Online Course Video Statistics

Key takeaways

1. 54.7% of U.S. postsecondary students took at least one distance course in fall 2024, and 26.5% took nothing else. (NCES Digest of Education Statistics) See details
2. Students enrolled exclusively at a distance grew 51.6% between fall 2019 and fall 2024. (NCES; our calculation) See details
3. Coursera learners watched approximately 657.0 million lectures in 2025, up 12% on the year before. (Coursera) See details
4. Udemy counted about 66 learners in its stated audience for every one monthly paying buyer. (Udemy; our calculation) See details
5. Median engagement time was at most 6 minutes across 6.9 million edX video sessions, regardless of video length. (Guo, Kim and Rubin, 2014) See details
6. Among enrolled Stanford students, the mean per-video median share watched was 91.3%, a different measure from a different audience. (ASEE, 2015) See details
7. One platform, one year, two correct completion rates: 3.13% of all participants and 46.02% of paid verified-track learners. (The MOOC Pivot) See details
8. 52% of people who registered for a HarvardX or MITx course never entered the courseware at all. (Reich and Ruiperez-Valiente) See details
9. 48% of U.S. faculty taught a fully online course in 2024-25, but online learning officers judged just 28% of faculty fully prepared to design one. (Bay View Analytics; CHLOE 10) See details
10. Udemy cut its instructor revenue share on subscriptions to 15%, effective January 2026. (Udemy 10-K) See details
11. The famous '25-60% retention' claim comes from a 2000 investment-bank paper that put classroom retention at 58%, not 8-10%. (WR Hambrecht + Co, 2000) See details
12. NTL Institute says it can no longer find the original research behind the learning pyramid it is credited with. (NTL Institute) See details

How many people take online courses?

In the United States, 54.7% of all postsecondary students took at least one distance education course in fall 2024, and 26.5% took nothing else, out of total enrollment of 19,659,633. The counts come from table 311.15 of the Digest of Education Statistics, published by the National Center for Education Statistics, which counts every degree-granting institution that takes part in federal student aid programs rather than sampling.

The series below is the part worth keeping. Distance enrollment did not appear in 2020 and it did not go back afterwards.

Fall 2020 is the emergency-remote term and should not be read as demand for anything. The useful comparison is 2019 against 2024: the number of students taking no campus courses at all rose 51.6% across those five years, while total enrollment was roughly flat. NCES revises these tables between editions, so a figure quoted from an older Digest will not always match the current one.

Graduate students are the heavier online population. In fall 2024, 3,888,681 undergraduates were enrolled exclusively at a distance against 1,325,644 postbaccalaureate students — but as a share of their own group that is about a quarter of undergraduates and roughly two in five graduate students. Of the exclusively online students, 1,781,956 were in a different state from the institution teaching them, which is the clearest sign that an online course is not a local product.

Outside the United States, Eurostat's household survey found 34.8% of EU internet users took an online course or used online learning material in 2025, up from 21.4% in 2019. The narrower question — did you take an online course — returns 17.3% of internet users, or 16% of everyone aged 16 to 74. The gap between those numbers is a denominator, not a disagreement, and articles quoting "a third of Europeans take online courses" have taken the broadest reading of the broadest question. India's All India Survey on Higher Education counted 44.0 lakh students in distance mode for 2023-24 and 2.8 lakh in fully online approved programmes — about 4.4 million and 280,000 respectively.

One more U.S. denominator, from the body that administers interstate online delivery: NC-SARA reported 4,657,856 exclusively distance students at participating institutions in fall 2025, up 4.9% from 4,442,246 in fall 2024, and more than 1.7 million students attending an out-of-state institution entirely online in fall 2025.

How big are the online course platforms?

Coursera reported approximately 197 million cumulative registered learners as of 31 December 2025, 54.4 million course enrollments during the year, and — the figure that matters most to anyone editing lessons — approximately 657.0 million lectures watched in 2025, up 12% on the previous year. That works out at 12.1 lectures watched per course enrollment. Its catalogue held more than 13,500 courses from over 375 content creators, sold to more than 1,700 paid enterprise customers.

Udemy's last annual report describes a library of more than 290,000 courses, a stated audience of nearly 84 million learners, more than 1 billion cumulative course enrollments, and an average of 40 million monthly unique visitors. New supply arrives at over 6,300 courses a month, in 78 languages, and instructors collectively earned $168 million in 2025.

Set those learner counts against what people actually pay for. Udemy averaged 1,275 thousand monthly buyers on its consumer marketplace in 2025 — roughly 66 counted learners for every one monthly paying buyer. Neither number is wrong; they answer different questions, and only one of them describes a customer.

Other platforms publish their own counts. Skillsoft’s Learner Platform served approximately 59 million registered learners; its separate Skills Management Platform served over 3,000 enterprise customers, and its Percipio platform held more than 431,000 learning assets. At the end of 2025, Docebo's platform was used by nearly 3,500 companies and Duolingo reported more than 130 million monthly active users. Khan Academy, a nonprofit that publishes an annual report, recorded 189.6 M registered users, 104.9 M yearly active learners and 66.8 B learning minutes across 842 courses in school year 2024-25. Registered Moodle sites reported 59,195,241 courses and 3,647,757,049 course enrolments in a snapshot taken on 14 September 2026; that is a live counter and it moves daily. MIT OpenCourseWare publishes materials from over 2,500 MIT courses and its YouTube channel has over 6 million subscribers. LinkedIn states that about 119 hours of learning content are consumed on its platform every minute, and edX states on its own site that 100M learners use it. Thinkific says its customers have earned $4.23B selling courses; that is a platform marketing claim with no date attached, and it is gross, not take-home.

The more important fact about this section is that it is getting harder to write. Coursera completed its acquisition of Udemy on 11 May 2026; Udemy was delisted the same day and deregistered on 21 May 2026, so the fiscal 2025 annual report cited above is the last one that will ever exist. Coursera's own quarterly report for the period ended 30 June 2026 no longer carries a registered-learner metric at all. 2U, which owned edX, filed for Chapter 11 in 2024 and deregistered; Instructure was taken private; Chegg has stopped disclosing subscriber counts. Class Central's annual MOOC census appears to have stopped after its 2021 edition. Several widely quoted platform numbers are now frozen at whatever the last filing said, and no newer audited figure is coming.

