Which Languages Have the Best YouTube Subtitle Coverage? We Checked 201,104 Videos
“Does this video have subtitles in my language?” is one of those questions everyone answers from gut feeling. YouTube publishes a list of languages its automatic captions support — around 70 of them — but a support list tells you what is possible, not what you actually get when you paste a URL into a transcript tool.
We can answer the second question, because we do it a few hundred thousand times a month. Over the 30 days from July 20 to August 19, 2026, youtube-transcript.ai resolved caption tracks for 201,104 unique YouTube videos. Here is what those videos actually carried.
The short answer
If your language is English, Spanish, Korean, Hindi, Japanese, Russian, Portuguese, Arabic, German, Vietnamese, French, Italian, Turkish, Thai, Indonesian or Polish, subtitle coverage on YouTube is effectively solved — not because creators write captions, but because YouTube's speech recognition covers those languages well and runs on almost everything.
If your language is Mandarin Chinese, the picture is strange and specific: you almost always get a human-uploaded track or nothing at all. And if your language sits outside roughly the top 25, coverage becomes a genuine coin flip.
How we measured this
Every time someone asks our tool for a transcript, we record what YouTube returned: how many caption tracks the video has, which track was used, and whether that track was auto-generated (YouTube marks these with an a- prefix, e.g. a-en) or uploaded by a human (en).
Two honest caveats before the numbers, because they change how you should read them:
- This is a demand-weighted sample, not a random sample of YouTube. These are videos real people wanted transcripts of. So the language ranking reflects what the world asks for, not YouTube's full corpus.
- The auto-vs-human numbers, however, are clean. For the 80% of videos that carry exactly one caption track, the track we report is the only track available — there is nothing for our language-preference logic to bias. Every “machine vs human” figure below is computed on that subset.
Finding 1: one video in eight has no captions at all
Of the 201,104 videos, 25,010 returned no caption track of any kind — 12.4%. That is the true floor on subtitle coverage, and no language choice can rescue it.
The usual reasons a video comes back empty:
- No intelligible speech — music videos, ambient footage, gameplay with no commentary.
- Very short clips — YouTube often skips captioning Shorts-length content.
- Captions disabled by the creator — a per-video setting that overrides everything.
- A language YouTube's speech recognition does not handle, or audio so accented or noisy that it gives up.
- Livestreams whose replay track has not finished processing yet.
Finding 2: “multilingual subtitles” is the exception, not the rule
Among videos that did have captions, 80% carried exactly one caption track. Only 20% had two or more, and the long tail — videos with 20, 30, 40 language tracks — is a few hundred videos, almost all of them from large media brands and institutional channels that pay for localization.
| Caption tracks on the video | Share of videos with captions |
|---|---|
| Exactly 1 | 80.0% |
| 2 | 10.7% |
| 3–5 | 2.0% |
| 6 or more | 7.3% |
This matters more than it looks. When YouTube's player offers you subtitles “in your language” on a video that only has one real track, what you are being offered is usually an auto-translation of that track — generated on the fly, not authored. More on that below.
Finding 3: almost nobody writes captions any more
Among single-track videos, 96.3% of the tracks were machine-generated by YouTube's speech recognition. Just 3.7% were uploaded by a human.
That is the single most useful fact in this whole analysis. “Subtitle coverage” on YouTube in 2026 is, for practical purposes, a question about which languages Google's speech recognition is good at — not about creator effort.
Finding 4: the actual language table
Here is the distribution of the one caption track that single-track videos carry. The last column is the share of that language's tracks a human actually wrote.
| Language | Code | Videos | Share | Human-made |
|---|---|---|---|---|
| English | en | 83,565 | 60.2% | 2.3% |
| Spanish | es | 7,954 | 5.7% | 0.4% |
| Korean | ko | 7,306 | 5.3% | 0.3% |
| Hindi | hi | 6,985 | 5.0% | 0.0% |
| Japanese | ja | 6,747 | 4.9% | 0.8% |
| Russian | ru | 3,661 | 2.6% | 0.0% |
| Portuguese | pt | 3,397 | 2.4% | 0.0% |
| Chinese (all variants) | zh | 2,826 | 2.0% | 99.8% |
| Arabic | ar | 2,820 | 2.0% | 0.0% |
| German | de | 2,155 | 1.6% | 3.1% |
| Vietnamese | vi | 1,921 | 1.4% | 1.1% |
| French | fr | 1,905 | 1.4% | 0.9% |
| Italian | it | 1,240 | 0.9% | 0.0% |
| Turkish | tr | 966 | 0.7% | 0.0% |
| Thai | th | 929 | 0.7% | 0.0% |
| Indonesian | id | 757 | 0.5% | 0.0% |
| Polish | pl | 634 | 0.5% | 0.0% |
| Bengali | bn | 429 | 0.3% | 0.0% |
| Dutch | nl | 356 | 0.3% | 0.0% |
| Hebrew | iw | 286 | 0.2% | 0.0% |
| Cantonese | yue | 246 | 0.2% | 6.5% |
| Ukrainian | uk | 203 | 0.1% | 0.0% |
A few things worth pulling out of that table:
- English is not merely first, it is the entire first tier. At 60% it outweighs every other language combined by a factor of 1.5.
- Korean and Hindi punch far above their share of global YouTube uploads — both outrank Japanese, Russian, German and French here. That is a demand signal as much as a supply one: Korean content and Indian-language content are what people most want to read in translation.
