Video Thumbnail and CTR Statistics

Key takeaways

1. YouTube says half of all channels and videos have an impressions CTR between 2% and 10%. It publishes no average. (YouTube Help) See details
2. The 2% to 10% range was on YouTube's help page by March 2018, the month the metric launched, and has not been updated since. (YouTube) See details
3. In YouTube's own example, CTR falls from 9% to 3.5% as impressions grow tenfold, while views from impressions rise 3.9 times. (YouTube; our calculation) See details
4. YouTube counts an impression only when a thumbnail is on screen for more than 1 second with at least 50% of it visible. (YouTube) See details
5. When YouTube first added watch time to its recommendations in 2012, it saw an immediate 20% drop in views. (YouTube) See details
6. YouTube's A/B test compares up to 3 titles and thumbnails for up to 2 weeks and picks the winner by watch time, not CTR. (YouTube) See details
7. Creators had run more than 40 million title and thumbnail experiments on YouTube by September 2026. YouTube has not said how many found a winner. (YouTube) See details
8. YouTube's recommended thumbnail size went from 1280x720 to 3840 x 2160 in March 2026, and desktop uploads can now be up to 50 MB. (YouTube) See details
9. Netflix's 82% is the share of members' browsing focus that went to artwork in its 2014 research, and its 1.8 seconds is the average time spent per title. (Netflix) See details
10. In Netflix's 2026 artwork-personalization tests, a unified artwork model lifted the core discovery metric by +0.127%, and a holdback during the launch of a new TV interface showed +0.184% more streaming hours. (Netflix researchers) See details
11. Videos on homepages hosted by Wistia averaged a 24% play rate in 2025. (Wistia) See details
12. The popular claim that faces get 38% more clicks comes from a 2014 study of likes on Instagram photos, not clicks on YouTube. (Georgia Tech and Yahoo Labs) See details
13. "90% of the best-performing videos have custom thumbnails" dates to YouTube Creator Academy in 2015 and has never had a definition or a sample. (YouTube) See details

What click-through rate is normal on YouTube?

YouTube says half of all channels and videos on YouTube have an impressions click-through rate (CTR) between 2% and 10%, in its Help Center FAQ on impressions. It is the only CTR benchmark YouTube publishes. It describes where half of channels and videos fall. It is not an average: YouTube gives no mean, no median, no "good" CTR, and no method or date for the range.

YouTube attaches two caveats. New videos or channels less than a week old, and videos with fewer than 100 views, can see an even wider range. The same page says CTR varies with the type of content, the audience and where on YouTube the thumbnail was shown.

The range has not been updated since the metric launched. YouTube announced impressions and impressions click-through rate on March 7, 2018, in a YouTube Official Blog post. The sentence giving the range, between 2% and 10%, was already on its help page in an Internet Archive capture dated March 14, 2018, one week later, and reads the same today.

A falling CTR can mean a video is growing

CTR usually drops as a video reaches people who do not already follow the channel. In a worked example on a YouTube help page on CTR and impressions, a video gets 10,000 impressions in its first period, mostly among loyal fans, and its CTR "sits at a high" 9%. A week later it has 100,000 impressions and its CTR has settled at 3.5%, which YouTube calls a sign of success rather than a decline. By our arithmetic the views the video gets from impressions rose 3.9 times, from 900 to 3,500, while its CTR fell by more than half. The example is illustrative, not measured data.

The YouTube Official Blog made the same point in September 2025: when a video goes viral its CTR can drop while views rise, because 3% of a million impressions is a lot more than 10% of a thousand. YouTube also says search results tend to bring fewer impressions and a higher CTR than the homepage, where viewers are browsing rather than looking for something specific.

What does YouTube count as an impression?

YouTube counts an impression when a thumbnail is shown for more than 1 second with at least 50% of it visible on screen, according to its help page on impressions. Impressions are counted on YouTube's own apps and site: search, the homepage, feeds, "Up next", and playlists. The same page lists what is not counted, including embeds and links on external websites and apps, the YouTube mobile website, the YouTube Kids and YouTube Music apps, cards and end screens inside the player, email and notifications, and videos playing in a background tab.

The rule used to be softer. The March 2018 version of the page said YouTube "may count" a thumbnail shown for more than one second with half of it visible, per an Internet Archive capture. YouTube's Reporting API documentation now defines video thumbnail impressions with the same thresholds, more than 1 second and at least half visible, and on January 15, 2026 the API added reach reports that give channels bulk impressions and CTR data.

The 2026 view count change did not change CTR

Beginning August 24, 2026, YouTube counts a public view the moment a video starts to play, across Shorts, long-form videos and live streams, according to its help page on the change. YouTube's developer revision history, in an entry dated August 27, 2026, says click-through rate and average view duration stay anchored on engaged views, and that the definition of a thumbnail impression is unchanged. So a video's public view count divided by its impressions should not be expected to match the CTR shown in YouTube Studio.

Does YouTube rank videos by click-through rate?

Not by clicks alone, according to YouTube. In a September 2021 post on its recommendation system, YouTube's vice president of engineering wrote that YouTube learned in 2011 that a click on a video did not mean the viewer watched it, and that it added watch time as a signal in 2012. When it did, it saw an immediate 20% drop in views. The same post says the system learns from over 80 billion signals, clicks among them, and that YouTube counts as "valued watchtime" only time spent on videos that surveyed viewers rate four or five stars out of five.

YouTube's 2012 announcements set out the change as it happened:

YouTube publishes no weights for its signals, so no public source can say how much CTR counts. Its own tools point the same way as the 2012 change: YouTube's thumbnail test picks a winner by watch time (see the next section), and YouTube Studio judges a video's opening against the channel itself. Its retention card treats a typical intro as about 30 seconds long, lists a video as "above typical" when 50% or more of viewers are still watching at the 30-second mark, and defines typical retention from the channel's last 10 videos of similar length, per its help page.

How does YouTube's thumbnail test pick a winner?

By watch time, not click-through rate. YouTube's A/B testing tool lets a creator test up to 3 titles and thumbnails on one video, then shows the option with the highest watch time to all viewers, according to its help page on testing titles and thumbnails. YouTube says it optimizes tests for overall watch time over other metrics, like click-through rate. A test can take a few days or up to 2 weeks, depending on impressions and how recently the video was published, and the help page says it is normal for a test to end without a winner.

YouTube's Creator Liaison gave the reason in October 2023: "If A/B testing returned CTR, creators would focus on CTR, and we'd very likely see a rise in click-bait and a drop in performance." YouTube has not published data showing that CTR-judged tests produce more clickbait, and it has never published the share of tests that find a winner.

The two usage counts cannot be read as growth. YouTube dates the launch to 2023 in one post and to 2024 in the other, and counts "uses" of thumbnail testing in the first and experiments on titles and thumbnails in the second. Dynamic thumbnails also drop the idea of a single winning thumbnail: different audience segments are shown different options.

Two rules limit the tool. Shorts are not eligible, and YouTube counts uploads up to three minutes long that are square or vertical as Shorts. And if any thumbnail in a test is below 720p (1280 x 720), YouTube downscales every thumbnail in that test to 480p (854 x 480).

How much confidence does each testing tool need to name a winner?

Each tool sets its own bar, from 65 percent confidence for a Meta A/B test to 95% for a conclusive Google Ads video experiment, and YouTube's creator tool publishes no confidence level at all. The ad platforms' tools test ad creative, including video and thumbnails, against ad outcomes; YouTube's tool tests an organic video's title and thumbnail against watch time. A "winner" means something different in each.

Google has published what successful creative tests were worth on YouTube ads. In global studies from 2019 and 2020, advertisers who successfully used video experiments saw a 30% lower median cost per acquisition from the better-performing creative for lower-funnel goals, and 60% higher ad recall for upper-funnel goals, per Google. Both figures count only successful experiments, so they describe the gain among winners, not the average test.

The only published win rate we found for YouTube thumbnail tests comes from a vendor. NoteLM.ai, which sells creator tools, reported that 89 of 127 tests across 15 channels (70%) produced a clear winner between August 2025 and January 2026, and that tests with a winner averaged a +37% CTR improvement, per its own write-up. Its tests ran one version after another rather than at the same time and were judged on CTR, the method YouTube's own tool avoids.