How long should an online course video be?

We found no research finding that names a best length. What exists is a set of platform rules, a set of platform recommendations, and one well-known 2013 measurement of how long students keep watching — which is not the same question. Take them in that order.

The rules are hard numbers and they are published by the platforms themselves. Udemy will not publish a paid course with less than 30 minutes of video or fewer than 5 separate lectures, requires 720p or 1080p at a 16:9 ratio, and caps free courses at under 2 hours. Skillshare requires a class of at least 10 minutes, describes its typical class as 20 to 60 minutes split into 2- to 8-minute lessons, and allows short-form classes from 2 minutes up to 15.

Two of those are recommendations rather than rules, and one of them carries a number worth flagging: Coursera stated in 2020 that keeping lectures under ten minutes raises a course's completion rate by 16%. Coursera sells course hosting, so that is a platform's claim about its own product and it publishes no method alongside it.

Now the measurement everyone cites. Across 6.9 million video watching sessions on edX, median engagement time was 6 minutes at most, regardless of how long the video actually ran. The dataset was 6,902,358 sessions across 862 videos and 127,839 students in four Fall 2012 maths and science courses. Engagement fell as videos lengthened: for the shortest videos, 75% of sessions ran past three quarters of the way through, and 56% of sessions were followed by an attempt at the assessment that came next, against 31% for the longest. Students engaged with tutorial videos for only 2 to 3 minutes.

A companion second-by-second analysis of the same 862 videos found that about 55.2% of viewing sessions ended before the video did, rising to 78.6% among students re-watching, and 61.3% for tutorials against lectures. Its regression predicted a 71% dropout rate for a twenty-minute video against 53% for a five-minute one.

Then the finding that almost never gets quoted alongside it. When Stanford researchers measured enrolled, fee-paying students rather than open registrants, the numbers inverted. Across 13,107 watching sessions of 28 videos from an introductory computer science course in autumn 2014, median watching time was 12.1 minutes on lectures running 50 minutes — and once a student's repeat visits to one video were stitched together, the mean of the per-video median share watched was 91.3%. Most students watched either almost all of a video or almost none of it. Short videos hold open-registration audiences; they are not a universal law of attention.

Two vendors publish length-versus-engagement benchmarks from their own hosting data. Wistia reports a 52% average engagement rate for videos under a minute, over 50% watched for educational and tutorial videos in the one-to-five-minute band, and an 11% drop once a video passes thirty minutes. For videos made in 2024, Vidyard reports 65% of viewers staying to the end under one minute against 20% past twenty minutes. Both companies sell video hosting, both measure their own customers' libraries rather than course platforms, and neither publishes a sample frame.

Three different quantities are moving through that evidence and roundups routinely blend them: engagement time is how long a session lasts, dropout is whether a session reaches the end, and watch percentage is how much of the runtime is consumed. A claim about one is not a claim about the others.

Do people finish online courses?

There is no single completion rate for online courses, and the spread between the published figures is caused almost entirely by who is counted in the denominator. In 2017-18, 3.13% of HarvardX and MITx course participants completed their course. In the same year, on the same platform, 46.02% of learners who had paid to enter the verified track completed theirs. That is a factor of 14.7 between two correct numbers.

The reason the low numbers are so low is that registration on an open platform costs nothing. Across six years of HarvardX and MITx courses, to May 2018, 52% of people who registered never entered the courseware at all, out of 12.67 million registrations. In their earlier report on 68 courses, Harvard and MIT excluded 1.3 million such registrations before calculating anything, counting as a participant only someone who opened at least one chapter. A completion percentage that includes people who never arrived is measuring the price of the sign-up button.

The first-year numbers, fall 2012 to summer 2013, show the shape clearly: 841,687 registrations produced 292,852 registrants who never engaged with any content, 35,937 who explored half or more of a course without certifying, and 43,196 certificates. The two-year sample of 64 certificate-granting courses held 1.47 million participants, and the certification rate rose to 8% in year two. Over the four years to summer 2016 the two universities recorded 4.5 million participants and 245 thousand certificates.

Attrition is front-loaded and it is mostly a first-week phenomenon. Around 50% of registrants stopped being active within the first week, and the review of 221 MOOCs found the first two weeks to be the critical window, after which the shares of active students and of students submitting assessments level out, within 3% of each other. Engagement and certification are also not the same thing: 44% of learners who explored half or more of a HarvardX or MITx course went on to certify, so most deeply engaged learners never appear in a completion statistic at all. Certification rates were 14% in the humanities against 7% in computer science. In the largest single cohort, learners who started in 2015-16, only 12% of 1.1 million unique learners took another HarvardX or MITx course the following year.

Enrollment on open platforms is also not the same commitment as enrolling at a university, which is the comparison people reach for. 61.1 percent of U.S. students who started postsecondary education in fall 2019 completed a credential within six years, falling to 34.1% for those who began part-time; NCES puts the six-year graduation rate of full-time bachelor's students who started in fall 2014 at 64 percent, and first-year retention at four-year institutions at 82 percent. At the start of the 2023-24 academic year, 43.1 million U.S. adults held some college credit and no credential. Formal education does not have a completion problem that online courses uniquely lack, nor the reverse — the two are counted differently.

One thing neither large platform discloses: neither Coursera's nor Udemy's annual report states a completion rate, completions per learner, or any persistence metric. For 2025, Coursera publishes 197 million cumulative registered learners against 54.4 million course enrollments in the year and 657.0 million lectures watched, along with 107.2 million assessments completed, and no completion rate. Any per-platform completion figure in circulation is somebody's estimate.

Who makes online course video, and are they ready to?

In the United States the supply side is large and mostly institutional. There were 3,896 degree-granting postsecondary institutions in 2022-23, of which 1,599 were public and 683 private for-profit, employing 1,530,513 faculty in fall 2023 — 670,688 of them part-time. Average salary for full-time instructional faculty on nine-month contracts was $101,925 in 2023-24.

