Skip emoji in captions when CPS is tight
Emoji is a character with width and parse time. Paid and SDH usually skip it. An organic hook can keep one if the mute pass still finishes.
Emoji in a caption is not free personality. It is a glyph that occupies width and time. When characters-per-second is already tight, skip it. Paid social and SDH usually skip it even when CPS is loose. An organic hook can keep one if the mute pass still finishes the line.
A glyph with a width
Reading speed is characters per second. Spaces count. Punctuation counts. A face, a fire, a shopping bag counts. Some emoji render wider than a Latin letter. Some arrive as a colour blob the eye has to recognise before it can continue the sentence. That recognition is time you already spent on the claim.
Auto-transcribe will not usually invent emoji. People add them in the edit because the line “felt flat.” Flat is often a CPS problem, a contrast problem, or a dull sentence. A sticker does not fix those. It adds a fourth job to a two-line block: read the words, parse the icon, ignore the chrome, still finish before the next cue.
On a mute feed the caption is the soundtrack. Spending the soundtrack on a cartoon is how the verb, the price, or the name falls off the end of the hold. Reading speed does not care that the blob was on-brand.
Word-highlight makes this louder. Karaoke type already asks the eye to track a live token. An emoji in that line is a token with no spoken timestamp, or a spoken word replaced by a picture. Either way the clock and the picture disagree. If you need pop, pop a word. Do not pop a sticker.
Line length is the other meter. Latin-script cues are commonly kept near 42 characters and two lines. A wide emoji can wrap a line you thought was safe. The wrap then collides with a face or with platform chrome. You did not add warmth. You added a third line.
Paid and SDH skip
Paid placements are a review surface. Legal, brand, and the platform’s ad tools will look at the frame. Emoji in the speech track reads as unreviewed copy. It also sits in the same band as the disclosure, the CTA, and the offer. A bag icon next to “£19.99*” is how the asterisk becomes the joke.
SDH is captions for people who cannot rely on the mix. The convention is words, speaker IDs, and labelled sounds in brackets: [music], [door]. An emoji is not a labelled sound. A music note is not [music under]. A door is not a door. Deaf and hard-of-hearing viewers, and anyone on mute who is actually reading, should not have to decode a pictogram to get the beat the edit is built on. If the sound matters, write the label. If it does not, write nothing.
Courses, webinars after the cut, and anything that will be reused as a master inherit the SDH habit even when you are not delivering a full SDH file. Sentence case, two lines, no stickers. The clip may later become the course snippet. You do not want to recaption emoji out of thirty files because a buyer asked for a clean track.
Brand safety is the quiet version of the same skip. Emoji palettes differ by OS and by app. The glyph you previewed on a desktop is not the glyph a phone will draw. Skin tones, vendor-specific designs, and “simple” icons that render as tofu boxes are all ways a paid caption becomes a different sentence than the one you approved.
BASIC presets (simple, plain, corpo) exist for this intent. They are 1×, line-level, readable. They are not a sticker kit. DYNAMIC at 2× is already motion. Adding emoji on top of Glass is two decorations on a line that still has to be read.
Organic can keep one
Organic mute-autoplay is allowed a little performance. One emoji, on a short spoken hook, when CPS still clears the mute pass, can work as a beat: the punchline is a word, the icon is a wink the voice already made. That is a choice you can see in one second.
Rules that keep it a choice instead of a habit:
- One, not a row. Three emoji is a reaction gif sitting in the caption track.
- After the claim, not instead of it. If the icon replaces “sale” or the product name, you have hidden the only reason to stay.
- Not on names, SKUs, or prices. Those strings are already ASR’s failure mode. Do not dress them.
- Not on a maxed line. If sentence case without the icon already fills two lines, you are done adding.
- Re-check on a phone. Vendor glyphs change size. If the wrap appears only on iOS, that is still your wrap in the feed.
If the hook is a line nobody said, the icon belongs on an overlay, not in the transcript. Captions follow the mouth. Decorating a transcript with pictures the talent did not say is how organic still looks like a caption file someone doodled in.
A/B tests that “include emoji in captions” are testing a different object if the control was already unreadable. Fix CPS first. Then, if you still want the icon, test one on the hook cue only, not on every line in the file.
The mute pass is the policy
You do not need a house ban. You need a hold.
Lock the cut. Transcribe. Fix names and numbers. Split clauses so two lines can be finished. Burn a candidate look — add captions to a video is 165 codes, 21 BASIC at 1×, 9 DYNAMIC at 2×. Then watch on mute at phone size.
If you cannot finish the line, remove the emoji before you reach for a punchier preset. If you can finish it and the icon is doing a job the word is not, keep one on the hook. If the file is paid, SDH, or a course snippet, remove it even if you could finish it. Those briefs are words.
The AI caption generator will not make this decision. Caption styles will not either. Snap and Reveal highlight words. They do not need a sticker to prove they are karaoke.
Skip is the default. One is the exception you can still read. A row of emoji is a different graphic, and it does not belong in the transcript.
Caption the next cut without the extra glyphs in Versely.
FAQ
Do auto captions insert emoji on their own?
Usually not. People add them in the edit. That addition still counts toward CPS and line length. If the mute pass fails after you add one, the icon is the miss.
Why do paid and SDH skip emoji even on a short line?
Paid is a review surface; pictograms next to offers and legal lines look unreviewed and steal width from the claim. SDH is labelled speech and sound, not pictograms. A music emoji is not [music].
Can an organic Reel keep emoji in every cue?
No. One on the hook, if the line still finishes on mute, is the exception. Repeating it on every cue is a sticker track. It spends reading time the scroller needed for the words.