How to add subtitles to a video (and why auto-captions get it wrong)

There are two completely different things people mean by "add subtitles to a video", they need different tools, and picking the wrong one is why a lot of people end up with captions that don't show up where they wanted them. Start there, because everything else follows from it.

Burned-in vs soft subtitles

Soft subtitles are a separate track (an .srt or .vtt file, or a platform's own caption layer) that the viewer can switch on and off. They're searchable, translatable, and the accessible choice. They're also off by default for most viewers on most platforms.

Burned-in subtitles (sometimes called open captions or hardsubs) are drawn into the pixels of the video itself. Nobody can turn them off, they look identical everywhere, and they survive being re-uploaded, downloaded, or reposted by someone else.

The rule of thumb: short-form gets burned in, long-form gets soft subtitles, and there's no reason you can't do both. A YouTube video should have a real caption track for search and accessibility. A TikTok, Reel or Short should have the words drawn on, because the median viewer is scrolling with sound off and will not go looking for a caption toggle.

The free options, honestly assessed

• The platform's own caption tool. TikTok, Instagram and YouTube all auto-caption on upload, and all three let you edit the text afterwards. Free, no extra software, decent accuracy on clear speech. The catch is that you're editing on a phone, the styling options are limited, and the captions only exist on that platform, so cross-posting the same video elsewhere means doing it again.
• YouTube's caption editor for the long version. Still the best free way to get a clean .srt for a long video: let YouTube auto-generate, fix the errors in its editor, then download the corrected .srt and reuse it anywhere.
• CapCut or your phone's editor. Free, burns in, gives you real styling control. Genuinely fine for a handful of videos. It stops scaling when you're doing this weekly across several clips.
• Whisper locally. If you're technical, OpenAI's Whisper runs on your own machine for free and is more accurate than most platform auto-captions. You get an .srt and then still need something to burn it in.

Why automatic captions get things wrong

Every automatic option shares the same three failure modes, and knowing them turns proofreading from "rewatch everything" into a thirty-second check.

Names, jargon and acronyms. Speech-to-text models guess from context, and they have no context for your company name, your guest's surname, or an industry term. This is where nearly all embarrassing caption errors live.

Invented text over silence. Transcription models do not return nothing when there's nothing to transcribe. Run a video with music, background noise or plain silence through one and it will confidently produce short phrases nobody said. "Thank you.", "Bye." and "I don't know." are the classics. If your video has non-speech stretches, read the captions in those stretches specifically.

Timing drift after an edit. Captions are timed against the video they were generated from. Cut a clip out of the middle of that video and every caption is off by however far in the clip started, unless whatever cut it re-timed them. Check the end of a clip, not the start, because drift accumulates and is most visible at the end.

What actually makes captions readable

Once the words are right, the styling decides whether they get read. The defaults that work almost everywhere:

• White text with a dark outline, not a background box. An outline stays readable over anything; a box eats the frame.
• One or two lines at a time, never three. If a line needs to wrap twice, it should have been two captions.
• Off the very bottom. Every platform overlays its own interface along the bottom edge. Captions sitting there get covered by usernames and buttons. Roughly the lower third, not the lower edge.
• Big. Most people are watching on a phone at arm's length. If it looks slightly too large on your monitor, it's about right on a phone.

Doing it at volume

The free options are fine for one video. They stop being fine when you're producing several clips a week, because the work is identical every time and none of it is creative. That's the point where an automated pipeline (transcribe, style, burn in, export per platform) pays for itself, whether that's ours or anyone else's.

What doesn't change is the proofread. Automatic captions are good enough to post and not good enough to post unread, and the three failure modes above are where to look. More on why they matter so much in captions aren't optional anymore, and on the platform-by-platform differences in TikTok vs Reels vs Shorts.