You get about three seconds to say it
A GIF doesn't get read the way a sentence does. It gets glanced at while someone scrolls, and if the text needs a beat to build before the payoff arrives, the payoff arrives after the thumb has already moved on. Whatever you're saying has to work on first contact: short words over clever ones, the punch loaded at the front of the line, one idea per caption instead of a paragraph you're hoping someone pauses to finish.
Two looks, and each one is a promise before anyone reads a word
The editor gives every caption a choice of two styles, and the choice itself communicates something before the words register. Meme style, white letters with a heavy black outline, auto-uppercased, is the internet's oldest shorthand for "this is a joke, relax." Put it on a serious message and you've undercut yourself before the sentence lands. Clean style, a plain brand-weight font with a color picker, reads as information: a label, a callout, a name on a screen. Put it on a punchline and the joke goes flat, because the visual language told the viewer to take it seriously.
Match the style to the register you actually want, not to habit. A product demo captioned in meme font looks like it's mocking itself.
Starting from a GIF you already have
If a GIF already exists and just needs words on it, drop it straight onto the tool. It's decoded into its individual frames right there in the tab, after which it behaves exactly like any freshly imported clip: trim it, crop it, and add up to three captions, each with its own style and position. Re-export and the words are burned into the pixels, not floating in some layer a repost could strip away. Two honest caveats: transparency flattens onto white the moment the GIF is decoded, and a very long existing GIF imports only about its first two minutes.
Timing turns three captions into a structure
Each caption can carry an optional window, visible from one second of the loop to another. Skip it and the line holds for the whole GIF, the safest default for a single idea. Set windows on all three and you're pacing a joke: setup first, reaction on the beat, tag right before the loop resets. That's a complete structure in one file, and it only works because the timing is exact rather than approximate.
Where timed captions get you into trouble
Timing has one blind spot worth planning around: the older desktop versions of Outlook, roughly the 2007 to 2019 builds, freeze an animated GIF on its very first frame and never move past it. If your funniest line is timed to arrive two seconds in, a chunk of any corporate audience never sees it at all. The fix costs nothing: front-load your best line into frame one, and everyone downstream gets the version that works everywhere.
The other honest limit is the cap itself. Three captions is enough for a real structure, not enough for a paragraph. If the message needs more than that, it needs a different format, and a GIF was probably the wrong container for it anyway.