“Transcript not available” on YouTube: why it happens and four ways round it

Through 2025 and 2026 YouTube tightened access to its caption endpoints. A caption URL that used to return timed text now frequently returns HTTP 200 with an empty body unless the request carries a token that the YouTube web player generates in the browser. No token, no error message — just nothing.

This is why a transcript site that worked last year now shows a spinner and then a vague failure. It is also why the same tool works on some videos and not others: several different client endpoints exist, they are gated inconsistently, and which one succeeds depends on the video.

Worth saying plainly, since this site runs one of those tools: our YouTube transcript tool tries the routes that are not token-gated and tells you honestly when a video cannot be read, rather than showing a generic error. When it cannot, the audio route below always works.

If "Show transcript" appears under the video, you already have the text.

Open it, click the three-dot menu inside the transcript panel, and turn off timestamps. Then select the whole panel and copy. You get a plain block of text, no timestamps, no punctuation improvements, and no speaker labels.

This is free, immediate and limited: it is the automatic caption track, which means run-on text with no sentence boundaries on most videos. For reading, fine. For quoting or subtitling, you will want punctuation, which means one of the next two routes.

A transcript tool that reads the caption track can do two things the panel cannot: restore punctuation and sentence breaks on an auto-generated track, and give you the text in a file rather than a selection.

That is worth the extra step for anything you intend to edit — a summary, a blog post, subtitles for a re-upload, a set of study notes. Auto-captions arrive as a stream of words; punctuating them is the difference between a transcript and a wall.

It still depends on captions existing and being reachable, so it fails on the first, second and fourth cases from the list above.

Captions are a convenience. The audio is the source, and transcribing it does not depend on YouTube's endpoints at all.

If the video is yours, or you have it as a file, upload it to the transcription tool and you get a timestamped transcript with optional speaker labels, regardless of whether captions ever existed. This is also the only route that works on a video whose captions are disabled, and the only one that gives you speaker separation — YouTube's captions never distinguish voices.

If the video is not yours and you do not have the file, you are into YouTube's terms of service, which is a decision rather than a technical problem. Downloading other people's videos to extract audio is restricted; the answer for research usually lies in fair-use quoting from the transcript panel rather than reproducing the whole thing.

For a video you own, this route is strictly better: punctuation, speaker labels, and an SRT you can upload back to YouTube to replace the automatic captions.

Upload your own caption file. It takes one upload, it makes the transcript panel work for everyone, and it is the single biggest accessibility win available on a video.

Generate an SRT from the audio, read it once to fix names and terms, then add it in YouTube Studio under Subtitles. Your caption track replaces the automatic one, the transcript panel starts working, and search engines index real text instead of guesses.

Two details people miss. Automatic captions keep running alongside your upload in other languages, so set the language of your file correctly. And if your video is music-only or has long silent stretches, YouTube may never generate automatic captions at all — an uploaded file is the only way that video will ever have a transcript.

Private and members-only videos. If you cannot watch it signed out, no tool can read it signed out either, and any site claiming otherwise is either lying or asking for your Google credentials. Never hand those over.

Videos with no speech. A silent timelapse has no transcript to produce, and a tool returning "no speech found" is being accurate rather than broken.

Poor source audio. A phone recording of a conference room with three people talking over each other transcribes badly whether the words come from captions or from audio. The ceiling is set by the recording.