All applies the correction to matching instances across the project, so its scope deserves attention when similar model names appear in several scenes.

Neither operation rerecords a sentence. A transcript that displays the approved model number while the audio says something else creates a misleading disagreement between text and sound.

Choose among correction, media editing, and hiding

Descript's document-style editing connects ordinary script changes to underlying media. Deleting spoken words through the regular editing workflow can remove the corresponding audio and video. Cutting and pasting words can also move media.

Remove from transcript serves another purpose: it hides the selected transcript content while retaining the corresponding media. This affects script and caption presentation rather than removing the spoken statement.

For a transcription error, use Correct. For genuinely unwanted recorded material, assess whether a media edit is intended. For content that should remain audible but not appear in transcript presentation, evaluate the hiding operation carefully.

Hiding essential product information from captions may create an incomplete experience for viewers who depend on text. It should not be used to disguise an incorrect spoken claim.

The key decision is which representation must change: displayed text, recorded media, or transcript visibility.

Illustrative exercise: two model-number mistakes

Illustrative exercise — fictional merchant scenarios. No Descript operation, audio edit, video edit, or AI generation was executed.

Input A is a coffee-machine demonstration. The narrator clearly says model K3, but automatic transcription displays K8. The approved product record also identifies K3.

The appropriate branch is transcript correction. If the error appears repeatedly, inspect each intended occurrence before considering project-wide Correct All.

Input B is a different advertisement where the narrator actually says K8, although the product being advertised is K3.

Correcting the transcript alone would leave the incorrect spoken statement. The merchant needs an approved replacement recording or another appropriate speech-production workflow, followed by a review of the revised audio and corresponding text.

The two cases look similar in the script but require different production decisions.

Review what viewers will hear and read

For every disputed model number, compare the approved product record, audible narration, visible captions, and exported transcript.

If the recording is correct, verify that transcript corrections do not unintentionally change other product identifiers through a project-wide replacement.

If the recording is wrong, confirm that the approved correction changes the actual spoken content rather than merely presenting different words on screen.

If transcript text is intentionally hidden while speech remains, review whether viewers still receive the necessary information through an appropriate accessible presentation.

The handoff is complete when the team can explain which representation was corrected and why. A visually correct script alone is not evidence of corrected narration.

Illustrative exercise — Product-model correction decision

These are illustrative exercise rows. Save a copy and record your own observations; the example is not an executed acceptance test.

Download review worksheet (CSV)