YouTube automatic captions need to be proofread before you use a video as instructions. First of all, check negations, numbers, units of measurement, and names: an error in these places can completely change what the viewer does. Checking is needed both for a long video and for Shorts.
Two passes instead of reading at random
YouTube warns that recognition depends on pronunciation, background noise, and other characteristics of the recording. Captions may appear after processing, rather than immediately upon upload. In YouTube Studio, open the captions section of the desired video and check the available track.
The first pass is checking the meaning against the original audio. For an automatic track, the help documentation indicates Duplicate and edit; after making edits, click Publish and check the already published track. The second pass is watching without sound: are the boundaries of phrases clear, does the viewer have time to read the line, is an important object hidden by text on the screen. Fixing words does not guarantee successful timing.
Error card
A hypothetical line from an instructor: “Do not tighten the screw all the way.” If the “not” is missed, the viewer receives the opposite instruction. Write down the time, the recognized text, the correct version, and the result of the recheck. In exactly the same way, verify “fifteen” and “fifty,” the model name, and its index.
If the speaker themselves misspoke, do not correct only the caption as if the audio were correct. Add an explicit correction or prepare a correct version of the video: viewers with sound and without it should receive the same action.
Readiness boundary
The video is ready on this point when all lines have been checked and the corrected places have been rewatched. The absence of auto-captions is not a reason to publish guesses instead of what was heard: prepare your own verified text in a supported way. Captions do not automatically describe everything that happens silently in the frame; necessary actions need to be voiced or explained separately.
Diagram: 1 — compare the speech and the recognized words; 2 — fix the meaning and the timing; 3 — watch the finished video without sound.
Related task: Shorts without sound and interface cues: checking whether a video is understandable.
Fix a simple file and check the offset
If you have your own SRT or WebVTT track, upload it to the caption editor. First check the text against the audio, then, if there is a consistent mismatch, set a global offset. For example, +1,5 seconds moves the line 00:00:01–00:00:03,5 to 00:00:02,5–00:00:05; its duration remains the same.
A global shift is only suitable when all cues are offset by the same amount. If the error grows toward the end of the video, such a shift does not fix the track. The tool's export keeps plain text and the external boundaries of the cues. A file with tags or internal timecodes is rejected: first prepare plain text without these constructs. WebVTT styling and settings are not preserved. After downloading the SRT or VTT, check the beginning, middle, and end in a player, then check the finished track on the platform.
Sources
YouTube: automatic captions · YouTube: editing captions. Checked on 21 September 2026. Practical examples and diagrams are the editorial methodology; the hypothetical numbers are not client data.
For a specific finished public video, you can check the view terms. This service does not perform the work described and does not guarantee organic interest, leads, or sales.