In one national faculty survey, about half taught online. Bay View Analytics found 48% of faculty taught at least one fully online course in 2024-25, against 72% teaching at least one face-to-face course and 23% teaching blended; 65% taught in a single modality only. The survey ran in April 2025 with 3,447 faculty responding and is published under an open licence with its method attached.

Readiness is the gap. Chief online learning officers surveyed for the 2025 CHLOE 10 report judged only 28% of their faculty fully prepared for online course design and 45% fully prepared for online teaching. Institutional intent nevertheless points one way: 88% plan to expand online offerings within three years and 74% report rising interest from graduate students, even as the share offering online-specific student orientations fell to 83% from 93% in 2021.

Instructor preference is moving the other way, and it is worth stating plainly because it cuts against the institutional plans above. Tyton Partners found the share of instructors preferring to teach face-to-face rose to 64% in 2025 from 55% in 2023, with the face-to-face share of courses rising to 72% from 56%. In the UK, Jisc found 64% of higher education teaching staff rated their digital teaching environment positively and 53% rated institutional support above average, but only 10% had any formal recognition for digital teaching skills.

Two production details from the 2012-13 edX corpus are useful to anyone cutting lessons: instructors spoke at a mean of 156 wpm, ranging from 48 to 254, and 89% of the videos studied were ordinary lectures or tutorials rather than any more elaborate format.

What is the money in online course video?

Two platforms carry most of the public financial record, and both filed for fiscal 2025. Coursera reported total revenue of $757.5 million, split $502.2 million consumer and $255.3 million enterprise, at a 54.6% gross margin. It paid $271.1 million in segment content costs to the universities and companies supplying its catalogue — 35.8% of total revenue. In the consumer segment alone, content costs were 38.6% of segment revenue. Udemy reported fiscal 2025 total revenue of $789.8 million, split $524.1 million enterprise and $265.8 million consumer, at a 66% gross margin, with content costs at 32% of consumer segment revenue.

The instructor share is the number that is actually moving. Udemy cut the instructor revenue share on its subscription offerings to 15% effective 1 January 2026, down from 20% in 2024 and 17.5% in 2025. Coursera, separately, disclosed that it has met opposition to its content fee terms. Anyone building a business on course video for a marketplace should read both filings rather than either one.

Elsewhere in the sector the direction is down. Chegg's net revenues fell to $376.9 million in 2025 from $617.6 million in 2024 and $716.3 million in 2023. Skillsoft reported $512.7 million for the twelve months to 31 January 2026, a third consecutive decline, and spent $20.3 million on proprietary content development, down year over year. Docebo, which sells to employers rather than learners, grew: annual recurring revenue of $238.1 million with an average contract value of $66,500 per customer.

Employer training budgets, the other buyer of course video, are not expanding either. The UK Department for Education's employer survey puts total training spend at £53.0 billion in the year to 2024, equivalent to £2,710 per person trained and £1,700 per employee — each down from 2022. Eurostat's continuing vocational training survey put the EU average at 1 441 PPS per participant and 64 euros per training hour in 2020. On the creator side, Hotmart, which owns Teachable, announced in March 2024 that creators on its platforms had collectively exceeded $10 billion in gross merchandise volume; that is gross sales across all product types, announced by the platform, not instructor take-home from courses.

What the popular online course numbers really say

Four statistics do most of the work in articles about course video. Each one is repeated with a confident attribution, and each attribution is wrong in a way that changes what the number means. They are traced here to the earliest document we could date.

"The optimal course video is under six minutes"

The finding reached the public in autumn 2013, before the paper. Philip Guo, the researcher who ran the analysis, wrote it up on the edX blog in a post usually dated November 13, 2013, which says the findings had already appeared in a Wall Street Journal article on online courses and its infographic; a blog repeating the post the next day credits "Philip Guo's website, Oct 2013". The post states that the optimal video length is 6 minutes or shorter and that average engagement time maxed out at 6 minutes regardless of how long the video ran. Those two sentences sit in the same paragraph, and they do not agree: a ceiling on how long people watch is not an optimum, and nothing in the study compared a six-minute video against a five- or seven-minute one on any outcome.

The peer-reviewed version, published four months later, reports the same result as a median engagement time and is explicit about its limits. Its authors write that it is important not to draw any conclusions about student learning from findings about video engagement. The paper also records that edX staff were already recommending videos of less than 6 minutes before the data existed, based on their prior intuitions — the rule predates the evidence now used to justify it.

What to use instead: median engagement time was at most six minutes across 6.9 million sessions in four Fall 2012 maths and science courses on one platform, and it describes watching, not learning. The Stanford measurement of enrolled students in the section above found median watch times of twelve minutes and per-video median watch shares above 90%, which suggests the six-minute ceiling describes open-registration audiences rather than human attention.

The guideline also travels through a teaching guide. Vanderbilt University’s Center for Teaching published Cynthia Brame’s "Effective Educational Videos" as an online guide, usually cited as 2015, and university teaching centers still link it; that address now redirects to the center’s home page. Brame published the guide as a peer-reviewed essay in 2016. It organizes video design around three elements, managing cognitive load, maximizing student engagement and promoting active learning, and it lists "short videos (6 minutes or less)" as one way to manage cognitive load, citing the Guo study above (Brame, CBE—Life Sciences Education, 2016). That is a recommendation drawn from the same engagement data, not a separate measurement.

"Online learners retain 25 to 60% more, against 8 to 10% in a classroom"

This is the most-cited claim in e-learning marketing and it is wrong twice over at its own source. The earliest datable document is a March 2000 equity research white paper, Corporate e-Learning: Exploring a New Frontier, written by two investment analysts at WR Hambrecht + Co. Its actual sentence puts average content retention for an instructor-led class at 58% and says online learning enhances the retention rate by 25 – 60%.