What size should a YouTube thumbnail be?

YouTube now recommends custom thumbnails of 3840 x 2160 pixels for videos and 2160 x 3840 for Shorts, according to its help page on video thumbnails, read on September 24, 2026. Until March 2026 the same page recommended 1280x720 (see the change history below).

A vertical video with a 16:9 custom thumbnail is shown on the Home, Explore and Subscriptions pages with an auto-generated 4:5 thumbnail instead, the same page says.

How YouTube's thumbnail spec changed in 2026

The spec most guides still quote, 1280x720 and under 2 MB, was correct until March 2026. Internet Archive captures of YouTube's help page and YouTube's own API changelog date the change:

By our arithmetic, the recommended size now has 9 times the pixels of the old one, and the desktop file-size limit is 25 times the old 2 MB. Uploads from a phone are still capped at 2 MB for videos.

To check a thumbnail against these numbers and preview it on the home feed, in search and on a phone, use our free YouTube thumbnail checker. It runs in your browser and never uploads the image.

What thumbnail sizes do other platforms ask for?

Most platforms ask for a thumbnail with the same aspect ratio as the video, and the published file-size caps run from 2 MB for LinkedIn video ads to 50 MB for YouTube uploads from desktop. The table gives each platform's own documentation as read in September 2026; the Meta rows come from Internet Archive captures because Meta's pages refuse automated requests.

LinkedIn and X publish thumbnail rules only for video ads, not for ordinary posts. Meta's API documentation still says a Reel shared to the feed uses the middle square as its cover, which does not match the taller profile grid Instagram introduced in 2025; we found no Meta page we could check that states the grid size. TikTok's cover rules sit in a help center that renders only with JavaScript, so TikTok is not in the table.

What happens when a thumbnail breaks YouTube's rules?

A first violation of YouTube's thumbnail policy usually brings a warning, which expires after 90 days if the creator completes a policy training, according to YouTube's thumbnail policy. Repeat offenses can cost a channel its custom-thumbnail privileges for 30 days, or the account, per YouTube's thumbnail help page, and a channel with three strikes in 90 days may be terminated. The policy covers sexual imagery, violent or graphic imagery, vulgar language, and thumbnails that mislead viewers into thinking they are about to see something that is not in the video.

A strike stops a channel from posting. After one Community Guidelines strike a channel cannot post new content, custom thumbnails included, for a week; after a second, for two weeks, per Google's Transparency Report help.

On 18 December 2024 YouTube announced stronger enforcement against "egregious clickbait": videos whose title or thumbnail promises something the video does not deliver, such as a thumbnail reading "top political news" on a video with no news in it. Enforcement began in India, removing such videos without a strike at first, per the Google India blog. YouTube has published no count of videos or thumbnails removed under either rule, and its Transparency Report does not break thumbnails out.

When does Google Search show a video thumbnail?

Only when the video is the main content of the page. Google made that change on April 13, 2023, dropping the format that showed a thumbnail after a result when a video was on the page but was not its main element, per the Google Search Central blog. Google said that in experiments the change had "minimal impact on overall engagement for publishers" and gave no numbers. On December 4, 2023 it applied the same rule to video mode, and told site owners to expect fewer pages with indexed videos, per a second post.

Before December 2023 Google judged prominence by size, and stated the rule two different ways in one year:

The February version applies the 1080-pixel limit to width; the June version applies it to height. Three other rules still apply: Google indexes only one video per page, per Search Console Help; key-moments markup needs a video at least 30 seconds long, per Google's video structured data guide; and the Search Console video indexing report shows 3 months of history. Google publishes no click-through data for video results.

How often do people press play on a video embedded on a website?

Videos on homepages averaged a 24% play rate in 2025, according to Wistia's 2026 State of Video report, the largest published source on the question. Play rate is the share of people who see a video player and press play, which makes it the closest thing to a CTR for embedded video. Wistia sells video hosting, and its figures describe videos hosted on Wistia, which recorded 2.5 billion plays in 2025.

Wistia's earlier editions are no longer online: its 2024 and 2025 State of Video pages were replaced by the 2026 report, and the Internet Archive no longer serves captures of them (checked October 25, 2026), so their findings are not repeated here. In Wistia's 2023 survey, 13% of respondents named play rate as the most important success metric for a video.

The only direct comparison Wistia has published is from 2012. Across all Wistia-hosted videos, those with a custom thumbnail had a 34% play rate against 25% for automatic thumbnails, according to a November 2012 Wistia blog post. It gave no number of videos and no time window.

Play rates are not comparable across hosts or years. Wistia counts a play only when someone interacts with the player, so for autoplaying video its play rate can read far below or above 100%, per its help center. Autoplay was common: in August 2017, 65% of plays in Chrome and 53% in Safari on JW Player's network started automatically, per JW Player. What happens after the play depends on length: in Vidyard's 2025 benchmark report, 65% of viewers watched videos under a minute to the end, against 20% for videos over 20 minutes (2024 data, mostly business and sales videos).

What have Netflix and other platforms learned from testing artwork?

Netflix says artwork was the biggest influence on its members' decision to watch, and made up over 82% of their focus while browsing, in consumer research it ran in early 2014 and described in May 2016 in "The Power of a Picture". The same post says members spent an average of 1.8 seconds considering each title, and that a member whose attention Netflix does not capture within 90 seconds will likely lose interest. Netflix published no sample, country or method for any of the three.

The 90 seconds is softer in its source. A December 2015 paper by two Netflix executives in ACM Transactions on Management Information Systems says a typical member loses interest after "perhaps" 60 to 90 seconds of choosing, having reviewed 10 to 20 titles and looked at perhaps 3 in detail. The same paper says Netflix's recommender influences about 80% of hours streamed, a figure about recommendations, not artwork, that is often blended with the 82%.

What Netflix found about the images themselves

One of Netflix's earliest artwork-testing trends was that an image's tendency to win dropped dramatically when it showed more than 3 people, per its 2016 post. In 2023 Netflix repeated that artwork with more than 3 people tends to perform worse, per its TechBlog, and reported that single-person portraits tend to beat images with more than one person (TechBlog). Netflix has not published an effect size for either finding.

The scale is large. When Netflix described its artwork A/B tests in May 2016, it had 81.5m members, and a title's explore test typically took a few days to find a winning image, per its TechBlog. By 2017 it personalized artwork per member from up to a few dozen candidate images per title (TechBlog), and several Stranger Things images each received over 5% of impressions. A 2026 Netflix paper says some titles have 40+ candidates (arXiv). One hour-long Stranger Things episode holds nearly 86,000 frames, and an average TV season of about 10 episodes nearly 9 million, per Netflix's 2018 post on its image tools.

The measured gains are small

Netflix researchers published their first hard A/B numbers on artwork personalization in 2026, and every online gain is under half a percent. Netflix now serves artwork on five canvas types, and one model serves all of them, per the paper.

At Netflix's size, the authors put a gain of under a fifth of a percent at hundreds of millions of hours a year, so these tests are worth running. They are also nowhere near the double- and triple-digit uplifts quoted for thumbnails in the claims traced below.

YouTube's automatic thumbnails and Spotify's Canvas

YouTube's automatic thumbnailer, launched in 2015, samples each uploaded video at one frame per second and scores each frame with a quality model, per Google Research. Human raters preferred its picks to the previous system's in more than 65% of side-by-side comparisons, a measure of quality, not clicks. YouTube's help page said in 2022 that creators could pick from the three options YouTube generates automatically, per an Internet Archive capture; the current page gives no number.

Spotify's Canvas, the short looping video shown on a track, shows how far a "best case" can sit from an average. In 2019 Spotify said a high-quality Canvas had increased streams by up to 120% and saves by up to 114% (Spotify for Artists). In 2020 its beta tracking against a control group found listeners who saw a Canvas were more likely to keep streaming by +5% on average, to share the track by +145%, to add it to playlists by 20%, to save it by +1.4%, and to visit the artist's profile by +9% (Spotify for Artists). Both are Spotify's own unaudited figures about a packaging feature, not a thumbnail, and the 2020 post also promoted a paid Canvas-design marketplace Spotify owns.

What do independent studies find about faces, text and clickbait in thumbnails?