- Human-written captions have essentially vanished everywhere except Chinese. For Hindi, Russian, Portuguese, Arabic, Italian, Turkish, Thai, Indonesian, Polish, Bengali, Dutch and Hebrew, the human share of single-track videos rounds to zero.
Finding 5: Chinese is the odd one out
Chinese breaks the pattern hard enough to deserve its own section. Out of 201,104 videos:
- 7 videos returned an auto-generated Mandarin track (
a-zh-Hantora-zh-Hans). - 2,819 videos returned a human-uploaded Chinese track (
zh,zh-TW,zh-Hans,zh-Hant,zh-CN,zh-HK). - Meanwhile Cantonese auto-captions work fine —
a-yueshowed up on 230 videos.
Chinese appears on YouTube's own list of supported auto-caption languages, but at our volume Mandarin auto-captions are effectively not being produced. The consequence for anyone working with Chinese-language video: if the creator did not upload captions, there is nothing to extract. That is a large part of why Chinese-language channels so often ship burned-in or manually uploaded subtitles — they have to.
Auto-captions vs. auto-translate: the distinction that trips everyone up
YouTube has two separate features that both look like “subtitles in your language”:
- Automatic captions. Speech recognition run on the audio, in the language actually spoken. YouTube supports roughly 70 languages here. This is a real, downloadable track.
- Auto-translated subtitles. Machine translation applied on top of an existing track, offered into 100+ languages. This is generated at view time and is derived, not authored.
If you watch an English video with Tamil subtitles turned on, you are usually reading machine translation of machine transcription — two lossy steps stacked. Recognition errors get faithfully translated into confident nonsense. It is why auto-translated captions so often feel almost right and then say something absurd.
Coverage tiers: what to actually expect
Tier 1 — assume captions exist
English, Spanish, Korean, Hindi, Japanese, Russian, Portuguese, Arabic, German, Vietnamese, French, Italian. High volume, reliable automatic captioning, present on ordinary videos from ordinary creators. If a video in these languages has no captions, the cause is almost always the audio — music, no speech — rather than the language.
Tier 2 — usually fine, occasionally not
Turkish, Thai, Indonesian, Polish, Bengali, Dutch, Hebrew, Cantonese, Ukrainian, Telugu, Persian, Tamil, Malayalam, Filipino, Romanian, Hungarian, Czech, Greek. Automatic captioning works, but volume is thin enough that short or noisy videos drop out more often. Worth trying; do not build a workflow that assumes success.
Tier 3 — a coin flip
Mandarin Chinese (human-uploaded or nothing), plus the long tail: Amharic, Punjabi, Mongolian, Burmese, Kannada, Swahili, Bulgarian, Uzbek, Azerbaijani, Sinhala, Nepali and similar. Each of these appeared on fewer than 100 videos in a full month of traffic. Captions do exist — just plan for the case where they do not.
How to check a specific video
You do not have to guess. Paste any YouTube URL into youtube-transcript.ai and the result lists every caption track the video has, with the language name and whether it is auto-generated. That tells you in one step what YouTube's player only hints at:
- Whether the video has captions at all.
- Which language was actually spoken — the automatic track is always in the source language.
- Whether a human wrote the track. Prefer it when one exists: human captions carry punctuation, speaker labels and correct proper nouns.
- What else is on offer, so you can pick a real track instead of an auto-translation.
Check any video's caption languages in seconds
See every available track, pick the one you want, and copy the full text — free, no sign-up.
Get the Transcript FreeFrequently Asked Questions
Q: Which languages have the best YouTube subtitle coverage?
English by a wide margin — 60% of single-track videos in our sample carried an English track. After that, Spanish, Korean, Hindi, Japanese, Russian, Portuguese, Arabic, German, Vietnamese and French all appear at real volume with reliable automatic captioning. Coverage thins noticeably past roughly the top 25 languages.
Q: What percentage of YouTube videos have no subtitles at all?
About 12.4% — 25,010 out of 201,104 videos over 30 days returned no caption track. Roughly one video in eight. Music, very short clips, creator-disabled captions and unsupported or noisy audio are the main causes.
Q: Are YouTube subtitles written by humans or machines?
Machines, almost entirely. Among videos with exactly one caption track, 96.3% of tracks were auto-generated by speech recognition and only 3.7% were human-uploaded. Chinese is the one large exception, where the ratio is reversed.
Q: Why can't I get auto-captions for a Chinese video?
Because in practice YouTube does not appear to generate them. We saw 7 auto-generated Mandarin tracks across 201,104 videos, versus 2,819 human-uploaded Chinese tracks. If the creator did not upload subtitles, there is no track to extract — you would need to transcribe the audio directly.
Q: Are auto-translated subtitles as good as native ones?
No. Auto-translated subtitles are machine translation layered on top of machine transcription, so recognition errors get translated into fluent-sounding mistakes. Extract the transcript in the spoken language and translate that text with an AI assistant instead — one lossy step rather than two, and you can check the source.
Q: How was this data collected?
From caption-track metadata on 201,104 unique YouTube videos processed by youtube-transcript.ai between July 20 and August 19, 2026. It is a demand-weighted sample — videos people asked for — so language rankings reflect real-world demand. The machine-versus-human ratios are computed only on videos carrying a single caption track, where no track-selection logic applies.