So the classroom baseline in the origin document is 58%, not 8 to 10%, and the 25-60% figure is a relative lift, not an absolute retention rate. The two halves of the popular claim are not even measuring the same thing. The "8 to 10%" figure appears nowhere in the document it is attributed to, and we could not locate any originating source for it.

The Research Institute of America, credited almost everywhere as the author of this research, did not produce it. The white paper credits that firm only with a forgetting-curve claim, and cites three other works for it. The publisher was an investment bank that made a market in the securities of e-learning companies profiled in the same report. A 2019 investigation that chased the claim to its source concluded the supposedly recent research was at least 20 years old.

What to use instead: we found no sound measurement of comparative retention, but there is one of test results. The U.S. Department of Education’s 2010 meta-analysis of 50 effects from controlled studies found that outcomes in purely online and purely face-to-face conditions were statistically equivalent, with a mean effect size of +0.05 (p = .46). Courses that blended online and face-to-face elements did better than face-to-face alone, at +0.35, but many of those studies gave the blended group more learning time and materials, so the report does not credit the medium itself (Evaluation of Evidence-Based Practices in Online Learning, page xv). Most of the studies involved college students and adult learners, and they measured learning outcomes, not retention.

"IBM found e-learning participants learned five times more material" and "every $1 of online training returns $30"

These two lines usually arrive together, credited to a single IBM study. They come from two different IBM sales documents describing two different things. The five-times figure is about one IBM course, Basic Blue for new managers, launched in 1999 as a blend of online modules, a five-day classroom workshop and more online work. IBM’s June 2004 white paper, IBM’s learning transformation story, says studies “conducted by Harvard Business School and other organizations determined that Basic Blue enables managers to learn five times as much material at one-third of the cost of a classroom-only approach”. The footnote for that sentence is a February 2004 magazine article, not a study. The same paper’s summary describes Basic Blue as carrying “five times as much content” as IBM’s previous new-manager course. That measures how much material the course holds, not how much anyone learned, and nothing in either sentence is about time spent.

The $30 figure is not about online training, and it is not IBM’s. IBM’s 2008 white paper The value of training says that “according to a Merrill Lynch study, Motorola estimated that every dollar spent on training yielded US $30 in productivity gains within three years”, citing a Merrill Lynch report from 2000. It is Motorola’s own estimate for training of any kind, passed on by an investment bank and then by an IBM sales paper. The version most pages repeat, that “every dollar invested in online training results in $30 in productivity”, comes from SHIFT eLearning’s 2016 statistics post. That post credited both lines to “the same IBM study” and linked a 2014 IBM training paper that contains neither figure. The SHIFT post is now offline; our link goes to the Internet Archive’s copy.

What to use instead: nothing general. Basic Blue’s cost figures describe one company’s blended course for its own managers, and the $30 is one company’s unpublished estimate.

"Online courses use 90% less energy and produce 85% less CO2"

This comes from a 2005 Open University study, and the 90% is not about online learning. Robin Roy, Stephen Potter, Karen Yarrow and Mark Smith compared the environmental impact of 20 UK university courses surveyed in 2000/01 and 2002/03: ten full-time campus courses, three part-time campus courses, three mainly print-based distance courses and four distance courses with some electronic delivery. Their final report found that the distance learning courses “consumed nearly 90% less energy and produced 85% fewer CO2 emissions (per student per 10 CAT points) than the conventional campus-based university courses”. The report puts most of that down to students no longer travelling to a campus or living in term-time housing.

The online part of the comparison was small. The same report found that e-learning courses offered “only a small reduction in energy consumption and CO2 emissions (20% and 12% respectively) when compared to mainly print-based distance learning courses”, partly because students used computers heavily and printed out the web material. The large saving belongs to studying at a distance, much of it from printed course books, not to putting a course online. The report’s PDF now returns a 404 on the Open University’s server; our link goes to the Internet Archive’s copy.

What to use instead: the study itself, quoted as a distance-versus-campus comparison of 20 UK courses from the early 2000s. It says nothing about modern online video courses or about learning outcomes.

The learning pyramid: "people remember 10% of what they read, 90% of what they do"

The numbers are usually credited to the NTL Institute. NTL's own form letter to enquirers gives the figures, beginning with 10% for reading, and claims they were developed at its Bethel, Maine campus in the early sixties. The same letter concedes that NTL no longer has, and cannot find, the original research supporting them.

They are also much older than that claim. A quantified version appears in print as early as 1906, in tenths rather than percentages: "We remember one tenth of what we hear, five tenths of what we see, seven tenths of what we say, nine tenths of what we do" (Letrud and Hernes, 2018). Percentage versions appear by 1914, presented even then as long-established. A 2018 peer-reviewed analysis found versions circulating for more than 160 years and concluded the models did not originate from empirical research; a 2014 peer-reviewed review concluded the retention chart cannot be supported in terms of scientific validity or logical interpretability. Edgar Dale's Cone of Experience, frequently shown with the percentages attached, carried no numbers at all in any edition.

What to use instead: if the point is that doing beats listening, cite a measured comparison of teaching methods, not a retention chart. A 2014 meta-analysis of 225 studies of undergraduate science, engineering and mathematics courses, led by Scott Freeman at the University of Washington, found exam scores about 6% higher on average in active-learning sections, and average failure rates of 21.8% under active learning against 33.8% under traditional lecturing (PNAS, via PubMed Central). It compares ways of teaching in STEM classrooms. It does not measure how much people remember from reading, watching or doing, and it says nothing about video.

"Only 3% of people finish an online course"

Both the very low and the moderately low versions are real numbers with different denominators, as the completion section sets out. The figure usually quoted as "about 15%" comes from Katy Jordan's public MOOC dataset, which she compiled from news stories and blog posts rather than platform data, and which reports 15%. Her peer-reviewed research note on the same work gives the defensible version: a median of 12.6% across 221 MOOCs, with a range from 0.7% to 52.1%. That dataset has not been updated since 2015.

The Harvard and MIT figures show why one course produces three headlines. In a typical course across the four years to 2016, 7,900 people accessed some content after registering while 500 certified. A 2019 analysis of the same platform found completion rates had not improved across 6 years. All of these describe free registrations, which is the whole story.

