Why this usually goes wrong
Three failures stack, and each one hides the one before it.
- Coverage. Transcript services index what is popular, recent and English. A 2019 lecture in Spanish with 30,000 views is exactly the kind of video they never crawled — so the tool reports no transcript, though YouTube has one.
- Double translation. Some tools read YouTube’s auto-translated captions rather than the original, then translate again into your language. Two machine passes over the same sentence, and the errors compound.
- Re-transcription. Tools that run their own speech recognition are measurably weaker outside English. You cannot see that happening, because the output is fluent in your language either way.
The common thread: you end up reading a translation of a translation of a guess, with nothing to check it against.
Ask for the original line alongside the translation when a claim matters. “Quote the original sentence and then translate it” gives you both, with the timestamp — and the original is the thing that is actually verifiable.
Prompts that work across languages
- Both versions. “Summarize this in English, and for each main point quote the original sentence with its timestamp.”
- Terminology. “List the technical terms used, in the original and in English, so I can recognise them later.”
- Check a claim. “At 14:20 — what exactly was said, literally, and what does it mean?”
- Register. “Is the speaker being formal, casual or sarcastic here? That does not come through in a plain translation.”
- Your own language. Ask in Spanish, Japanese, Portuguese or anything else — the answer comes back in the language you asked in.
The honest limit
If YouTube has no captions in any language for that video, there is nothing to read and the plugin will say so instead of producing something. Where the only captions are machine-generated, expect proper nouns and jargon to be unreliable — which is exactly when asking for the original line next to the translation is worth the extra seconds.
Related: what ChatGPT can and cannot do with a video, and turning a lecture into study notes.