Studies of YouTube videos find that thumbnails with faces and strong emotion go with more views, but every one of them is observational, not an experiment. The largest, by Lingnan University, analysed 16,215 YouTube videos published February 16 to 28, 2023, in the Journal of Business Research: 51% used thumbnails featuring human faces and 38% had text in the thumbnail, both associated with more viewers, though long text went with fewer clicks, per the university's summary. The 18.3% of videos whose thumbnails expressed strong sentiment had higher view counts, while the 5% with titles phrased as questions had lower ones. Our own YouTube thumbnail study counts how often the biggest channels’ thumbnails use faces, big faces and text, and whether the text repeats the title.

The British Columbia data came from BBTV, a company that sells channel optimization. Pew Research Center's 2019 study of popular YouTube channels measured titles, not thumbnails: across videos posted by the 43,770 channels with at least 250,000 subscribers in the first week of 2019, videos whose titles mentioned words like "Fortnite", "prank" or "worst" got more than five times as many views at the median, per Pew. Among U.S. politics and current-events videos, 36% mentioned "Trump" or "president" in the title, and those got nearly twice the median views.

How common is clickbait on YouTube?

No study measures it for YouTube as a whole, and the published shares run from one video in 40 to 45.45% depending on how the videos were picked:

The Cyprus study also shows how clickbait behaves once found. Its classifier reached 0.924 accuracy on held-out labelled videos. In its hand-checked set, 81% of videos whose thumbnail an image-tagging service labelled "pretty" were clickbait, and comments mentioning the thumbnail were 2.5 times as common on clickbait videos. A viewer watching a clickbait video was recommended 4.1 clickbait videos on average, against 2.73 others, and only 10% of the confirmed clickbait videos were offline by April 30, 2017.

What the popular thumbnail and CTR numbers really say

Most thumbnail statistics in circulation come from creator-tool blogs and agency pages that copy each other. We traced the most repeated ones to the earliest document we could date. Pages that repeat a claim are linked below only as dated evidence of how it changed, never as a source for the number. Several were published by companies that sell thumbnail tools or channel services.

"90% of the best-performing videos on YouTube have custom thumbnails"

This one is YouTube's, but it has never had a method. It first appears in YouTube Creator Academy's thumbnail lesson; the earliest English capture we found, dated December 26, 2015, reads: "90% of the best-performing videos on YouTube have custom thumbnails." (Internet Archive capture). YouTube's earlier Creator Playbook and Creator Hub pages do not contain it. YouTube never defined "best-performing" or gave a sample or a period, and the same lessons said their tips were not promises of success.

The line survives on YouTube's current thumbnail and title tips page, labelled a "fun fact": "90% of the best-performing videos have custom thumbnails." Repeaters have made it causal. A March 2020 post by an editing service said custom thumbnails "helped" 90% of the best performing videos (VidChops).

What to use instead: quote it, if at all, as YouTube's unsourced remark. At most it describes a correlation among successful videos, and custom thumbnails were already the norm among them.

"The average YouTube CTR is 2%" (or 3-4%, 4-5%, 4-6%)

YouTube has never published an average CTR. Its only figure is the statement that half of channels and videos fall between 2% and 10%, unchanged since March 2018 (see above). Each version below puts a different number on it:

• 4-5%: the earliest instance we found is a January 4, 2021 Databox article built from an informal poll of nearly 25 marketers: "According to the marketers we surveyed, the average CTR on YouTube is 4-5%." (Internet Archive capture). It published no questions, dates or data. Databox sells marketing-analytics dashboards.
• 2-10% as "the average": a March 2026 agency post says "the average YouTube CTR across all channels is around 2-10%" (Market Maker Management), turning YouTube's spread into an average.
• 2-10% as "good": Teleprompter.com's CTR guide presents YouTube's 2-10% spread as the definition of a good click-through rate (Teleprompter.com).
• 3-4%: a March 2026 creator-coaching page says "YouTube's own data shows" most channels achieve 2-10% CTR, "with the average sitting around 3-4%" (Alan Spicer). YouTube has published no such average.
• 4-6%: a September 2026 SEO-agency "benchmark" says average organic YouTube CTR "falls between 4-6% across all niches" from a "proprietary" study with no published sample (Focus Digital). Its own source list includes other marketing blogs and a company that sells YouTube views.

What to use instead: YouTube's 2% to 10% statement with its caveats, and a channel's own history. YouTube compares a video's retention with the channel's last 10 similar videos, not with a platform average, and says CTR normally falls as a video reaches new viewers.

"Thumbnails with faces get 38% more clicks"

The 38% is about likes on Instagram photos, not clicks on YouTube thumbnails. It comes from a 2014 Georgia Tech and Yahoo Labs study of a random selection of 1.1 million Instagram photos, which found photos with faces were 38% more likely to receive likes and 32% more likely to receive comments, after controlling for the poster's followers and activity (CHI 2014 paper). Georgia Tech's news release put it as 38 percent more likely to receive likes (Georgia Tech). The paper also found that the number, age and gender of the faces made no difference.

By 2026 thumbnail-tool and agency blogs were stating it as "thumbnails with faces get 38% more clicks" (ThumbnailCreator, February 2026) and faces that "get up to 38% more clicks, according to research" (Market Maker Management, March 2026).

What to use instead: the YouTube evidence is observational. Lingnan University found 51% of 16,215 videos used faces and that they were associated with more viewers, and Netflix found artwork with more than 3 people tended to lose its tests. Neither gives a click uplift for faces.

"The brain processes an image in 13 milliseconds" and "visuals are processed 60,000 times faster than text"

The 13 milliseconds is a real lab result about something else, and it is disputed. MIT reported in January 2014 that people could identify images seen for as little as 13 milliseconds (MIT News). The task was spotting a named target, such as "picnic", in a rapid stream of six or 12 pictures shown for between 13 and 80 ms each (Potter et al., Attention, Perception, & Psychophysics). A 2016 University of Melbourne replication with better masking found no evidence that people could detect the target even at 27 ms per picture (Maguire & Howe).

Netflix carried the MIT figure into artwork in May 2016, writing that "the human brain can process images in as little as 13 milliseconds" (Netflix). By January 2026 a thumbnail-tool blog was telling creators "your thumbnail has 13 milliseconds to stop the scroll" (SkySnail).

The 60,000 figure has no study behind it that we could find. Its earliest known appearance is an advertisement, not research: a Special Advertising Section in the June 28, 1982 issue of Business Week, “Computer Services: Management’s Answer to Data Overload”, quotes Philip Cooper, president of Computer Pictures Corporation, as saying “people assimilate visual information about 60,000 times faster than they assimilate printed copy”, with no source. Alan Levine, who blogs as CogDog and has chased the line since 2012, reported the find after a reader spotted it and a librarian confirmed the issue; Levine wrote to Cooper in 2015 and got only an out-of-office reply (CogDogBlog, March 2015). Later repeats include a 1998 presentation by a 3M product specialist (CogDogBlog, April 2014) and a 3M Meeting Network presentation tip sheet captured in January 1999, which credits unnamed "behavioral research". The only 3M-funded study we traced, a 1986 University of Minnesota study, found presentations with visual aids were 43% more persuasive; it measured persuasion, not speed (University of Minnesota / 3M), and its lead author, Douglas Vogel, told Levine his research “had nothing to do with visual processing speed”. HubSpot still states it as fact: "research has determined that the human brain processes visuals 60,000 times faster than text" (HubSpot, 2023).

What to use instead: neither number says anything about thumbnails. For how long people look at a title, Netflix's 1.8 seconds per title on its own service is the closest published figure.

"Netflix found 82% of viewers choose what to watch by the thumbnail"

Netflix measured something narrower. Its May 2016 post says artwork made up over 82% of members' focus while browsing Netflix, in consumer research run in early 2014, and that members spent an average of 1.8 seconds considering each title. It is a share of attention inside the Netflix app, with no published method, not a share of viewing decisions. The 90 seconds began as "perhaps 60 to 90 seconds" in a 2015 Netflix paper.