What to use instead: the median of 12.6% across 221 MOOCs, always stated with its denominator.

"The human attention span is eight seconds"

This one is borrowed from elsewhere to justify short lessons. The BBC traced it to a 2015 Microsoft Canada report that did not conduct the research but cited a statistics website, whose own listed sources had no record of it. Page 6 of the report puts the average human attention span at 12 seconds in 2000 and 8 seconds in 2013, against 9 seconds for a goldfish, credited "Source: Statistic brain" (Microsoft Canada, Internet Archive capture). Microsoft’s own page for the report no longer opens. The Microsoft document drew on a survey of 2,000 people for its other findings. Attention researchers quoted in the same investigation say an average attention span is not a measurable quantity, because attention depends on the task. There is also no evidence that goldfish have short attention spans.

What to use instead: if a number is needed, use one that says what it measured. In a 2016 study, Gloria Mark of the University of California, Irvine, with researchers from Microsoft Research and MIT, logged the work computers of 40 information workers at a large U.S. technology company for about 12 working days. They stayed on one screen for a mean of 47 seconds before switching, with a median of 40.2 seconds (CHI 2016 paper, Table 1). That counts how often people switch windows at work, not how long anyone can pay attention, and it does not say how long a lesson video should be. Mark told the American Psychological Association that her team’s first measurement, timed by stopwatch in 2004, averaged two and a half minutes, and that computer logging found 75 seconds around 2012 (Speaking of Psychology, 2023). Those figures come from different groups of workers measured in different ways. If a lesson needs to be shorter, the reason is the structure of the material, not a number about goldfish.

"The global e-learning market will be worth $X billion"

Every version of this figure traces to a market-research reseller rather than to any statistical agency, filing or survey owner, and the resellers do not agree with one another. For the same market in the same base year, the published estimates we found differ roughly threefold, and the ten-year projections roughly eightfold. One figure in wide circulation is not a market size at all — it is a projected increase, requoted as a total. None of the firms publishes a method that can be checked.

No figure from those sources appears anywhere on this page, and none is reproduced here. Where a real denominator is needed, the filings and government series above supply one.

Every popular claim our research pages trace, 766 in all, is on one page: popular video statistics, fact-checked.

How we built this page

We started from the questions people search for about online course video statistics and looked for the organization that actually produced each answer. Every figure on this page sits in an evidence ledger with its source, the exact table or paragraph it came from, the measurement period, the population measured, and the date we checked it. This edition holds 201 figures from 59 publishers.

On September 22, 2026, every figure was re-fetched from its source on a cleared cache and matched against the recorded wording; all 179 sourced figures then on the page re-verified automatically, with no failures. Sources that refuse automated requests — Udemy, Skillshare, Thinkific and Coursera help documentation, and the 2013 edX blog post, all behind Cloudflare or since removed — were read from dated Internet Archive captures of the publisher's own pages, recorded as such in the evidence table. Four U.S. Bureau of Labor Statistics wage and employment figures were collected but are not published here: the public data API's daily request quota was exhausted before they could be independently re-checked on publication day, and we do not print a figure we could not verify ourselves. EDUCAUSE research, the ATD State of the Industry report and the paywalled journal version of The MOOC Pivot were left out or routed to an openly published copy rather than obtained another way.

We only cite the producer: government agencies, regulators, company filings and official posts, and the survey owner's own report. We do not cite statistics roundups, Statista-style aggregators, or salary sites without a public method. We did not bypass paywalls or sign-up forms; where a full report was gated, we used only the figures its publisher states publicly.

Figures marked "our calculation" are arithmetic on the cited numbers, and anyone can reproduce them:

This edition contains no Mark Studios customer data.

Limitations

• Most engagement research comes from one platform and one era. The video-length findings quoted everywhere, and repeated here, are drawn from four Fall 2012 maths and science courses on edX. They do not describe humanities courses, corporate training, or paying students, and the page says so where it uses them.
• Open-platform enrollment is not a commitment. Registration is free, so completion percentages calculated against registrations are measuring sign-up friction as much as learning. Every completion figure here is labelled with its denominator.
• Vendor benchmarks are vendor benchmarks. Wistia, Vidyard, Coursera and Thinkific figures come from companies selling video hosting or course platforms, measuring their own customers, usually without a published sample frame. Each is labelled.
• Platform counts are shrinking as a public record. Udemy was acquired and deregistered in May 2026, 2U and Instructure no longer file, Coursera has dropped its registered-learner metric, and Chegg has dropped subscriber counts. Several figures here are final disclosures that will not be updated.
• Live counters drift. The Moodle totals are a dated snapshot taken on 14 September 2026, not a stable figure.
• No market-size estimate appears on this page. Every version we found traces to a market-research reseller with no checkable method, and the published estimates for the same year differ roughly threefold.

How to cite this page

For a single figure, cite the original publisher named next to it, and link here if the verification trail is useful to your readers. Each figure has its own anchor, so you can link straight to it, for example https://www.markstudios.com/research/online-course-video-statistics#ORG-007. To cite the page as a whole:

Mark Studios. "Online Course Video Statistics." Last verified Sep 22, 2026. https://www.markstudios.com/research/online-course-video-statistics

@misc{markstudios_online_course_video_statistics_2026,
  author = {{Mark Studios}},
  title = {{Online Course Video Statistics}},
  year = {2026},
  month = sep,
  url = {https://www.markstudios.com/research/online-course-video-statistics},
  note = {Last verified 2026-09-22}
}

Our calculations and the evidence table are free to reuse with attribution and a link to this page. Each underlying figure remains the claim of its original publisher, and some publishers ask for a link when their figures are reused.

