1. Check the video before changing the workflow
For a one-off lookup, you may not need an API. Open the video description and look for Show transcript. YouTube documents this option for videos that have captions, with transcript lines that link back to the corresponding moment. YouTube's transcript instructions.
If the video is newly uploaded, captions may still be processing. YouTube also lists unsupported language, poor audio, overlapping speech and other audio conditions among reasons automatic captions may not appear. Waiting is a reasonable option when you control the video and the task is not urgent. YouTube's automatic-caption guidance.
If captions work in the player but your integration fails, keep the error and check the integration separately. Do not turn an access or software error into a conclusion that the creator supplied no captions. For an open-source library, check its current API and installed version before copying an older code example. youtube-transcript-api documentation.
2. Pick a route, rather than retrying everything
| What you found | Useful next step | What not to assume |
|---|---|---|
| A usable caption track exists | Retrieve it and inspect the returned text and language. | Existing captions are necessarily accurate. |
| Captions exist, but not in your preferred language | Decide whether another available language fits the task. Review the returned language. | A language preference guarantees the requested language. |
| There are no usable captions, but permitted audio is accessible | Consider AI transcription with an explicit duration limit and budget. | Speech-to-text will recover every word or fix inaccessible media. |
| The video or request is unavailable, restricted or failed | Save the diagnostic, investigate the specific cause, or exclude that source. | Repeating the request will create a transcript. |
On a narrow screen, scroll the table sideways to read all columns.
These are workflow decisions, not a list of guaranteed API error codes. Different services report failures differently. Keep the original status and message rather than rewriting every failure as “no captions.”
3. Start with an explicit captions-only input
TubeText's YouTube Transcript Actor retrieves caption text and offers optional AI transcription when captions are absent. Set aiFallback explicitly so a future reader of your workflow can see which behavior you chose.
Configuration example, not an executed request. Replace the placeholder with a supported YouTube video URL you are permitted to process:
{
"urls": ["YOUR_YOUTUBE_VIDEO_URL"],
"languages": ["en"],
"preferManualCaptions": true,
"aiFallback": false,
"includeSegments": true,
"includeTimestampedText": true
}This disables the Actor's optional audio-transcription fallback. It does not make the Actor run free. The language list is a preference order: the documented behavior can return the original automatic track when none of your preferred caption languages is available. Inspect language instead of labelling every output English. Current input reference.
Use this route when your project needs existing source captions, when you want to review missing-caption cases before spending on transcription, or when generated text would not meet your evidence requirements.
4. Enable AI fallback deliberately
If your task permits generated transcripts, the alternative input below allows AI transcription for captionless videos within the chosen limit:
{
"urls": ["YOUR_YOUTUBE_VIDEO_URL"],
"languages": ["en"],
"preferManualCaptions": true,
"aiFallback": true,
"maxAiMinutesPerVideo": 15,
"includeSegments": true,
"includeTimestampedText": true
}The current documented defaults are AI fallback on and a 15-minute AI limit, checked on 23 October 2026. Both are written explicitly here. This is a fallback when captions are absent, not an instruction to replace every caption track with AI text. YouTube Transcript documentation.
maxAiMinutesPerVideo is an eligibility limit: longer videos are skipped by AI transcription. It does not mean “transcribe the first 15 minutes,” and it is not a total-spend limit for a multi-video run. Review Apify's run settings and the Actor's current pricing events before execution. Transcript delivery, started AI minutes and Actor Start are separate billing considerations; a waived result fee does not imply a zero-cost run.
AI fallback does not grant access to private or restricted videos. Process only material you have permission to use, and review the resulting text against the audio before quoting or publishing it.
5. Review the result before sending it onward
Use the documented output fields to make a small triage record:
- Identity: retain
videoIdandurl, plus the run and dataset identifiers from your workflow. - Outcome: inspect
statusanderror. Do not treat an error message as transcript text. - Content: require a nonempty
transcriptbefore summarizing or indexing it. - Provenance: retain
sourcewhen present. The documented values distinguish retrieved captions from AI transcription. An unknown source should remain unknown. - Language and timing: review
languageand keep requested timestamp data with the original export.
Download the blank caption-triage CSV. It is a suggested review worksheet, not an Actor output file. Its review_decision, review_notes and retrieval-date columns are for your workflow to fill in.
A useful downstream rule is to send only a successful, nonempty transcript with no reported error to a content destination. Send everything else to a separate review queue. Then apply any stricter language or provenance requirements your project has. For example, a caption-only corpus should not quietly accept an AI-generated replacement.
If a client request times out, inspect the original Apify run and dataset before launching another run. A missing response does not prove that processing stopped. If the text already exists but your database write failed, recover from the retained export instead. The incremental RAG and export-recovery guide covers that separate problem.
Common questions
Can I get a transcript when YouTube has no captions?
Potentially, by transcribing accessible audio rather than retrieving a caption track. The audio must be available, the input must be supported, and your chosen limits must allow it. Music-only, silent, restricted or failed inputs can still leave you without usable speech text.
Will AI fallback translate a video into English?
Do not treat it as a guaranteed translation step. Check the returned language and text. Translation, if your project needs it, should be an explicit, separately reviewed downstream operation rather than an assumption based on languages: ["en"].
Do timestamps prove that a transcript is accurate?
No. They help you locate the source passage. Verify names, numbers and quotations against the video, especially when the source captions or AI transcription are uncertain.
Open the tool with your policy decided
Review YouTube Transcript on Apify, choose captions-only or optional fallback, and begin with a small permitted input. Keep the diagnostic rows as well as the text.
Need visual information rather than speech? The video-to-shot-list guide explains how to retain and review a visual breakdown.
TubeText Labs maintains the Actor linked in this guide. This page was prepared with AI assistance from the linked documentation; the configurations and worksheet are not a live test or a coverage guarantee. Product execution and billing happen in your Apify account. About TubeText.