The direct answer
To add captions to a podcast clip properly: get a transcript with a timestamp on every word, group the words into short lines of roughly three to five words broken at phrase boundaries, burn them into the video so each word appears when it is spoken, check the words that matter most, which are names, numbers and anything in a second language, and only then choose a style. Captions on a short-form clip are not an accessibility layer added at the end; for the many viewers who meet the clip muted, they are the clip.
Why timing beats everything else
A viewer will forgive a plain font. They will not forgive a caption that runs ahead of the voice. Half a second of lead is enough to make a clip feel dubbed, and it is the single most visible defect a clip can have. Timing has to come from the audio at the level of the word, not from dividing a sentence's duration evenly across its words. When the transcript comes from a source that chains its cues end to end, with the next line starting exactly where the previous one ended, the first word of every line inherits the silence before it and appears early; that has to be corrected before the captions are burned in.
Grouping: how many words per line
Three to five words per line is the range that reads at a glance on a phone. Lines should break at phrase boundaries, never inside a name, a number or a fixed expression, and a group so short that it flashes for a fraction of a second should be merged with the next rather than shown. When captions are grouped by a fixed word count instead of by phrase, you get lines like "the fifteenth of / March" and viewers stumble.
The words that must be right
Automatic transcription is good on ordinary words and predictably bad on three kinds, which are exactly the kinds viewers check:
- Proper names. The guest's name, a company, a city. A name spelled phonetically on the guest's own clip is the mistake most likely to be shared.
- Numbers. Dates, prices, percentages. Decide on one convention, digits or words, and check that the caption matches the hook.
- Words from another language. An English term in a French clip, a French word in a Darija clip. Write them as spoken, in their own script, in the right place in the line.
Filler words are a judgment call. Removing pure hesitations makes captions readable; removing repetitions that carry emphasis flattens the speaker. Whatever you remove, confirm the remaining words still appear when they are said.
Rules that depend on the language
Captions are language-specific in ways a generic tool misses:
- French puts a space before question marks, exclamation marks, colons and semicolons, and that space must never be a line break. Elisions like "l'épisode" are one word.
- Spanish opens questions with an inverted mark; a caption or hook with the opening mark and no closing one is a visible error.
- Arabic reads right to left and its letters change shape depending on their neighbours. Letter spacing that looks stylish in Latin script breaks the joins in Arabic and makes captions unreadable, and a mixed line with a French word in it has to be laid out in visual order. A dialect like Moroccan Darija must be transcribed as spoken rather than rewritten into Standard Arabic; see Darija vs Modern Standard Arabic subtitles.
The hook card is not a caption
The title shown for the first seconds is a separate text, written separately, and placed separately: at the top for a single speaker, at the seam for a stacked two-person layout, below the webcam on a gaming layout, always off the face. It is also the text most likely to need editing, because it is generated rather than transcribed, and it should be re-rendered on its own when it changes rather than by re-cutting the clip.
Style, last
Once timing, grouping and correctness are right, the style is a matter of brand: a highlighted current word, a solid block, an outline, a pill behind the active word. The constraint is legibility on a phone over a moving background, which argues for a heavy face, strong contrast and captions kept in the lower third but clear of the platform's own overlays. On a two-person clip the captions sit at the seam. Whatever style you pick, keep it the same across the channel; consistency is what makes a clip recognisably yours before the viewer reads a word.
In ClipFlap
Every clip from the AI podcast clip generator comes with word-timed captions burned in, grouped by phrase, with the language rules above applied and a choice of styles set before the run. The review step is yours: names, numbers, second-language words and the hook. The rest of the workflow is in the complete guide to podcast clips.