Sources

Download the evidence table (CSV)

1. National Center for Education Statistics (NCES), U.S. Department of Education, Digest of Education Statistics, Table 311.15. Number and percentage of students enrolled in degree-granting postsecondary institutions, by distance education participation, location of student, level of enrollment, and control and level of institution (2025-12-01). Accessed 2026-09-22.
2. National Center for Education Statistics (NCES), U.S. Department of Education, Digest of Education Statistics, Table 311.15. Number and percentage of students enrolled in degree-granting postsecondary institutions, by distance education participation, location of student, level of enrollment, and control and level of institution (2019-12-01). Accessed 2026-09-22.
3. National Center for Education Statistics (NCES), U.S. Department of Education, Digest of Education Statistics, Table 311.15. Number and percentage of students enrolled in degree-granting postsecondary institutions, by distance education participation, location of student, level of enrollment, and control and level of institution (2021-12-01). Accessed 2026-09-22.
4. National Center for Education Statistics (NCES), U.S. Department of Education, Digest of Education Statistics, Table 311.15. Number and percentage of students enrolled in degree-granting postsecondary institutions, by distance education participation, location of student, level of enrollment, and control and level of institution (2024-12-01). Accessed 2026-09-22.
5. Eurostat, Online learning increasingly popular in the EU (2026-03-19). Accessed 2026-09-22.
6. Eurostat, Digital economy and society statistics - households and individuals (Statistics Explained) (2026-03-01). Accessed 2026-09-22.
7. Ministry of Education, Department of Higher Education, Government of India, Report of the All India Survey on Higher Education (AISHE) 2023-24 (2026-07-08). Accessed 2026-09-22.
8. National Council for State Authorization Reciprocity Agreements (NC-SARA), NC-SARA Annual Data Report: Technical Report for Fall 2025 (2026-10). Accessed 2026-11-04.
9. National Council for State Authorization Reciprocity Agreements (NC-SARA), Data Executive Summary Report: Fall 2025 Exclusively Distance Education Enrollment & 2025 Out-of-State Learning Placements (11th Annual Data Report) (2026-08-01). Accessed 2026-09-22.
10. Coursera, Inc., Coursera, Inc. Annual Report on Form 10-K for the fiscal year ended December 31, 2025 (2026-02-23). Accessed 2026-09-22.
11. Udemy, Inc., Udemy, Inc. Annual Report on Form 10-K for the fiscal year ended December 31, 2025 (2026-02-19). Accessed 2026-09-22.
12. Skillsoft Corp., Skillsoft Corp. Annual Report on Form 10-K for the fiscal year ended January 31, 2026 (2026-04-07). Accessed 2026-09-22.
13. Docebo Inc., Docebo Inc. Annual Information Form for the year ended December 31, 2025 (filed as part of Form 40-F) (2026-02-27). Accessed 2026-09-22.
14. Duolingo, Inc., Duolingo, Inc. Annual Report on Form 10-K for the fiscal year ended December 31, 2025 (2026-02-27). Accessed 2026-09-22.
15. Khan Academy, Khan Academy Annual Report: SY24-25 (undated). Accessed 2026-10-03.
16. Moodle Pty Ltd, stats.moodle.org — Moodle statistics (2026-09-14). Accessed 2026-09-22.
17. MIT OpenCourseWare, About | MIT OpenCourseWare (undated). Accessed 2026-09-22.
18. LinkedIn Corporation, About LinkedIn — LinkedIn Pressroom (undated). Accessed 2026-10-29.
19. edX LLC, edX | Build new skills. Advance your career. (undated). Accessed 2026-09-22.
20. Thinkific Labs Inc., Thinkific — Online Course Platform (undated). Accessed 2026-09-22.
21. Udemy, Udemy course quality checklist (undated). Accessed 2026-09-22.
22. Udemy, Video Standards (undated). Accessed 2026-09-22.
23. Skillshare, How long should a class be? (2021-12-21). Accessed 2026-09-22.
24. Skillshare, Class Quality Guidelines (2025-08-14). Accessed 2026-09-22.
25. Axim Collaborative (Open edX / edX partner course staff documentation), 9.2.2. Create a Video - Building and Running an edX Course (undated). Accessed 2026-09-22.
26. Coursera, Building skills through online learning: 4 data-driven best practices (2020-10-01). Accessed 2026-09-22.
27. Teachable, Content File Types (2025-07-01). Accessed 2026-09-22.
28. Thinkific, Recommended Specs for Exporting Videos (undated). Accessed 2026-09-22.
29. Kajabi, Maximum video file size (undated). Accessed 2026-09-22.
30. Guo, Kim and Rubin (ACM Learning@Scale 2014), An empirical study of MOOC videos (ACM Learning@Scale 2014) (2014-03-04). Accessed 2026-09-22.
31. Kim, Guo, Seaton, Mitros, Gajos and Miller (ACM Learning@Scale 2014), Understanding In-Video Dropouts and Interaction Peaks in Online Lecture Videos (2014-03-04). Accessed 2026-09-22.
32. American Society for Engineering Education (Lagerstrom, Johanes and Ponsukcharoen, Stanford University), The Myth of the Six-Minute Rule: Student Engagement with Online Videos (2015-06-14). Accessed 2026-09-22.
33. Wistia, How to Choose the Right Marketing Video Length for Any Goal (2025-07-01). Accessed 2026-09-22.
34. Vidyard, Video in Business Benchmark Report (2025). Accessed 2026-09-22.
35. Justin Reich and José A. Ruipérez-Valiente, Massachusetts Institute of Technology (post-print in DSpace@MIT), The MOOC Pivot: From Teaching the World to Online Professional Degrees (post-print manuscript) (2019-01-11). Accessed 2026-09-22.
36. Open Praxis (International Council for Open and Distance Education), Uncovering MOOC Completion: A Comparative Study of Completion Rates from Different Perspectives (2024-08-29). Accessed 2026-09-22.
37. Harvard University (HarvardX Research Committee) and MIT Office of Digital Learning, HarvardX and MITx: Two Years of Open Online Courses Fall 2012-Summer 2014 (HarvardX Working Paper No. 10) (2015-03-30). Accessed 2026-09-22.
38. The International Review of Research in Open and Distributed Learning (IRRODL), Athabasca University Press, Massive open online course completion rates revisited: Assessment, length and attrition (2015-06-19). Accessed 2026-09-22.
39. Harvard University (HarvardX Research Committee) and MIT Office of Digital Learning, HarvardX and MITx: The First Year of Open Online Courses, Fall 2012-Summer 2013 (HarvardX Working Paper No. 1) (2014-01-21). Accessed 2026-09-22.
40. Isaac Chuang (MIT) and Andrew Ho (Harvard University), HarvardX Working Paper series, HarvardX and MITx: Four Years of Open Online Courses -- Fall 2012-Summer 2016 (SSRN abstract page) (2016-12-23). Accessed 2026-09-22.
41. National Student Clearinghouse Research Center, Yearly Progress and Completion (2025-12-04). Accessed 2026-09-22.
