Captions are not the CTA card

Captions are not the CTA card. Speech type is a transcript of what was said. A CTA card is an ask: shop, follow, swipe, click. An end screen is a last-seconds overlay YouTube draws. Three objects. If you make the caption track carry the ask, you get a karaoke “link in bio” nobody said, or an end screen fighting a two-line block. Keep the jobs apart.

A transcript is not an ask

Auto captions transcribe. Nobody has to have said “shop the SKU” for you to need that line. If they did not say it, ASR will not write it. If they did say it, it is still speech, timed to the mouth, replaced every clause. It is a bad place to park the only ask, because the cue will yield when the next sentence arrives.

The card is a request, not a caption. “Link in bio,” “shop the SKU,” “follow for part two,” a YouTube info card, a Spark-style product tag — something the viewer can do. Text overlay is the wider craft of burning words you wrote. A hook line is not a CTA because it does not ask. Hook-body-CTA is the structure that saves the ask for a beat. The card is the object that carries it.

Timing is the whole skill. Drop the card in the first second and you have covered the hook. Drop it only after the file has already lost the people who would have acted and you are decorating a skip. Completion rate grades whether they were still there to see it. Safe area grades whether the platform chrome covered it. A YouTube card that sits under the player UI is a card nobody clicked.

Captions occupy the same band. Two-line speech plus a CTA card plus a legal row is three graphics. Count them as one budget. Sequence or stack on purpose. Do not hope the DYNAMIC pop “feels like a button.”

Paid social already has a platform CTA: a shop row, a Demand Gen button, a sponsored chip. Your burned “SHOP NOW” on top of that chip is two asks. Let the platform button be the button when it exists. Use the caption for the spoken claim. Use a small authored card only if the platform does not draw one.

End screens are a different slot

YouTube draws an end screen. You pick elements — videos, playlist, subscribe, a channel, a link on eligible accounts — and they sit on top of the last five to twenty seconds of the file. The picture underneath still has to exist. A talking head that keeps talking into the overlay, or a caption that sits where the elements land, is a fight with your own file.

It only works if they finish. A cold open that holds and a body that leaks will never show the end screen to the people you wrote it for. Short-form players often loop instead of showing a next-up mosaic, so an end screen designed for long-form YouTube is the wrong object on a Reel. There the last beat is a CTA card in pixels, or a join back to frame one for a rewatch.

Do not confuse an end screen with a logo hold. A three-second sting with no element on it is a branding clip, and it is also a loop-killer. If you need the YouTube elements, leave picture that can survive being covered. If you need a short-form ask, burn it in as a card and skip the empty mosaic.

Do not burn a fake end screen into a Short and also run two-line captions through those last seconds. The loop will show a covered pack and a leftover “subscribe.” Empty the caption track for the hold, or empty the hold.

Sidecar SRT will not draw a CTA. It will not draw an end screen. It will transcribe speech for a player. Reels will not play it. Keep the objects in their tools.

Layout: speech, then ask, then chrome

Decide the stack before you render.

Speech is the soundtrack for mute. Two-line BASIC for ads and demos. DYNAMIC only on a spoken hook you can finish. Off the SKU. Off the bottom ~200px keep-out. Off the Shorts AI overlay when that badge applies.

The ask is one beat. Authored overlay or a real platform card. One ask per file. Two cards is two jobs, and the viewer will do neither. Place it inside the current safe area, then check it on the phone UI. If the platform has no mid-roll card, burn the CTA into the last beat instead of hoping the last caption is read as a button.

The end screen is YouTube long-form. Last five to twenty seconds, elements on top, captions yielded or raised. Not a Reel. Not a Short.

Legal / Article 50 marks are a fourth graphic when they apply. They are not a CTA and they are not a caption. First exposure, in the picture. Do not hide the mark in a subtitle cue.

Karaoke on an ask the mouth never said is a category error. Timed kind needs word timestamps on speech. “LINK IN BIO” popping syllable-by-syllable is an overlay wearing a karaoke costume. Use a static card.

Numbers in the ask must still match the PDP. “Shop at $19” on a $21 page is the caption-numbers problem on a CTA. Same pass.

Burn the transcript; author the ask

Order:

1. Lock the cut.
2. Transcribe speech. 165 codes. Fix names and numbers.
3. Burn captions as captions. 21 BASIC at 1×, 9 DYNAMIC at 2× if the hook is spoken.
4. Author the CTA card as overlay, timed to a beat that is not the hook and not the chrome.
5. If the destination is long-form YouTube, design the last seconds for end-screen cover and yield the caption track.
6. Mute phone pass: can you read the claim, see the pack, and find the one ask.

The AI caption generator will not write the card. Add captions to a video will not place a YouTube end screen. Different tools. 45 Studio CSS presets are caption looks, not buttons.

Captions are not the CTA card. Speech, ask, end screen. One job each.

Keep the ask off the transcript in Versely.

FAQ

Can the last caption line be the CTA?

It can, if someone said it, and if that cue holds through the ask beat without covering the pack or the platform button. Usually they did not say it. Author a card. Do not force ASR to invent a button.

Do Reels have end screens?

Not as YouTube draws them. Short-form loops. Put the ask in a CTA card in pixels, or in the platform’s own button. A fake mosaic is a covered last frame.

Should karaoke animate the shop line?

No. Karaoke is a word clock on speech. A shop line is an overlay. Static card, one ask, in the hole chrome leaves.