YouTube datasets
The data behind Viewershift, available for research, analytics and AI work. Each dataset says what it holds and, just as plainly, what it does not.
Channels
Every YouTube channel we have recorded, with its profile and latest public counts.
Fields
- Channel ID, name and handle
- Subscriber, view and video counts
- Country, creation date and last upload
- Main category and spoken language
- Avatar and banner images, creator keywords
- Rank by category, country, language and size
Coverage
- About 250 million distinct channels, about 4.9 million of them with 10,000 or more subscribers.
- For channels with 10,000+ subscribers: creation date on about 97%, country on about three quarters.
- Subscriber counts are as YouTube shows them, rounded to three significant figures.
Videos
Video records with their metadata and latest public counts.
Fields
- Video ID, channel, title
- View count, likes and comments
- Duration, Shorts and live flags
- Upload date (exact, or approximate where only relative text is known)
- Category and language
Coverage
- About 8 billion distinct videos; about 1% of stored records are repeat versions of the same video.
- View count on about 98% of videos and duration on about 80%. An exact upload date on about 43%, a date of either kind on about 94%.
- Likes, comments, category and language on roughly a tenth of videos, mostly the more popular ones.
Video view-count history
Repeated view-count readings of individual videos, for growth and trend analysis.
Fields
- Video ID
- Reading date
- View count at that reading
Coverage
- Recorded only for videos with 1,000 or more views; videos below that have no series.
- Readings arrive every few days, not daily: the median gap is 4 days. The median video has 2 readings; one in ten has 10 or more.
- Roughly 100 million videos have two readings within the last two weeks.
- History: Since January 2026.
Channel subscriber and view history
Channel-level time series: subscribers over years, channel views over months.
Fields
- Channel ID
- Reading date
- Subscriber count
- Channel view count
Coverage
- Subscriber history for about 98% of channels with 100,000+ subscribers, most with 50 or more readings, back to late 2018.
- Channel view counts since January 2026, read about once a week.
- Subscriber counts are rounded by YouTube, so they show the long trend, not day-to-day changes.
- History: Subscribers since November 2018; channel views since January 2026.
Captions and transcripts
Full caption text with timings, manual and automatic, in many languages.
Fields
- Video ID
- Language
- Manual or automatic
- Timed caption text (SRT)
- Which caption languages each video lists
Coverage
- About 440 million videos with at least one stored transcript, about 5% of all videos, and a third to a half of videos above 10,000 views.
- Most are a single automatic track; manual captions span about 300 languages.
- A typical transcript is about 14 KB of text.
Audio tracks and dubbing
Which audio languages a video offers, including YouTube's automatic dubbing.
Fields
- Video ID
- Audio language
- Original, dubbed or automatically dubbed
Coverage
- Collected since 9 September 2026, so far for about 18 million videos (a small share of the catalogue).
- History: Since September 2026.
Descriptions, chapters, tags and heatmaps
The extra detail of a video page.
Fields
- Video descriptions
- Chapters with titles and times
- Creator tags
- Most-replayed heatmaps
Coverage
- Descriptions for about 2.4 billion videos, tags for about 390 million, chapters for about 52 million.
- Heatmaps for about 92 million videos; YouTube shows them only on popular videos, so they are concentrated above 100,000 views.
Coverage measured 9 Oct 2026 on deduplicated records, rounded. Live counts are stored records, read 15 Nov 2026. Collection continues every day, so counts grow.
Delivery
Bulk files
Parquet or CSV extracts of whole datasets or a defined slice, delivered to your storage, once or on a schedule.
API access
Keyed access to channel, video and history endpoints, with rate limits agreed per contract.
Custom extracts
A list of channels, a category, a language or a market, joined across datasets and prepared for your use.
Start with a sample
Tell us which datasets and which slice you need: a set of channels, a language, a time window. We prepare a sample by hand so it matches what you plan to use, and reply within one working day.