The dated changes:

• July 2019: an agency post says thumbnails typically get a view time of around 1.8 seconds and that Netflix has "just 90 seconds" before a user leaves (Stickyeyes).
• October 2021: a design-tool blog says Netflix research "proved" thumbnails are the focus of 82% of browsing time (GoVisually), and turns the average into a maximum: the most any user spends looking at a thumbnail is 1.8 seconds.
• August 2025: a YouTube-tool blog tells creators "you have about 1.8 seconds to make an impression" on YouTube, with no source (1of10).

What to use instead: Netflix's own wording, labelled as Netflix research about Netflix's interface.

"Custom thumbnails increase clicks by 154%"

We found no study behind it. The earliest text version is an August 2016 Neil Patel post saying simple and beautiful thumbnails "can increase your engagement by 154%". It links to his own 2014 QuickSprout infographic, where the line is "+154% more clicks by women for YouTube videos with a thumbnail that stands out" (and "+64% for men"). None of the infographic's references that we could open contains it.

Since then it has been reattributed several times:

• March 2020: credited to BuzzFeed, "buzzfeed.com says thumbnails increase your channel's engagement by 154%" (VidChops).
• October 2024: "engagement" becomes click-through rate, "custom thumbnails can increase click-through rates (CTR) by up to 154%" (Motion Arts Media).
• March 2026: credited to YouTube's Creator Academy, 154% (Market Maker Management). None of the Creator Academy captures we checked, from 2015 to 2021, contains it.
• May 2026: credited to a "ThumbAI" study, "custom thumbnails averaged an 8.2% CTR, a massive 154% increase over the 3.2% CTR of auto-generated thumbnails" (ThumbnailCreator). The study page no longer exists and was never archived, and its two numbers give a 156% increase, not 154%.

What to use instead: nothing public. The only custom-versus-automatic comparison we found from a platform's own data is Wistia's 2012 play-rate figure, 34% against 25%, published with no sample.

"More than 70% of YouTube watch time comes from recommendations"

YouTube has never published this figure in its own posts. It first appears in an August 30, 2017 Verge feature built on interviews with YouTube staff, which reported that "more than 70 percent of the time people spend watching videos on the site is now driven by YouTube's algorithmic recommendations" (The Verge, quoted by Annotote). YouTube's chief product officer repeated it on a CES panel in January 2018, and a January 10, 2018 CNET report of that remark is where most repeaters find it. CNET first printed "more than 75 percent" and corrected it to "more than 70 percent" the same day, and a syndicated copy still carries 75 percent. YouTube's own 2021 post on recommendations gives no percentage; it says recommendations drive a significant amount of viewing, more than channel subscriptions or search.

What to use instead: YouTube's own wording, with no number, or leave it out.

"5% of YouTube videos get 95% of the views"

This one rests on a real count, but the count is from October 2015. ReelSEO, the video marketing site then owned by the analytics company Tubular Labs, reported Tubular’s tally that YouTube held at least 1.1 billion videos with 7.8 trillion views between them, and that the 58.6 million videos with 10,000 or more views had generated 7.4 trillion of those views, "around 95% of the total all time view count" (ReelSEO, as quoted by Music Ally). 58.6 million is about 5% of 1.1 billion, which is where the ratio comes from. It measures all-time views across everything uploaded up to 2015, so it describes the library, not how a new upload will do. ReelSEO’s article address now redirects to the Tubular Labs home page.

What to use instead: the dated 2015 count. In October 2015, videos with at least 10,000 views, about 5% of YouTube’s videos, had about 95% of all-time views.

"YouTube has over 2 billion users", "a billion hours are watched every day" and "70% of watch time is on mobile"

All three are YouTube’s own figures, from the press page it kept for journalists, and that page no longer carries them. Its old addresses, youtube.com/yt/about/press and youtube.com/about/press, now redirect to the YouTube blog, so statistics posts that cite them link to a page without the numbers. The figures changed as YouTube updated the page:

• 2017: YouTube "has over a billion users" who "watch a billion hours of video" a day, and "More than half of YouTube views come from mobile devices" (YouTube press page, Internet Archive capture).
• 2018: "Over 1.9 billion logged-in users visit YouTube each month", and "More than 70% of YouTube watchtime comes from mobile devices" (capture).
• 2020: "Over 2 Billion logged-in users visit YouTube each month and every day people watch over a billion hours of video", with the 70% mobile line unchanged (capture).

YouTube announced the billion hours a day on its official blog on February 27, 2017, as a milestone reached the year before: people were "now watching a billion hours of YouTube’s incredible content every single day" (YouTube Official Blog). The user counts are logged-in users, so they leave out everyone who watches without signing in, and the mobile share is of watch time, where the 2017 page counted views. YouTube’s press page, read in October 2026, gives no total, no daily hours and no mobile share; it says only "billions of monthly logged-in users". The reach figures did survive: it still says there are "localized versions of YouTube in over 100 countries around the world, across 80 languages", as the 2020 version did. Totals such as 2.3 or 2.7 billion users are not on the press page in any version we read.

What to use instead: the figure with its year, such as "over 2 billion logged-in users a month (YouTube, 2020)", linked to the archived copy of YouTube’s press page, or YouTube’s 2017 blog post for the billion hours.

"Over a billion users", "6 billion hours a month" and "5 billion videos a day": YouTube’s older statistics page

Before the press page above, YouTube kept a statistics page for journalists at youtube.com/yt/press/statistics.html (earlier youtube.com/t/press_statistics), and older statistics posts link that address. It now redirects to the YouTube blog, which carries none of these figures. The numbers changed as YouTube updated the page:

• December 2012: "Over 800 million unique users visit YouTube each month" and "Over 4 billion hours of video are watched each month" (YouTube statistics page, Internet Archive capture).
• April to early June 2013: "More than 1 billion unique users visit YouTube each month", while the hours line still read "Over 4 billion hours of video are watched each month"; by June 12, 2013 it said 6 billion (capture, April 2013).
• March to August 2012: “Over 3 billion hours of video are watched each month on YouTube” (capture, March 2012), the figure a 2013 statistics list was still quoting after the page had moved to 4 and then 6 billion.
• December 2013: "More than 1 billion unique users visit YouTube each month", "Over 6 billion hours of video are watched each month", and "Mobile makes up almost 40% of YouTube’s global watch time" (capture).
• December 2014: still "Mobile makes up almost 40% of YouTube’s global watch time" (capture, December 2014). By February 2015 the mobile line counted views instead: "Half of YouTube views are on mobile devices" (capture, February 2015).
• September 2015, the first version with this wording that we found: "YouTube has over a billion users — almost one-third of all people on the Internet", and in the mobile section, "The number of hours people spent watching videos on mobile is up 100% y/y" (capture, September 2015). The August 2015 version had said "YouTube has more than 1 billion users", and its 100% figure was mobile revenue, not viewing.
• December 2015: "YouTube has over a billion users", "every day people watch hundreds of millions of hours on YouTube and generate billions of views", "More than half of YouTube views come from mobile devices", and "On mobile, the average viewing session is now more than 40 minutes, up more than 50% y/y" (capture).

Two figures that statistics lists credit to this page are not on any version we read (December 2012, April and June 2013, December 2013, 2014, August 2015, December 2015 and 2016): "more than 1.3 billion people use YouTube" and "almost 5 billion videos are watched every day". The page never gave a daily view count beyond "billions of views". Two lines from the 2015 versions changed as they were copied. "Almost one-third of all people on the Internet" became "one-third of internet users have a YouTube account", although YouTube counted users, not accounts, and never said what a user was. One year’s growth in YouTube mobile viewing became a permanent rate for all mobile video: a January 2016 list by the agency Insivia wrote that "YouTube reports mobile video consumption rises 100% every year" (Insivia, January 2016, Internet Archive capture), and later lists drop "YouTube reports". A third line splices two versions together: “Mobile makes up more than half of global watch time on YouTube”, as a January 2016 statistics deck put it (Adelie Studios deck, SlideShare, Internet Archive capture), is on no version. The page gave mobile as a share of watch time only as “almost 40%” (2013 and 2014); the “half” of 2015 and the “more than half” that followed count views. What to use instead: the figure with the year of the version it came from, such as "over a billion users (YouTube, 2015)", linked to the archived copy of the page.

