How to Fix Auto-Generated Captions Before Publishing Shorts

1. Check the words a spellchecker cannot decide for you

Start with people, organizations, product names, specialist terms, numbers, and negations. A plausible word can still identify the wrong thing. An omitted “not” can reverse a claim while leaving a perfectly grammatical sentence.

In our example from The Build Log, guest Colin Sidoti discusses his company, Clerk, and later mentions React. The original transcript lowercases both. Listening helps establish what is said; the recording’s title and the surrounding technical discussion help establish the intended proper names. Do not replace an unfamiliar term solely because a more familiar one looks right.

For your own episode, keep a short reference list supplied by the host or guest: names, brands, and recurring terms. Verify each relevant occurrence. Search for repeated occurrences so a name stays consistent throughout the clip.

2. Edit the captions that appear in the video

Open the candidate in Edit clip. Open the Captions task. Its Transcript tab shows read-only source context; Edit captions contains the text shown in the exported video. This distinction matters when the source transcript still displays an old spelling after you have corrected the captions.

1. Settle the clip’s range first, so you are reviewing the words that will actually be included.
2. Select Edit captions and find the line containing the term. Use Find in this video… to help locate text, then listen to the relevant passage.
3. Correct the text in that caption line. Keep the speaker’s meaning and level of certainty.
4. Play the line with the neighboring lines before you save.

Preserve phrases that qualify the answer: “an initial question” means something different from “the only question.” A post title can summarize an idea; captions represent the speech. If a word is genuinely unclear, return to the audio or ask the speaker instead of silently guessing.

3. Use Professional terms for a correction; Account glossary for recurring vocabulary

A single typo can be edited directly in its caption line. When a name or technical term repeats, open Professional terms in the caption toolbar. Enter the observed spelling in Find text and the verified spelling in Replacement, then choose Preview suggestions. For a caption that still contains the original lowercase term, an example pair is clerk → Clerk.

Read the Original and Suggestion differences. Select only the changes that fit the spoken context, then use Apply selected · Don’t save. That updates the unsaved draft. Finish by choosing Save changes and refreshing the final preview. If the caption is already corrected, there may be no matching change to apply. Professional terms finds matching text, so review each suggestion in context before applying it.

For vocabulary that recurs across recordings, Account glossary stores incorrect and preferred spellings for reuse. Choose Add term, enter the pair, and set the relevant category, video language, and case-matching option. Save the glossary, then use Preview for this video to inspect matching suggestions for the captions you are working on. The glossary is available with Standard.

Saving the glossary does not rewrite existing captions. Its matches generate reviewable suggestions, which still need to be selected, applied to the draft, and saved as caption changes. This is useful for recurring guest names, brands, and technical vocabulary without treating every occurrence as interchangeable. For example, an ordinary use of “react” should not automatically become the product name React.

For a recurring show, keep verified guest names, company names, and specialist vocabulary in the glossary. Use Professional terms for a correction specific to the current video. Both workflows let you review the matching lines before changing them.

4. Separate source time from caption time

The source range begins at 07:15.1 in the interview. In the edited clip, the line about Clerk begins at 00:12.570 and ends at 00:16.130. In this example, caption timestamps count from the selected clip’s start. With confirmed cuts or an independent intro, use the finished timeline shown by the editor. Use source time to locate the passage in the interview and the indicated caption time to align its text; an imported SRT also needs the correct time reference.

Use the line’s start and end fields or the small timing controls to align the text with the speech. Moving an entire line and moving only one boundary solve different problems: the first shifts the line; the second changes how long it remains visible. Listen around the adjustment so you do not create a collision with the neighboring caption.

Do not use a timing change to conceal missing words. To divide a long caption, pause at a natural phrase boundary and choose Split at playhead. Check the words and start and end times in both new captions, then replay the transition at normal speed. Keep names and closely related words together, and confirm that every spoken word is still represented.

5. Import a corrected SRT without losing your edits

If you or your editor have already corrected the captions in an SRT file, open the SRT control in Edit captions. You can also attach an optional corrected SRT when submitting a source. Attaching a file prepares it for matching; it does not replace the transcript or automatically change every Clip.