42. U.S. National Center for Education Statistics (NCES), Institute of Education Sciences, Fast Facts: Undergraduate graduation rates (40) (2022-05-31). Accessed 2026-09-22.
43. U.S. National Center for Education Statistics (NCES), Institute of Education Sciences, Condition of Education: Undergraduate Retention and Graduation Rates (2022-05-01). Accessed 2026-09-22.
44. National Student Clearinghouse Research Center, Some College, No Credential Student Outcomes: 2025 Report for the Nation and the States (2025-06-04). Accessed 2026-09-22.
45. U.S. Department of Education, National Center for Education Statistics, Digest of Education Statistics, Table 317.10. Degree-granting postsecondary institutions, by control and level of institution: Selected academic years, 1949-50 through 2022-23 (2024-08). Accessed 2026-09-22.
46. U.S. Department of Education, National Center for Education Statistics, Digest of Education Statistics, Table 315.10. Number of faculty in degree-granting postsecondary institutions, by employment status, sex, control, and level of institution: Selected years, fall 1970 through fall 2023 (2025-01). Accessed 2026-09-22.
47. U.S. Department of Education, National Center for Education Statistics, Digest of Education Statistics, Table 316.10. Average salary of full-time instructional faculty on 9-month contracts in degree-granting postsecondary institutions, by academic rank, control and level of institution, and sex: Selected years, 1970-71 through 2023-24 (2025-01). Accessed 2026-09-22.
48. Bay View Analytics, Deeply Digital: Educational Resources in U.S. Higher Education, 2025 (2025-09-16). Accessed 2026-09-22.
49. Quality Matters, Encoura Eduventures Research and EDUCAUSE (CHLOE 10), 2025 CHLOE 10 Report - CHLOE 10 Meeting the Moment: Navigating Growth, Competition, and AI in Online Higher Education (2025-08-12). Accessed 2026-09-22.
50. Quality Matters, Encoura Eduventures Research and EDUCAUSE (CHLOE 10), Nearly 9 in 10 Colleges Plan to Expand Online Programs as Student Demand Soars, New Report Finds (2025-08-12). Accessed 2026-09-22.
51. Tyton Partners, Time for Class 2025: Empowering Educators, Engaging Students (2025-06-11). Accessed 2026-09-22.
52. Jisc, 2024/25 UK higher education teaching staff digital experience insights survey findings (2025). Accessed 2026-09-22.
53. Chegg, Inc., Chegg, Inc. Annual Report on Form 10-K for the fiscal year ended December 31, 2025 (2026-03-09). Accessed 2026-09-22.
54. Docebo Inc., Docebo Announces Fourth Quarter and Fiscal Year 2025 Results (press release furnished on Form 6-K) (2026-02-27). Accessed 2026-09-22.
55. Department for Education (UK), Employer skills survey, calendar year 2024 (2025-07-24). Accessed 2026-09-22.
56. Eurostat, Statistics on continuing vocational training in enterprises - Statistics Explained (2022-11). Accessed 2026-09-22.
57. Hotmart Company (Teachable), Hotmart Company, Home to Teachable, Announces Record-Breaking $10 Billion in Global Creator Earnings (2024-03-21). Accessed 2026-09-22.
58. edX (post authored by Philip J. Guo), Optimal Video Length for Student Engagement (2013-11-13). Accessed 2026-09-22.
59. CBE—Life Sciences Education (American Society for Cell Biology), Effective Educational Videos: Principles and Guidelines for Maximizing Student Learning from Video Content (2016). Accessed 2026-10-17.
60. WR Hambrecht + Co (Trace A. Urdan and Cornelia C. Weggen), Corporate e-Learning: Exploring a New Frontier (2000-03-01). Accessed 2026-09-22.
61. Learning Accelerators (Adrian Snook), RIA Window - Learning Accelerators (2019-11-25). Accessed 2026-09-22.
62. IBM Global Services, IBM's learning transformation story (2004-06). Accessed 2026-10-20.
63. IBM, The value of training (IBM Training white paper) (2008-10). Accessed 2026-10-20.
64. SHIFT eLearning, Facts and Stats That Reveal The Power Of eLearning [Infographic] (2016-04-07). Accessed 2026-10-20.
65. The Open University (Design Innovation Group), Towards Sustainable Higher Education: Environmental impacts of campus-based and distance higher education systems (2005-03). Accessed 2026-10-20.
66. NTL Institute for Applied Behavioral Science (document hosted by Work-Learning Research), Thanks for your inquiry of NTL Institute (NTL's learning pyramid letter) (2006-11-03). Accessed 2026-09-22.
67. Work-Learning Research (Will Thalheimer), quoting NTL Institute, NTL continues its delusions (2006-11-03). Accessed 2026-09-22.
68. Learning Accelerators (Adrian Snook), Moonrakers! - Learning Accelerators (2019-12-05). Accessed 2026-09-22.
69. Letrud and Hernes, Cogent Education (Taylor & Francis), Excavating the origins of the learning pyramid myths (2018). Accessed 2026-09-28.
70. Work-Learning Research (Will Thalheimer), Mythical Retention Data & The Corrupted Cone (2015-01-05). Accessed 2026-09-22.
71. Kåre Letrud and Sigbjørn Hernes, Cogent Education (record via ERIC, U.S. Department of Education), Excavating the Origins of the Learning Pyramid Myths (2018-01-01). Accessed 2026-09-22.
72. Katy Jordan, MOOC Completion Rates: The Data (2015-06-12). Accessed 2026-09-22.
73. Harvard University (Harvard Gazette), reporting the HarvardX/MITx four-year report, Grading the MOOCs (2017-01-12). Accessed 2026-09-22.
74. MIT Teaching Systems Lab (Justin Reich and José A. Ruipérez-Valiente), The MOOC pivot (2019-01-11). Accessed 2026-09-22.
75. BBC News (More or Less, BBC World Service), by Simon Maybin, Busting the attention span myth (2017-03-10). Accessed 2026-09-22.
76. Microsoft Canada (Consumer Insights), Attention spans, Spring 2015 (Internet Archive capture) (2015). Accessed 2026-10-28.
77. Gloria Mark (University of California, Irvine), Shamsi T. Iqbal, Mary Czerwinski, Paul Johns and Akane Sano, ACM CHI 2016, Neurotics Can’t Focus: An in situ Study of Online Multitasking in the Workplace (2016-05). Accessed 2026-10-28.
78. American Psychological Association (Speaking of Psychology, episode 225), Why our attention spans are shrinking, with Gloria Mark, PhD (2023-02-06). Accessed 2026-10-28.
79. U.S. Department of Education, Office of Planning, Evaluation, and Policy Development (Barbara Means and others, SRI International), Evaluation of Evidence-Based Practices in Online Learning: A Meta-Analysis and Review of Online Learning Studies (revised 2010-09). Accessed 2026-10-28.
80. Scott Freeman and others (University of Washington), Proceedings of the National Academy of Sciences, via PubMed Central, Active learning increases student performance in science, engineering, and mathematics (2014-06-10). Accessed 2026-10-28.