"YouTube is the second largest search engine"

This rests on one month of U.S. data from 2008, and the company that measured it ranked Yahoo second. comScore’s August 2008 search report counted 2,564 million U.S. searches on a line it labelled "YouTube/All Other" against 2,427 million on Yahoo! Sites (comScore release, September 2008, Internet Archive capture). That line is Google Sites minus Google search, so it counts every Google property other than the search engine itself, and the survey covered U.S. home, work and university computers only. comScore’s own text says "Yahoo! Sites ranked second with 2.4 billion searches". Lists now give the claim as a present fact about YouTube alone, worldwide. The release’s old address now redirects to comScore’s list of press releases.

What to use instead: the dated comparison. In August 2008, comScore counted slightly more U.S. searches on Google’s non-search properties, YouTube among them, than on Yahoo, while still ranking Yahoo second.

"YouTube is the second most visited website in the world"

The ranking still holds in 2026, but the source most lists link to no longer exists. Lists credit Alexa, Amazon’s web-traffic ranking, which ordered sites by a mix of average daily visitors and pageviews estimated from its own panel. In its last capture before the service closed, Alexa’s global top-sites list ranked youtube.com 2, after google.com (Alexa top sites, April 1, 2022, Internet Archive capture). The "minutes a day on YouTube" figures credited to Alexa come from the same service’s Daily Time on Site estimate, the average time a visitor spent on youtube.com each day: 14:55 in April 2022 (youtube.com on Alexa, April 2022, Internet Archive capture) and 17:12 in January 2021 (Alexa top sites, January 2021, Internet Archive capture), so each list’s number depends on the month it was copied. Both counted visits to the website in a browser, not viewing in YouTube’s apps or on TVs. Amazon retired Alexa.com on May 1, 2022 (Alexa’s end-of-service notice, Internet Archive capture), and its old addresses now open an Amazon sign-in page.

What to use instead: a current ranking, with its month and who made it. Similarweb’s ranking for September 2026 still puts youtube.com 2nd worldwide, after google.com (Similarweb Top Websites Ranking, updated October 1, 2026), and Semrush estimates 59.97 billion visits to youtube.com that month, second to google.com’s 108.9 billion (Semrush, Most Visited Websites in the World, September 2026). Both are third-party estimates of website traffic, not YouTube’s own counts, and both change monthly. For time spent, Similarweb’s nearest figure is an average visit of 00:18:50 on youtube.com in September 2026, which is per visit, not per day.

"How-to searches on YouTube grow 70% a year"

This is a one-year growth rate from 2015, still quoted in the present tense. In May 2015 Think with Google published an article on "I-want-to-do moments" by Google’s David Mogensen, which said that "searches related to ‘how to’ on YouTube are growing 70% year over year", sourced to Google data for the U.S. from the first quarter of 2014 to the first quarter of 2015, and that more than 100M hours of how-to content had been watched in North America so far that year (Think with Google, May 2015, Internet Archive capture). No base volume was given. The same article’s "67% of millennials agree that they can find a YouTube video on anything they want to learn" comes from a Google Consumer Survey of 385 U.S. adults aged 18 to 34 in April 2015. Its summary line, “Of smartphone users, 91% turn to their devices for ideas while completing a task”, comes from a different survey: Google/Ipsos’s “Consumers in the Micro-Moment” study of March 2015, with 9,598 respondents from the online population (the article at its later address, Internet Archive capture). That is what smartphone users said they do, not a count of how-to searches. Its line that “nearly one in three millennials say they’ve purchased a product as a result of watching a how-to video” is sourced to yet another Google Consumer Survey, of 1,128 people in April 2015: a self-reported purchase at any time, not a conversion rate. A growth rate measured over a single year in 2014–15 says nothing about how-to searches today. The article’s address, and the later Think with Google pages that repeated the figure, now redirect to the home page of Think with Google’s successor at business.google.com.

What to use instead: the figure with its period, "U.S. how-to searches on YouTube grew 70% from the first quarter of 2014 to the first quarter of 2015 (Google, May 2015)", linked to the archived article.

"There are 51 million YouTube channels" (or 37 million, or 31 million)

YouTube has never published how many channels exist. The figure most pages repeat comes from "How Many YouTube Channels Are There?", a post by the YouTube marketing agency tubics that was rewritten each year with a new headline number: more than 23 million as of 2018 (2018 version), 31 million as of 2019 (2019 version), 37 million as of 2020 (2020 version), and from January 2022 "more than 51 million YouTube channels" as of 2022 (2022 version). The counts came from Social Blade, which only tracks channels with at least five subscribers, and the 2022 version's own data paragraph still described the 38 million channels it had counted in 2020, so nothing on the page shows where 51 million came from. tubics.com now returns an error for every page (checked October 19, 2026), so the articles that cite it link to nothing; the Internet Archive keeps each version.

What to use instead: YouTube's own count of channels that earn money on it, more than 3 million creators in the YouTube Partner Program in May 2026 (YouTube Official Blog), and say that no official total exists.

Every popular claim our research pages trace, 853 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 video thumbnail and ctr 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 268 figures from 49 publishers.

On September 24, 2026, every figure then on the page was re-fetched from its source on a cleared cache and matched against the recorded wording; all 220 sourced figures re-verified automatically. Figures added since were checked on the dates in the change log. Pages that block automated requests or no longer exist, including Netflix's TechBlog, Meta's developer and business help pages, TubeBuddy, YouTube Creator Academy, and earlier versions of YouTube's and Google's help pages, were read from dated Internet Archive captures of the publisher's own pages. We left out TikTok's cover-image rules, because its help center renders only with JavaScript; the size of Instagram's profile grid, which we found on no Meta page we could check; Wistia's full reports, which sit behind a form, and its thumbnail tests that give no sample size; a 2026 journal study of YouTube thumbnails whose full text is behind a browser check; vendor totals with no method, such as counts of "extra views" from thumbnail-testing tools; and a 2026 preprint's figure for how much misleading thumbnails raise CTR, because the paper it cites could not be found.

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 figures come from platforms describing themselves. YouTube, Google, Netflix, Meta, TikTok and Spotify state their own rules and results. No one audits them. YouTube's CTR range comes with no method, sample or date.
• CTR and play rate are different measures. YouTube's CTR counts views per thumbnail impression on YouTube's own surfaces; Wistia's play rate counts presses of an embedded player; other hosts define play rate differently. Do not compare them with each other.
• The research on thumbnails is observational. The studies here link faces, text and emotion in thumbnails to views, but none assigned thumbnails at random. The study of changed thumbnails used data from BBTV, a company that sells channel optimization.
• How common clickbait looks depends on how videos were sampled. The published shares differ widely with how the videos were chosen, and none samples YouTube as a whole.
• Vendor data describes the vendor's own customers. Wistia, Vidyard, JW Player, TubeBuddy and NoteLM.ai each sell into this market, and their figures cover the videos and channels that use their products.
• Specifications are a snapshot. YouTube changed its thumbnail size and file-size rules in March 2026 and its API limits in September 2026. Meta's rules come from Internet Archive captures dated 2025 and 2026.
• Netflix's figures describe Netflix. Its 2014 browsing research has no published method, and its 2026 test results come from research papers posted before peer review.