Choose the file’s time reference carefully. Full original video / audio uses absolute times from the original recording, even when only a later part of that recording was processed. Current finished Clip uses times from the saved finished Clip, including its confirmed cuts and independent intro. That file belongs to that Clip and saved version; it cannot be reused as another Clip’s timeline.

For example, a whole-source cue at 03:10–03:12 appears at 00:10–00:12 in a Clip beginning at 03:00 when there are no cuts or intro. Do not subtract the Clip start from a whole-source file yourself and then label it as whole-source time. The matching preview handles the mapping. If a cue crosses a cut, the Clip boundary or a region without source mapping, inspect the flagged issue before adopting it.

1. Choose the time reference and the file’s language, then read the SRT. Use UTF-8 or UTF-8 BOM encoding and a valid .srt file no larger than 2 MiB.
2. Review the matched lines, differences, uncovered intervals and original versus finished times. Temporarily preview the candidate to check where its captions appear.
3. Resolve boundary warnings. If the file leaves gaps, decide explicitly whether those gaps should remain blank; missing text is not silently filled from an older draft.
4. On Lite or Standard, explicitly adopt the candidate into the draft. Edit or undo it as needed, then choose Save changes through the normal Clip controls.
5. Generate the finished preview and check the actual export, particularly around cuts and the opening caption.

Importing and matching your own file is available without an account and on every plan. Adopting, editing and saving imported captions requires Lite or Standard. Standard can apply and save a whole-source candidate to explicitly selected safe Clips from that same source. Targets start unselected; protected edits, uncertain history and items needing review are excluded from the batch. This operation does not batch-render or export the Clips.

Existing manual captions, previously imported text and unknown editing history are protected. Review the full differences and explicitly confirm replacement when appropriate. Handle any unsaved Clip draft first. Updating the shared whole-source SRT does not rewrite previously saved Clip captions; those keep their saved snapshot until you choose another candidate.

A whole-source file can be matched to other Clips from the same recording without uploading it again. Current-clip files cannot. SRT import changes captions only: the trusted source transcript, original-transcript search, AI recommendations and AI Writer evidence remain unchanged. It uses no processing minutes or writing rounds. System translation and export permissions remain separate; bringing your own corrected text does not unlock those actions.

If parsing fails, fix the indicated entry or encoding and retry; the previous candidate and saved captions remain. If nothing matches, check the selected Clip and time reference before changing plans. If another window saves a newer version, compare again rather than overwriting its edits.

6. Check captions in the frame you will publish

Read the clip once with sound off. Can you follow the thought at normal speed? Are names readable? Does the caption cover the speaker’s mouth or a relevant screen? A long host question may need more attention than the short answer, even when the wording is correct.

Open Caption settings to adjust caption visibility, phrase segmentation, line limits, speaker labels and appearance for this clip. Check the longest line and each speaker change in the actual output ratio. A line that is comfortable in landscape can wrap into an awkward block in 9:16. Recheck it after changing the framing.

See the caption settings

Use Caption schemes for a reusable style. Compare the current and candidate previews, adjust the candidate if needed, then apply it to the draft. Saving a personal scheme does not save this clip’s edits. Schemes do not replace the caption text; language and translation remain in Edit captions with their own plan access. Keep any translated or bilingual captions aligned with the speech when reviewing line breaks.

Use any caption-quality flags to find lines that need attention, then read and listen through the whole selection. Pay particular attention to unfamiliar names and dense explanations.

7. Save, refresh, and read the rendered captions

Choose Save changes, then Generate clip preview in the preview header. After the preview is ready, review the clip and use Download MP4. Saving text and rendering text into the final video are separate steps; an instant preview is not the downloaded file.

Open the downloaded MP4 and check the corrected terms, the first caption, the final caption, and any dense line. If you change the wording or timing again, save and refresh again. Caption editing is available with Lite and Standard; consult the plan comparison for current access.