The one-second test
A social viewer makes a keep-or-scroll decision faster than a sentence can be spoken. That single constraint drives every other decision on this page: the hook, the speed, the voice, and the length.
It also means the opening is the entire video. If the first five seconds do not make a promise, nothing later in the clip can rescue it, because most viewers will not be there to be rescued.
Test it honestly. Watch your own output with sound off, then with sound on. If the on-screen text alone does not work without audio, you are relying on the audio to carry more attention than a feed will give it.
Writing for someone who is scrolling
Social copy is not written prose. It is built to survive a thumb moving at speed, which means short sentences, concrete nouns, and no subordinate clauses waiting on a payoff.
Specificity is what separates a clip that works from one that does not. A real number, a real name, or a real consequence outperforms an abstract claim almost every time, and it is the easiest thing to get right before generating anything.
Cut every word that does not earn its place. Social is the one context where the 2,000-character limit is genuinely generous and readers are genuinely impatient, so trim before you generate rather than after.
Keeping a consistent brand voice
Individual posts are cheap to produce, which is exactly why brand voice drifts. Each clip is made in isolation, the voice is chosen from memory, and within a month the feed sounds like three different accounts.
Pick one voice and one speed and use them for everything. Consistency is more noticeable to a follower than quality is, and an account that sounds like itself is doing something no single great clip can do.
Keep a short note of the settings you use. It sounds trivial, but the most common cause of an inconsistent feed is somebody reasonably assuming they remembered the right speed.
Fitting the platforms
Each platform has its own length limit and its own audience behaviour, so the same script does not transfer cleanly between them. A thirty-second Reel script is roughly 400 words, and it needs to be a complete idea rather than the first half of one.
Vertical video with a caption burned in is expected on every platform, and the caption is usually read rather than the audio. Write the caption as a summary and the narration as the fuller version, rather than making them identical.
Check the length after export rather than trusting the character count. Speech at 1.0x varies slightly with punctuation and with how numbers are written, and a clip that lands two seconds over gets cut off mid-sentence.
Step by step
Write the hook first, before anything else
The first sentence decides whether the rest is heard. Lead with the specific claim or question rather than a greeting. "Most transcripts fail on one setting" earns attention; "Hi, today we will look at" does not.
Keep it inside the platform limit from the start
Write to the limit rather than trimming afterwards. 2,000 characters is roughly thirty seconds of narration, which covers a Reel, a Short, or a LinkedIn voice note. Trimming a finished script usually breaks the sentence that was doing the work.
Use a bright voice at 1.0x or above
This is the one use case where a higher-energy voice is correct. Social feeds are fast and slightly manic, and a measured podcast-style read loses the viewer here for the same reason it would work in a documentary.
Front-load the value and end cleanly
Say the most interesting thing first rather than building to it. Viewers leave mid-video, so anything placed after a fifteen-second mark is optional. Finish on a complete sentence with no trailing clause.
Export MP3, since social platforms compress anyway
There is no point working in WAV for a platform that re-encodes to AAC on upload. Export MP3, upload, and keep the length comfortably under the limit so the platform does not trim you mid-sentence.
If you need captions for social clips you have already recorded rather than narration for new ones, see transcribing a video into subtitles.