Video captions best practices

The best captions are accurate to the word, split into short lines at natural pauses, held on screen long enough to read at about 160 to 180 words a minute, and placed near the bottom of the frame without covering a face, a mouth or text that matters. Everything else, the font, the colour, the animation, is secondary to those four.

The detail comes from two sources that disagree on almost nothing: broadcast subtitling guidance, where the BBC publishes the most complete public rulebook, and accessibility guidance from the W3C and the Described and Captioned Media Program. This post takes their rules and applies them to the short promos, tutorials and creator ads most businesses actually make.

Why captions

WCAG 2.2 success criterion 1.2.2 is Level A, the baseline: prerecorded video with sound needs captions. The criterion exists for people who are deaf or hard of hearing, and the W3C adds people who take in written information more easily than audio. In practice there is a third group, the viewer scrolling a feed with the sound off or in a place where sound is not an option, and social video is built around that viewer.

Captions come in two kinds. Closed captions are a separate track the viewer can switch on and off. Open captions are drawn into the picture and cannot be turned off, which is what “burned in” means. Open captions are the only kind guaranteed to show on every platform and every placement, including ads and reposts. The price is that they are permanent: a typo ships in every copy.

Line length: two short lines, broken at pauses

A caption is read in a glance, so its shape matters more than its font. The BBC limits online subtitles to 68 percent of the width of a 16:9 frame and 90 percent of a 9:16 frame, and 37 characters for broadcast. It recommends a maximum of two lines on landscape and square video and three on vertical, where lines are narrower. The DCMP Captioning Key also prefers no more than two lines per caption.

Where the line breaks is the part most tools get wrong. Both guides say to break at a natural pause, ideally at punctuation, and never between words that belong together: an article and its noun, a preposition and its phrase, a pronoun and its verb.

The fitness app UGC video shows why. Its opening line is “Okay I have twelve minutes a day, that’s it, that’s the whole window.” That is 13 words, too long for one caption. Split at the first comma it becomes two readable captions: “Okay I have twelve minutes a day,” then “that’s it, that’s the whole window.” Split by character count instead, it might break between “the” and “whole window”, which is exactly the break both guides forbid.

Timing: plan for reading, not for speaking

The BBC recommends a subtitle speed of 160 to 180 words per minute and a target minimum of about 0.3 seconds per word, so a four word caption stays up for 1.2 seconds. The DCMP Captioning Key, written for educational media, sets lower ceilings: 130 words per minute for lower levels, 140 for middle and 160 for upper levels.

Check the scripts you have against those numbers. The fitness app ad has three talking beats of 6, 6 and 4 seconds. The first two carry 13 words each, 130 words per minute across the beat, which sits comfortably under the BBC range. The dashboard tour video is a tutorial with 7 second steps; “Break it down by channel; paid is the one that fell.” is 11 words, under 100 words per minute, so a viewer can read the caption and still look at the screen it describes.

When a line is too fast to read, slow the voice or lengthen the scene before trimming words. The DCMP guide also notes a small trick: a caption can borrow about 15 frames before or after the audio without viewers noticing, which helps a short line stay readable.

Placement: low, but never over a face

The DCMP Captioning Key places captions on the bottom two lines, and says they must not cover names, faces, mouths, or text and graphics needed to understand the video. When they would, captions move to the top of the frame, or elsewhere if the top is also busy. The BBC gives the same rule and adds a note for vertical video: subtitles usually sit a little higher, still in the lower third, because faces are generally in the top half of a tall frame.

Two situations make placement hard in short video. Creator ads put a face in the middle of the frame, so a caption that drifts upward can cover a mouth, which lip readers rely on. And offer videos print prices, codes and dates near the bottom, so a caption there can hide the one fact the ad exists to deliver. Decide which text owns the bottom of each scene before exporting. In the fitness app ad, scene 6 prints FREE WEEK with the price after the trial underneath it, so a caption on that scene has to sit clear of both lines, or that scene should carry its message in on screen text alone. For the platform overlays on top of all that, see vertical video safe zones.

Accuracy: automatic is a draft

The W3C is blunt: automatically generated captions do not meet user needs or accessibility requirements unless confirmed to be fully accurate. The errors are not only cosmetic; a missed “not” reverses a sentence.

Accuracy also covers what is not speech. WCAG expects captions to identify who is speaking and to convey meaningful sound effects. The BBC asks for a label when music is part of the action and warns against captioning every trivial sound. A creator ad with one voice needs no speaker labels. A tutorial where a notification sound tells the viewer something happened needs a short label for it.

Names and numbers deserve a second read: product names, prices, dates, codes and street names. The restaurant opening video is narrated in Spanish, and a caption that misspells Calle de Sagunto sends people to the wrong street.

The style rules in one table

The table below the article lists each caption rule with its value and its source, so it can sit beside an edit. The two rows that change most with format are lines per caption and line width: a vertical frame can take three narrower lines, a landscape frame should stop at two.

Common mistakes

• One long caption per sentence. Split it at pauses into captions of two short lines.
• Word by word pop captions for everything. They suit a punchy hook but are tiring across a 40 second tutorial.
• Captions over the price. Move the caption or move the price; never let one hide the other.
• Publishing automatic captions unread. Read every line against the audio before export.
• Burned in captions plus platform captions. Two caption tracks on top of each other is worse than either alone; with open captions, leave the platform’s automatic captions off.
• Timing captions to the edit. A caption should follow the speech, and stay long enough to read even if the cut comes early.

Build it

Captions in the video builder are derived from each scene’s narration text, so they match the script exactly and carry no speaker labels or sound descriptions of their own. They are on by default and toggled per scene in the voice dialog, where each scene also takes a voice speed of 0.85x, 0.9x, 1x, 1.1x or 1.25x and a delay in seconds. They are burned in as one dark pill centred at the bottom, typing on as the line is spoken, capped at 72 percent of the frame width, with 26 pixel type on landscape and 30 on vertical, inset 44 pixels from the bottom on landscape and 120 on vertical. The captions page covers how they are drawn, the voiceover languages post covers captions in other scripts, and the steps are in the tutorial on adding voiceover, captions and music.