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/video-thumbnail-and-ctr-statistics#CTR-001. To cite the page as a whole:

Mark Studios. "Video Thumbnail and CTR Statistics." Last verified Sep 24, 2026. https://www.markstudios.com/research/video-thumbnail-and-ctr-statistics

@misc{markstudios_video_thumbnail_and_ctr_statistics_2026,
  author = {{Mark Studios}},
  title = {{Video Thumbnail and CTR Statistics}},
  year = {2026},
  month = sep,
  url = {https://www.markstudios.com/research/video-thumbnail-and-ctr-statistics},
  note = {Last verified 2026-09-24}
}

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. YouTube (Google), Impressions & click-through-rate FAQs - YouTube Help (Undated (Google Help Center pages carry no publication or update date)). Accessed 2026-09-24.
2. YouTube (Google), YouTube Studio: Better Insights, New Metrics & Faster Access to News - YouTube Official Blog (2018-03-07). Accessed 2026-09-24.
3. YouTube (Google), YouTube Help article 7577430 on impressions and click-through rate (Internet Archive capture of 2018-03-14) (2018-03-14 (capture date; the article itself is undated)). Accessed 2026-09-24.
4. YouTube (Google), Decoding CTR & impressions in your analytics - YouTube Help (Undated (Google Help Center pages carry no publication or update date)). Accessed 2026-09-24.
5. YouTube (Google), Stop guessing, start growing: master these 4 metrics - YouTube Official Blog (2025-09-03). Accessed 2026-09-24.
6. YouTube (Google), Check your YouTube thumbnail impressions and watch time - Computer - YouTube Help (Undated (Google Help Center pages carry no publication or update date)). Accessed 2026-09-24.
7. Google for Developers (YouTube Analytics and Reporting APIs), Metrics | YouTube Analytics and Reporting APIs | Google for Developers (YouTube Reporting API) (2026-09-14 (page 'Last updated' date)). Accessed 2026-09-24.
8. Google for Developers (YouTube Analytics and Reporting APIs), Revision History | YouTube Analytics and Reporting APIs | Google for Developers (2026-01-15 (entry date); page last updated 2026-09-14). Accessed 2026-09-24.
9. YouTube (Google), How engagement metrics are counted - YouTube Help (Undated (Google Help Center pages carry no publication or update date)). Accessed 2026-09-24.
10. YouTube (Google), On YouTube's recommendation system - YouTube Official Blog (Cristos Goodrow, VP of Engineering) (2021-09-15). Accessed 2026-09-24.
11. YouTube (Google), Changes to Related and Recommended Videos - YouTube Official Blog (2012-03-09). Accessed 2026-09-24.
12. YouTube (Google), YouTube Now: Why We Focus on Watch Time - YouTube Official Blog (2012-08-10). Accessed 2026-09-24.
13. YouTube (Google), YouTube search, now optimized for time watched - YouTube Official Blog (2012-10-12). Accessed 2026-09-24.
14. YouTube (Google), Content tab analytics tips - YouTube Help (Undated (Google Help Center pages carry no publication or update date)). Accessed 2026-09-24.
15. YouTube (Google), A/B test titles & thumbnails - YouTube Help (Undated). Accessed 2026-09-24.
16. YouTube (Official Blog, Rene Ritchie, YouTube Creator Liaison), Rene's Top Five on YouTube: October 26, 2023 Edition (2023-10-26). Accessed 2026-09-24.
17. YouTube (Official Blog, Rene Ritchie, YouTube Creator Liaison), Rene's Top Five on YouTube: August 3, 2023 Edition (2023-08-03). Accessed 2026-09-24.
18. YouTube (Official Blog, Rene Ritchie, YouTube Creator Liaison), Rene's Top Five on YouTube: November 2, 2023 Edition (2023-11-02). Accessed 2026-09-24.
19. YouTube (Official Blog, Rene Ritchie, YouTube Creator Liaison), Rene's Top Five on YouTube: January 18, 2024 Edition (2024-01-18). Accessed 2026-09-24.
20. YouTube (Official Blog, Rene Ritchie), Rene's Top Five on YouTube: June 14, 2024 edition (2024-06-14). Accessed 2026-09-24.
21. YouTube (Official Blog, Amjad Hanif, VP of Creator Products), Transform your creative journey with the latest YouTube Studio updates (2025-09-16). Accessed 2026-09-24.
22. YouTube (Official Blog, Aparna Pappu, VP), New tools to power your creation journey from start to finish (2026-09-23). Accessed 2026-09-24.
23. Google, Create a video experiment - Google Ads Help (Undated). Accessed 2026-09-24.
24. Meta, About confidence in your tests and experiments | Meta Business Help Center (Undated). Accessed 2026-09-24.
25. Meta, Best practices for A/B testing | Meta Business Help Center (2021-03-15). Accessed 2026-09-24.
26. TikTok, About Split Testing | TikTok Ads Manager (2026-01). Accessed 2026-09-24.
27. TikTok, Split Test Best Practices | TikTok Ads Manager (2026-01). Accessed 2026-09-24.
28. TubeBuddy (vendor), YouTube Thumbnail A/B Testing Tool | TubeBuddy (Undated). Accessed 2026-09-24.
29. Google, Video experiments boost creative performance on YouTube (2021-04-06). Accessed 2026-09-24.
30. NoteLM.ai (vendor), YouTube Thumbnail A/B Test 2026: Test & Compare Setup + 40% CTR Gains (2026-01-05). Accessed 2026-09-24.
31. YouTube (Google), Add custom thumbnails on YouTube (YouTube Help) (Undated, read 2026-09-24). Accessed 2026-09-24.
32. YouTube (Google), Add video thumbnails on YouTube (YouTube Help), Internet Archive capture (Undated (page as archived 2026-03-11)). Accessed 2026-09-24.
33. YouTube (Google), Add video thumbnails on YouTube (YouTube Help), Internet Archive capture (Undated (page as archived 2026-03-12)). Accessed 2026-09-24.
34. YouTube (Google), Videos | YouTube Data API (2026-09-16). Accessed 2026-09-24.
35. YouTube (Google), Revision History | YouTube Data API (2026-09-11). Accessed 2026-09-24.
36. YouTube (Google), Thumbnails: set | YouTube Data API (2026-09-14). Accessed 2026-09-24.
37. Vimeo, How to change the thumbnail image for my video (Vimeo Help Center) (2025-11-07). Accessed 2026-09-24.
38. Vimeo, About thumbnail options for your Vimeo OTT videos (Vimeo Help Center) (2026-09-22). Accessed 2026-09-24.
39. Meta (Instagram Platform), IG User Media reference, Instagram Platform documentation (Meta for Developers), Internet Archive capture (Undated (page as archived 2026-05-13)). Accessed 2026-09-24.
40. Meta (Facebook Graph API), Graph API Video Thumbnails reference (Meta for Developers), Internet Archive capture (Undated (page as archived 2025-06-29)). Accessed 2026-09-24.
41. LinkedIn, Video ads advertising specifications (LinkedIn Marketing Solutions Help) (Undated (page shows 'Last updated: 2 months ago' on 2026-09-24)). Accessed 2026-09-24.
42. X Corp., X Ads creative specs (X Business) (Undated, read 2026-09-24). Accessed 2026-09-24.
43. Spotify, Thumbnails (Spotify for Creators support) (Undated, read 2026-09-24). Accessed 2026-09-24.
44. Apple (Apple Podcasts for Creators), Show cover (Apple Podcasts for Creators artwork guide) (Undated, read 2026-09-24). Accessed 2026-09-24.
45. Google Search Central, Video SEO Best Practices (Google Search Central documentation) (2025-12-18). Accessed 2026-09-24.
46. YouTube (Google), Thumbnails policy (YouTube Help) (Undated, read 2026-09-24). Accessed 2026-09-24.
47. Google Transparency Report (YouTube), YouTube Community Guidelines enforcement FAQs (Transparency Report Help Center) (Undated, read 2026-09-24). Accessed 2026-09-24.
48. YouTube (Google), Google India Blog, Strengthening enforcement against egregious clickbait on YouTube (2024-12-18). Accessed 2026-09-24.
49. Google Search Central, Simplifying video presentation on Google Search Results (Google Search Central Blog) (2023-04-13). Accessed 2026-09-24.
50. Google Search Central, Video mode now only shows pages where video is the main content (Google Search Central Blog) (2023-12-04). Accessed 2026-09-24.
51. Google Search Console Help, Video indexing report - Search Console Help (Undated (help page; Internet Archive capture 2023-02-20)). Accessed 2026-09-24.
52. Google Search Central, Improving prominence reporting in the Search Console video indexing report (2023-06-12). Accessed 2026-09-24.
53. Google Search Console Help, Video indexing report - Search Console Help (Undated (live help page)). Accessed 2026-09-24.
54. Google Search Central, Video (VideoObject, Clip, BroadcastEvent) structured data (2026-09-24). Accessed 2026-09-24.
55. Wistia, The Key Takeaways From Wistia's 2026 State of Video Webinar (2026-04-22). Accessed 2026-09-24.
56. Wistia, State of Video Report: Video Marketing Statistics for 2026 (2026-04-22). Accessed 2026-09-24.
57. Wistia, 2023 State of Video Marketing Report - Wistia (2023-02-28). Accessed 2026-09-24.
58. Wistia, Two Thumbnails Up: Optimizing Video Play Rates (2012-11-05). Accessed 2026-09-24.
59. Wistia, Average User Engagement Analytics | Wistia Help Center (2026-07-08). Accessed 2026-09-24.
60. JW Player, Why Click-to-Play Video is a Digital Publisher's Best Friend (2017-08-01). Accessed 2026-09-24.
61. Vidyard, Video in Business Benchmark Report | Vidyard (2026-09-11). Accessed 2026-09-24.
62. Netflix, The Power of a Picture (2016-05-03). Accessed 2026-09-24.
63. Netflix (Gomez-Uribe & Hunt, ACM TMIS), The Netflix Recommender System: Algorithms, Business Value, and Innovation (ACM Transactions on Management Information Systems 6(4), Article 13) (2015-12). Accessed 2026-09-24.
64. Netflix (Netflix Technology Blog), Causal Machine Learning for Creative Insights (2023-01-11). Accessed 2026-09-24.
65. Netflix (Netflix Technology Blog), Discovering Creative Insights in Promotional Artwork (2023-01-30). Accessed 2026-09-24.
66. Netflix (Netflix Technology Blog), Selecting the best artwork for videos through A/B testing (2016-05-03). Accessed 2026-09-24.
67. Netflix (Netflix Technology Blog), Artwork Personalization at Netflix (2017-12-07). Accessed 2026-09-24.
68. Netflix (arXiv preprint), Netflix Artwork Personalization via LLM Post-training (arXiv:2601.02764v2) (2026-01-06 (v2 2026-08-26)). Accessed 2026-09-24.
69. Netflix (Netflix Technology Blog), AVA: The Art and Science of Image Discovery at Netflix (2018-02-07). Accessed 2026-09-24.
