Blog

The Podcast Has a Transcript. There Is Nowhere to Put a Caption.

November 8, 2026  ·  Tablingo

Apple and Spotify built transcripts. YouTube built captions. Neither is text you can read while an audio episode plays — the reason is geometry.

This was the year every platform announced podcast text. Apple generates transcripts automatically and has backfilled its catalogue. Spotify shows them in the Now Playing view. YouTube, now the most-used podcast service in the United States, has captions on the video and an auto-translate menu with a hundred languages behind it. By the count of announcements, the problem is closed.

Then you press play on an audio episode and look at the screen. Nothing appears. Not a line, not a word. The text exists somewhere, and it is not there.

Two words the industry treats as one

A transcript is a document. It has paragraphs, it scrolls, and it usually lives in a panel you have to open. A caption is a line of text timed to something you are watching, placed on the thing you are watching.

Every platform blog post and most news coverage uses the two interchangeably, and the difference is not editorial taste. It is geometry.

The accessibility standard split them first

The Web Content Accessibility Guidelines contain two separate success criteria, and they were never meant to be interchangeable. One covers audio-only content — a podcast, a radio recording — and asks for a transcript. The other covers video with sound and asks for captions.

The split is not arbitrary. A transcript is what you can offer when there is no picture: the whole text, in order, readable at your own pace. A caption is what you offer when there is a picture, because the picture gives the text somewhere to sit.

Read the two criteria together and you have the rule that explains everything else on this page. On an audio player there is nothing to draw on. Not because the platform was lazy, but because the surface does not exist.

So the text went into three different containers

Look at what each platform actually shipped and the same constraint shows up as three different compromises.

Apple generates the transcript automatically, in eleven languages: English, Danish, Dutch, Finnish, French, German, Italian, Norwegian, Portuguese, Spanish and Swedish. It appears behind a button in the player, follows playback, and lets you tap a paragraph to jump to that moment. It is also, by construction, one language — the one the episode is spoken in. There is no translate control, because this was built as an accessibility feature rather than a translation one. Listeners can copy roughly 200 words at a time and there is no full download. Episodes over ten hours may not be transcribed at all, and dynamically inserted ads may be missing from the text.

Spotify puts its version in the Now Playing view and on the episode page, as a card you scroll to. Availability varies by show and by episode, not every creator can manage it, and desktop is less consistent than mobile. Some shows also get a translated transcript. When it is there it is genuinely useful, and it is still a panel sitting next to the player rather than on it. A video episode can carry real captions, but that is a separate track and it is not guaranteed either.

YouTube is the only one of the three where the text lands on the thing you are looking at, because YouTube always has a picture. Two limits come with it. A video carries one active caption track at a time, so auto-translate replaces the original rather than adding a second line underneath — you can read the episode in your language or in its own, never both. And on a live stream the automatic captions are English only; auto-translated captions appear on the archive after the broadcast ends, not during it.

Where the text isWhile the episode playsWhich language
Apple PodcastsBehind a buttonThe episode's own
SpotifyA card belowThe episode's own
YouTube, uploadOn the pictureOne at a time
YouTube, liveEnglish onlyEnglish

Three containers, one shared property: at no point in that table do you get the original and your language at the same time. The formats are not built for that, and nothing in the last two years of announcements changed it.

The audience moved to the only place with a frame

Edison Research asks weekly US podcast consumers which service they use most. In the third quarter of 2026 the answer was YouTube at 40%, Spotify at 17%, Apple Podcasts at 9%, with Netflix entering the list in seventh place at 3%. In October 2024 the same measure read 31, 27 and 15. In two years YouTube gained nine points, Spotify lost ten, and Apple lost six.

I am not going to claim captions caused that. The reasons are recommendation feeds, shows built for cameras, and an audience that expects a face. But the side effect is worth naming precisely: the audience moved to the one surface where a caption can physically exist, and left behind the apps where text can only ever be a panel.

Even with a frame you get one line, in one language

Having a picture solves placement and leaves language untouched. Three details, all from platform documentation.

Auto-translate replaces. Pick a target language in the player and the captions switch to it. There is one track on screen, and it now holds one language. Putting the original beside the translation is not a setting you have not found; it is a shape the player does not have.

Live has no translate at all. Automatic captions during a broadcast are English only, and translated captions arrive once the stream ends and the archive is processed. If the show is live and you do not read English, the text layer is empty for the entire time it matters.

The newest feature changes the audio, not the words. Since February 2026, YouTube generates auto-dubbed tracks in 27 languages for eligible uploads, on by default, with a more natural voice mode in eight of them. YouTube says more than six million viewers a day watch at least ten minutes of dubbed content. Live dubbing is a pilot for early 2027, starting with English and Spanish. So the choice the platform now offers is narrower than it sounds: hear it in your language, or read it in a language you may not have. Not both, and not on the same screen.

Where we sit

Our extension does not ask the player for a caption track, and it does not need a frame to draw on. It reads the audio of the tab you are already playing and puts the original line above the translation, on the page.

Which makes the audio podcast the one case where the geometry stops applying. An episode playing in a browser tab has no picture, and it still gets two lines — the case that the guidelines, the players and the transcript panels all agree is impossible.

The boundary is the same as everywhere on this site: it happens in a tab. An episode playing in a phone app with the screen off, or through a car stereo, gets nothing from us and nothing from anyone — and that is not a failure to fix. That is what audio is for.

Two checks

  1. Play an episode and look at the screen without touching anything. If the only text in the player is a button you have to press, that player has no canvas. You are looking at the reason, not at a missing feature.
  2. On a video episode, turn captions on, then pick auto-translate, then try to get the original back on screen alongside it. You cannot. That is the shape of the track, not a setting you have not found yet.

Neither check requires trusting a vendor, including us.

What you actually wanted

Underneath all the announcements, the request is modest: I want the words in front of me while the words are being said, and I want them in a language I read. Every platform has now shipped one half of that sentence. The half that is still missing is not a feature. It is a surface — and the surface was never going to come from the side that owns the player.

If you want to try ours, it's called Tablingo — a browser extension that reads the tab you are already playing and puts the original line above the translation. The on-demand podcast case is here, the live one here, and the episode that changes language halfway through is here.