Step-by-step instructions
1. Check whether captions already exist
Most videos uploaded to YouTube in the last several years already have auto-generated captions, even if the creator never opened the captions tab. TikTok generates a spoken-word transcript the same way. If the CC button in the platform's own player actually does something, there's already something to extract — you don't need to hire anyone to create it from scratch.
2. Extract the captions as a file, not just on-screen text
Pull the existing captions out with SubExtract's YouTube caption extractor — paste the video URL, extract, and download as .srt (timestamped) or plain text. For TikTok, the TikTok transcript tool does the same for the spoken-word audio. This is extraction, not generation: the tool pulls text the platform already produced, it doesn't run its own speech recognition on the video.
3. Proofread against the accuracy baseline
Auto-captions run 90-95% accurate on clear audio, and worse with accents, technical jargon, or music under the dialogue. Read the extracted text against the video once, end to end, and fix names, product terms, and punctuation before publishing it anywhere. How to check caption accuracy covers a faster way to do that pass. Skip this step and you're publishing a "close enough" transcript, not an accessible one.
4. Attach or publish the corrected file
Two ways to use it, depending on where the video lives:
- Sidecar subtitle file — drop the
.srtnext to the video in your player, video editor, or CMS. Works for anything you host yourself: Vimeo, a website embed, an LMS. - On-page transcript — paste the plain-text version below the video on the page it's embedded on. Simpler if you don't control the player, and it doubles as page content.
Either way, you end up with a real, readable caption or transcript without paying a captioning vendor per minute of footage.
DIY extraction vs. a paid captioning vendor
| | Free extraction + proofread | Paid captioning vendor | |---|---|---| | Cost | $0 (free tier covers a handful of videos; Pro is $10/mo unlimited) | Per-minute rate, adds up at volume | | Turnaround | Minutes | Hours to days | | Starting accuracy | 90-95% (platform's own auto-captions) | Usually higher out of the box | | Your effort | You proofread | Vendor proofreads (sometimes) | | Works when... | The platform already generated captions | Always, including a source video with zero captions |
The trade is time for money. If you'd be proofreading a vendor's file anyway, extraction gets you most of the way there for free. If your volume is high enough that the proofreading hours cost more than a vendor invoice, that's the point to reconsider.
When there's nothing to extract
This whole approach depends on the platform having generated captions in the first place. A video with no auto-captions — music-only audio, captions disabled by the uploader, or an upload so recent the auto-generation hasn't finished — has nothing to pull. If /video-captions comes back empty on a YouTube video, what to do when a YouTube video has no captions covers the actual next step.
Frequently asked questions
Is this legally sufficient for ADA or WCAG compliance? It gets you most of the way there, but "sufficient" depends on your obligations, jurisdiction, and content. This is general information, not legal advice — see the video accessibility compliance guide for more, and talk to a lawyer or accessibility consultant for your actual situation. Responsible for accessibility across a whole video catalog? The accessibility teams page covers that workflow.
Does extracting captions cost anything? SubExtract's free tier gives 5 free credits, no credit card required — enough to extract captions for a handful of videos before you'd need Pro.
Can I just use the auto-captions without proofreading? You can, but 90-95% accuracy means roughly 1 in 10 to 20 words could be wrong, worse on technical terms, names, and accented speech. For anything published or citation-grade, proofread first.
Does this work for Instagram Reels or Twitter/X videos? Instagram's written captions are the post's caption text, not a spoken transcript — useful for context, not a substitute for accessibility captions on a Reel's audio. X/Twitter video transcripts work like TikTok's: spoken-word audio, extracted as text.
Can I translate the captions into another language? Pro users can translate an extracted transcript without re-extracting it from the source video.
Does this create burned-in captions on the video itself? No. This extracts existing text-based captions as a file you attach or publish separately. Burning captions into the video frame, live captioning, and generating captions where none exist are different problems this tool doesn't solve.