Update policy and change log

We re-verify every figure quarterly and when a major source publishes a new edition. The next scheduled review is December 2026. Corrections go to contact@markstudios.com.

• Sep 22, 2026: First published with 184 figures from 51 publishers, each checked against its original source.
• Sep 28, 2026: The six-minute video finding was public before November 2013: Philip Guo's edX post says it had already appeared in a Wall Street Journal article, and a repeat of the post credits his website in October 2013. We had called the November 2013 post the earliest datable document.
• Sep 28, 2026: The learning pyramid's numbers go back further than 1914: a 2018 peer-reviewed study found a version in tenths published in 1906.
• Oct 3, 2026: Khan Academy took its SY24-25 annual report offline (the address now redirects to its homepage). The four Khan Academy figures now link to an Internet Archive capture of the report from September 23, 2026, re-checked word for word.
• Oct 17, 2026: Added Cynthia Brame’s "Effective Educational Videos" to the six-minute section: Vanderbilt’s online guide now redirects, and its peer-reviewed 2016 version recommends videos of 6 minutes or less, citing the same edX engagement study.
• Oct 20, 2026: Traced "IBM found e-learning participants learned five times more material" to IBM’s 2004 white paper on its Basic Blue course for new managers, where it describes how much material that one blended course holds, and "every $1 of online training returns $30" to IBM’s 2008 white paper, where it is Motorola’s estimate for training of any kind, via a 2000 Merrill Lynch report. SHIFT eLearning’s widely linked post, now offline, combined both as one IBM study.
• Oct 20, 2026: Traced "online courses use 90% less energy and produce 85% less CO2" to the Open University’s 2005 study of 20 UK courses, where the 90% compares distance learning (mostly printed course books) with full-time campus courses, and e-learning saved only 20% of energy and 12% of CO2 against print-based distance courses. The report’s PDF is now a 404; the section links the Internet Archive’s copy.
• Oct 28, 2026: Linked the BBC investigation and an Internet Archive capture of Microsoft Canada’s 2015 attention spans report, whose page 6 credits the 8-second figure (12 seconds in 2000, 9 for a goldfish) to Statistic Brain. Microsoft’s own page for the report no longer opens.
• Oct 28, 2026: The eight-second section now gives a measured figure to use instead: a mean of 47 seconds on one screen before switching, logged on the work computers of 40 information workers in a 2016 University of California, Irvine and Microsoft Research study. It measures window switching at work, not attention span. We had said there was nothing to use instead.
• Oct 28, 2026: Two more sections now give a measured figure to use instead. The e-learning retention claim points to the U.S. Department of Education’s 2010 meta-analysis, where purely online and purely face-to-face outcomes were statistically equivalent. The learning pyramid points to the 2014 PNAS meta-analysis of 225 STEM course studies, where failure rates were 21.8% under active learning against 33.8% under lecturing.
• Oct 29, 2026: LinkedIn’s About page now says about 119 hours of learning content are consumed on LinkedIn every minute, down from about 123 when we published, so the platform section shows 119.
• Oct 30, 2026: Corrected two readings. After the first two weeks, the review of 221 MOOCs found the shares of active students and of students submitting assessments level out within 3% of each other; we had called that 3% a weekly loss. And Coursera’s $271.1 million in content costs is 35.8% of its total revenue: the 38.6% is the consumer segment’s own content costs as a share of consumer revenue. Also scoped several lines to their sources (Udemy’s revenue-share steps, Skillsoft’s separate platforms, the Bilgeİş completion bases, and the 2025 CHLOE and Tyton surveys).
• Oct 31, 2026: Added dates and scopes some sentences were missing: Docebo’s nearly 3,500 companies and Duolingo’s 130 million monthly users are end-of-2025 counts; Coursera’s 16% completion claim is from 2020; Vidyard’s length benchmark covers 2024 videos; the Stanford sessions are from autumn 2014; the HarvardX and MITx figures now carry their periods (first year fall 2012 to summer 2013, four years to summer 2016, six years to May 2018, the 2015-16 cohort); the 43.1 million adults with some college and no credential is as of 2023-24; Coursera’s learner and enrollment counts and Udemy’s revenue are for 2025; Eurostat’s training costs are EU averages; and CHLOE 10 is the 2025 report.
• Nov 4, 2026: Moved the NC-SARA exclusively-distance count to its fall 2025 report: 4,657,856 students, up 4.9%.