A failed transcript request does not tell you whether a video contains speech. First identify what you requested: an existing caption track, a translation of that track, or new speech recognition from audio. These are different operations with different failure modes.
Check the language and track
Start by checking which caption tracks are available for the video and which language your request selects. A video may have captions without having the requested language. Human-created and automatically generated tracks also need to be distinguished; they are not interchangeable quality labels.
The youtube-transcript-api repository documents listing transcripts, language preferences and track types. Follow the API for the installed version rather than copying an old tutorial. Do not silently substitute another language: record the requested language and the returned language separately so downstream search or summarization knows what it received.
Separate absence from access failure
If a track is unavailable, that is a content-availability outcome. If a request is blocked or times out, it is an access or transport outcome. Returning the same empty string for both loses the information needed to recover.
The repository describes RequestBlocked and IpBlocked errors, including problems with cloud IP addresses. A script working on a laptop therefore does not establish that an identical deployment will work from a server. Conversely, an access error is not evidence that the video has no captions. Preserve the error category and the environment where it occurred; avoid endless retries that repeat the same failure and consume resources.
Understand the official download route
The YouTube captions.download documentation specifies authorization requirements. Public playback does not itself establish permission to download that video’s captions through the official API. Check the account’s actual access and the documented operation before treating an authorization failure as a library bug.
Choose the next step explicitly
If the requested language is missing, offer available languages or an explicitly identified translation workflow. If captions are disabled or absent, speech-to-text may be a separate option where audio access and processing are appropriate. That introduces additional cost and quality questions; it is not simply another caption-download retry.
If access is blocked, report the access failure and investigate the deployment or supported retrieval method. If only part of a batch succeeds, preserve successful items and identify failed items individually. Never present an empty transcript as a complete, successful result.
For a production integration, a useful result record should explain the operation, requested and returned languages, track provenance, outcome and any recoverable error. These are design considerations, not a claim about an available TrueFetch endpoint. TrueFetch currently publishes product previews; consult the catalog’s release status before planning a live integration.