70. Netflix (arXiv preprint), Multimedia Asset Personalization via Multimodal Embeddings at Netflix (arXiv:2608.18322v1) (2026-08-18). Accessed 2026-09-24.
71. Google Research / YouTube, Improving YouTube video thumbnails with deep neural nets (2015-10-08). Accessed 2026-09-24.
72. YouTube (Google), Add video thumbnails on YouTube (YouTube Help) (Undated (help page; Internet Archive capture 2022-06-05)). Accessed 2026-09-24.
73. Spotify (Spotify for Artists), Canvas Metrics Guide (2019-10-10). Accessed 2026-09-24.
74. Spotify (Spotify for Artists), Introducing a New Way to Connect with Canvas Designers (2020-11-19). Accessed 2026-09-24.
75. Lingnan University (Journal of Business Research 2024), Clicks for money: Predicting video views through a sentiment analysis of titles and thumbnails (Lingnan Scholars abstract page) (2024-10). Accessed 2026-09-24.
76. Lingnan University (Journal of Business Research 2024), Lingnan study using AI analyses over 16,000 social media videos in order to show key marketing strategies that boost viewership (press release) (2024-08-06). Accessed 2026-09-24.
77. Korea University Business School et al. (Decision Support Systems 2022, author copy), An Exploration of the Relation between the Visual Attributes of Thumbnails and the View-Through of Videos: The Case of Branded Video Content (2022-02-27). Accessed 2026-09-24.
78. KAIST (The Web Conference 2026 / arXiv), Moral Outrage Shapes Commitments Beyond Attention: Multimodal Moral Emotions on YouTube in Korea and the US (2026-01-30). Accessed 2026-09-24.
79. University of British Columbia (IEEE TKDE / arXiv), Engagement dynamics and sensitivity analysis of YouTube videos (2016-11-02). Accessed 2026-09-24.
80. Duke University et al. (arXiv preprint), Auditing Engagement Incentives in the Kidfluencer Ecosystem: A Multimodal Weak Supervision Approach (2026-06-02). Accessed 2026-09-24.
81. Yahoo Research (CIKM 2016 / arXiv), To Click or Not To Click: Automatic Selection of Beautiful Thumbnails from Videos (2016-09-06). Accessed 2026-09-24.
82. Pew Research Center, A Week in the Life of Popular YouTube Channels (2019-07-25). Accessed 2026-09-24.
83. Pew Research Center, A Week in the Life of Popular YouTube Channels (2019-07-25). Accessed 2026-09-24.
84. Cyprus University of Technology (IEEE Security and Privacy Workshops 2018), The Good, the Bad and the Bait: Detecting and Characterizing Clickbait on YouTube (2018-05-24). Accessed 2026-10-21.
85. Bauhaus-Universitat Weimar & Leipzig University (HCOMP 2018 Works in Progress, CEUR-WS), Towards Crowdsourcing Clickbait Labels for YouTube Videos (2018-07). Accessed 2026-09-24.
86. Rajshahi University of Engineering & Technology and Qatar University et al. (arXiv preprint), YTClickbait21K: Human-Annotated Multimodal Dataset for YouTube Clickbait Detection Across Diverse Channels and Content Categories (2026-06-10). Accessed 2026-09-24.
87. YouTube (Google), YouTube Creator Academy, Make great thumbnails - YouTube (Creator Academy lesson) (Undated). Accessed 2026-09-24.
88. YouTube (Google), Thumbnail & title tips - YouTube Help (Undated). Accessed 2026-09-24.
89. VidChops, Do YouTube Thumbnails Increase Engagement (2020-03-15). Accessed 2026-09-24.
90. Databox, Improve YouTube Click-Through Rate (CTR) ... 11 Pro Tips (Jessica Malnik, Databox) (2021-01-04). Accessed 2026-09-24.
91. Market Maker Management, Why YouTube Thumbnails Matter More Than You Think (2026-03-27). Accessed 2026-09-24.
92. Teleprompter.com, What's a Good Click-Through Rate on YouTube? (2026 Benchmarks) (2025-03-17). Accessed 2026-09-24.
93. Alan Spicer, What Is a Good YouTube Click-Through Rate in 2026? (2026-03-31). Accessed 2026-09-24.
94. Focus Digital, Average YouTube CTR: Organic & Paid Benchmarks in 2026 (2026-09-11). Accessed 2026-09-24.
95. Georgia Tech and Yahoo Labs (Bakhshi, Shamma & Gilbert), CHI 2014, Faces Engage Us: Photos with Faces Attract More Likes and Comments on Instagram (2014-04-26). Accessed 2026-09-24.
96. Georgia Institute of Technology (News Center), Face It: Instagram Pictures With Faces are More Popular (2014-03-20). Accessed 2026-09-24.
97. ThumbnailCreator, Top 10 Thumbnail Trends from Leading YouTubers (2026-02-20). Accessed 2026-09-24.
98. MIT News Office, In the blink of an eye (2014-01-16). Accessed 2026-09-24.
99. Potter et al., Attention, Perception, & Psychophysics (Psychonomic Society/Springer), Detecting meaning in RSVP at 13 ms per picture (2013-12-28). Accessed 2026-09-24.
100. Maguire & Howe (University of Melbourne), Attention, Perception, & Psychophysics, Failure to detect meaning in RSVP at 27 ms per picture (PubMed abstract, PMID 27048442) (2016-07). Accessed 2026-09-24.
101. SkySnail, YouTube Thumbnail Psychology: The Science Behind Thumbnails That Get Clicked (2026-01-28). Accessed 2026-09-24.
102. CogDogBlog (Alan Levine), The 60,000 Times Faster Claim Gets Dialed Back to 1982 (2015-03-27). Accessed 2026-10-26.
103. CogDogBlog (Alan Levine), Tweet it, Blog It, Repeat It, 60,000 Times: Truthiness Achieved (2014-04-07). Accessed 2026-10-26.
104. University of Minnesota MIS Research Center with 3M (Vogel, Dickson & Lehman), Persuasion and the Role of Visual Presentation Support: The UM/3M Study (MISRC-WP-86-11) (1986-06). Accessed 2026-09-24.
105. HubSpot, How Video Can Enhance Your Landing Pages (2023-08-15). Accessed 2026-09-24.
106. Stickyeyes, The importance of the thumbnail in selling your content (2019-07-25). Accessed 2026-09-24.
107. GoVisually, Thumbnail Artwork: How It's Helping Million Dollar Businesses (2021-10-29). Accessed 2026-09-24.
108. 1of10, How to Boost YouTube CTR with Better Thumbnails (The psychology behind high-CTR thumbnails) (2025-08-12). Accessed 2026-09-24.
109. Motion Arts Media, YouTube Thumbnail Size: Best Thumbnail Dimensions & Resolution (2024-10-24). Accessed 2026-09-24.
110. ThumbnailCreator, Before-and-After Thumbnails: What Works and Why (2026-05-11). Accessed 2026-09-24.
111. Annotote (Anthony Bardaro), quoting The Verge, TLDR: How YouTube perfected the feed (2017-09-25). Accessed 2026-11-09.
112. ReelSEO (Tubular Labs data), as quoted by Music Ally, Report claims 5% of YouTube videos drive 95% of views (2015-10-21). Accessed 2026-10-15.
113. YouTube (Google), press page via Internet Archive, YouTube for Press (2017 capture) (2017-06-15). Accessed 2026-10-15.
114. YouTube (Google), press page via Internet Archive, YouTube for Press (2018 capture) (2018-10-15). Accessed 2026-10-15.
115. YouTube (Google), press page via Internet Archive, YouTube for Press (2020 capture) (2020-06-15). Accessed 2026-10-15.
116. YouTube (Google), YouTube Official Blog, You know what's cool? A billion hours (Cristos Goodrow) (2017-02-27). Accessed 2026-10-15.
117. YouTube (Google), YouTube Official Blog, YouTube Press (2026-10-15). Accessed 2026-10-15.
118. YouTube (Google), statistics page via Internet Archive, YouTube press statistics (2012 capture) (2012-12-01). Accessed 2026-10-16.
119. YouTube (Google), statistics page via Internet Archive, YouTube press statistics (April 2013 capture) (2013-04-02). Accessed 2026-10-20.
120. YouTube (Google), statistics page via Internet Archive, YouTube press statistics (March 2012 capture) (2012-03-01). Accessed 2026-10-25.
121. YouTube (Google), statistics page via Internet Archive, YouTube press statistics (2013 capture) (2013-12-14). Accessed 2026-10-16.
122. YouTube (Google), statistics page via Internet Archive, YouTube press statistics (December 2014 capture) (2014-12-30). Accessed 2026-10-25.
123. YouTube (Google), statistics page via Internet Archive, YouTube press statistics (February 2015 capture) (2015-02-14). Accessed 2026-10-25.
124. YouTube (Google), statistics page via Internet Archive, YouTube press statistics (September 2015 capture) (2015-09). Accessed 2026-10-25.
125. YouTube (Google), statistics page via Internet Archive, YouTube press statistics (December 2015 capture) (2015-12-01). Accessed 2026-10-16.
126. Insivia, 50 Must-Know Stats About Video Marketing 2016 (2016-01-12). Accessed 2026-10-25.
127. comScore, comScore Releases August 2008 U.S. Search Engine Rankings (2008-09-18). Accessed 2026-10-25.
128. Alexa Internet (Amazon), Alexa - Top sites (April 1, 2022 capture) (2022-04-01). Accessed 2026-10-29.
129. Alexa Internet (Amazon), youtube.com Competitive Analysis, Marketing Mix and Traffic - Alexa (April 1, 2022 capture) (2022-04-01). Accessed 2026-10-29.
130. Alexa Internet (Amazon), Alexa - Top sites (January 1, 2021 capture) (2021-01-01). Accessed 2026-10-29.
131. Alexa Internet (Amazon), Alexa.com End of Service Notice (May 2, 2022 capture) (2022-05-01). Accessed 2026-10-29.
132. Similarweb, Top Websites Ranking - Most Visited Websites in September 2026 (2026-10-01). Accessed 2026-10-29.
133. Semrush, Most Visited Websites in the World September 2026 (2026-10-01). Accessed 2026-10-29.
134. Think with Google (David Mogensen, Google), I Want-to-Do Moments: From Home to Beauty (2015-05). Accessed 2026-10-16.
135. Think with Google (Google), I-want-to-do moments: From home to beauty (2015-05). Accessed 2026-10-20.
136. tubics (Internet Archive capture), How Many YouTube Channels Are There? (tubics), Internet Archive capture of 16 July 2019 (2018-12-10). Accessed 2026-10-19.
137. tubics (Internet Archive capture), How Many YouTube Channels Are There? (tubics), Internet Archive capture of 1 January 2020 (2019-12-19). Accessed 2026-10-19.
138. tubics (Internet Archive capture), How Many YouTube Channels Are There? (tubics), Internet Archive capture of 3 January 2021 (2020-11-13). Accessed 2026-10-19.
139. tubics (Internet Archive capture), How Many YouTube Channels Are There? (tubics), Internet Archive capture of 14 January 2022 (2022-01-11). Accessed 2026-10-19.
140. YouTube Official Blog, The YouTube Partner Program is now live in Armenia (2026-05-11). Accessed 2026-10-19.

