5,000 AI-Powered Video Learning Examples: Exploratory Statistics Without Fixed Categories

Survey date: October 3, 2026

This time, rather than allocating samples by category, we collected candidates using a broad set of search terms related to AI video learning and deduplicated them by video ID. We retained only items whose titles or descriptions explicitly mentioned both AI-related terms and learning, education, or training-related terms, yielding a final set of 5,000 examples.

Key findings

• We were able to identify publisher names for 3,479 channels (one item lacked a publisher name). Even the combined total for the top 10 channels was only 298 videos, or 5.96% of the whole.
• The largest estimated category was general and cross-disciplinary content, with 989 videos (19.78%). This was followed by technical and vocational training with 775 (15.50%), primary and secondary education with 760 (15.20%), and corporate training with 743 (14.86%).
• The most common role for AI was interaction, questions, and assessment, at 1,405 videos (28.10%). Automated instructional-content production and editing accounted for 1,175 (23.50%).
• Interactive videos totaled 2,046 (40.92%).
• The median video length was 6 minutes 8 seconds. A total of 3,814 videos (76.28%) were under 15 minutes.
• In a rough language proxy based on character types, 4,908 videos (98.16%) were primarily in Latin-script languages. This is the study's largest source of bias.

Natural distribution of categories

Estimated category	Count	Share
General and cross-disciplinary	989	19.78%
Technical and vocational training	775	15.50%
Primary and secondary education	760	15.20%
Corporate training and organizational learning	743	14.86%
Educational content production and teacher support	530	10.60%
Language learning	377	7.54%
Healthcare and health	351	7.02%
University and higher education	250	5.00%
Lifelong and individual learning	225	4.50%

Unlike the previous study, which balanced categories evenly, this study preserved the volume that emerged naturally from search. Technical and vocational training, where practical procedures are easy to turn into videos; primary and secondary education, which has abundant repeatable materials; and corporate training, which is updated frequently, ranked near the top. Lifelong learning appears smaller under its dedicated label because its content is often absorbed into general and cross-disciplinary or language-learning categories.

AI's role

Primary role of AI	Count	Share
Interaction, questions, and assessment	1,405	28.10%
Other uses of AI	1,365	27.30%
Automated instructional-content production and editing	1,175	23.50%
Personalization and recommendations	355	7.10%
Translation, captions, and accessibility	333	6.66%
AI instructors, avatars, and video generation	314	6.28%
Summarization, search, and review support	53	1.06%

The large “Other” category reflects a decision not to force an assignment for examples where the title and description alone did not identify AI's specific role. Interaction, questions, and assessment ranked first, suggesting that publicly available examples make visible a substantial number of designs that use AI not merely as a production tool, but to respond to learners during or after viewing.

Interactivity

Format	Count	Share
One-way / production-side AI	2,954	59.08%
Interactive	2,046	40.92%

Interactive videos were identified from terms such as interactive, tutor, chat, question, quiz, feedback, conversation, and adaptive in titles, descriptions, and search context. These figures do not reflect a manual verification of whether the functionality was actually implemented within every video.

Video length

Length	Count	Share
Under 5 minutes	2,199	43.98%
5–15 minutes	1,615	32.30%
15–30 minutes	540	10.80%
30 minutes or more	646	12.92%

The median was 368 seconds (6 minutes 8 seconds), the first quartile was 156 seconds, and the third quartile was 848 seconds. Short instructional videos dominate, but videos lasting 30 minutes or more account for roughly 13%, forming a separate cluster of lectures, webinars, and long-form tutorials.

View scale and publisher concentration

Among the 4,959 videos with view counts available, the median was 157 views and the third quartile was 4,526 views. Rather than being dominated by a small number of hit videos, the landscape contains a large volume of small-scale videos for specific use cases. However, view counts are strongly affected by when a video was published and are not a measure of learning effectiveness.

We were able to identify publisher names for 3,479 channels (one item lacked a publisher name). Simplilearn had the most videos, with 66 (1.32%), followed by Safe AI for The Classroom with 56 (1.12%). Even the top 10 channels combined accounted for only 5.96%, with no extreme concentration around a single vendor or a small number of publishers.

Hypotheses suggested by this study

1. AI video learning is not a monolith; it divides into three layers: production, interaction, and personalization.
2. Lightweight interactivity—such as questions, quizzes, and feedback—has spread earlier than personalization.
3. The market is not centered on blockbuster learning materials; it consists of many small videos addressing narrow use cases.
4. Videos under 15 minutes are the main battleground, but long-form content remains in specialist lectures and vocational training.
5. The bias toward English and Latin-script languages is extremely strong; additional research in non-English-speaking regions is needed to discuss the global picture.

Methods and limitations

• We gathered approximately 19,800 candidate search results from public YouTube searches.
• We deduplicated videos by video ID.
• We retained only videos whose titles or descriptions themselves contained both AI-related and learning-related terms.
• Categories, AI roles, and interactivity were estimated using keywords.
• Each video was treated as one example; this is not a count of independent organizational adoptions or market share.
• The data are subject to bias from search rankings, publicly available examples, YouTube users, and English-speaking regions.
• This was not a study in which all 5,000 video bodies were watched manually.

Accordingly, these data are a map of AI video learning topics and practices observable through public search, not a random sample for estimating the full population.