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 24, 2026: First published with 224 figures from 40 publishers, each checked against its original source.
• Sep 24, 2026: Corrected three Mark Studios pages that disagreed with this one. A blog post said YouTube's test picks winners by click-through rate (YouTube says watch time), and our glossary and a KPI post gave CTR benchmarks with no source.
• Sep 28, 2026: Corrected the origin of "more than 70% of YouTube watch time comes from recommendations": it first appears in an August 30, 2017 Verge feature reporting YouTube staff's figure, four months before the January 2018 CES remark we had named.
• Oct 15, 2026: Traced "5% of YouTube videos get 95% of the views" to Tubular Labs’ October 2015 count, reported by ReelSEO, which has since closed.
• Oct 15, 2026: Added the figures from YouTube’s old press page (over 2 billion logged-in users, a billion hours a day, 70% of watch time on mobile), with archived copies of the 2017, 2018 and 2020 versions; the page’s old addresses now redirect to the YouTube blog.
• Oct 16, 2026: Traced "how-to searches on YouTube grow 70% a year" to a May 2015 Think with Google article: one year of U.S. growth, early 2014 to early 2015.
• Oct 16, 2026: Added YouTube’s older statistics page (2012 to 2016 versions: 800 million and then over a billion users, 6 billion hours a month, almost 40% and then more than half on mobile, 40-minute mobile sessions), with archived copies. The "1.3 billion users" and "5 billion videos a day" figures credited to it are not on any version we read.
• Oct 19, 2026: Added the origin of "there are 51 million YouTube channels": a post by tubics, now offline, whose headline number changed every year and whose last version gave no basis for 51 million. Each version is linked through the Internet Archive.
• Oct 20, 2026: Added the source of "91% of smartphone users turn to their phones for ideas while doing a task" to the how-to section: the same May 2015 Think with Google article credits it to a March 2015 Google/Ipsos survey of 9,598 people online, separate from the how-to search data.
• Oct 20, 2026: Added the source of "nearly one in three millennials bought a product after watching a how-to video" from the same article: a Google Consumer Survey of 1,128 people in April 2015.
• Oct 20, 2026: Added the spring 2013 version of YouTube’s old statistics page, which paired "more than 1 billion unique users" with "over 4 billion hours" a month before the hours line moved to 6 billion in June 2013.
• Oct 21, 2026: The Cyprus University of Technology copy of the clickbait study ("The Good, the Bad and the Bait") went offline, so its figures now link to the Internet Archive’s copy of the same PDF; every figure was re-checked against it.
• Oct 25, 2026: Added the September 2015 version of YouTube’s old statistics page ("almost one-third of all people on the Internet"; mobile watch hours "up 100% y/y") and how lists changed it into "one-third have a YouTube account" and "mobile video consumption rises 100% every year".
• Oct 25, 2026: Traced "YouTube is the second largest search engine" to comScore’s August 2008 U.S. report, whose "YouTube/All Other" line counted all of Google’s non-search properties and which itself ranked Yahoo second.
• Oct 25, 2026: Removed five figures from Wistia’s 2024 and 2025 State of Video reports (play rates, contact-page videos, custom-thumbnail play rate and A/B testing). Wistia replaced those pages with its 2026 report and the Internet Archive no longer serves copies of them, so they can no longer be checked.
• Oct 25, 2026: Added the December 2014 and February 2015 versions of YouTube’s statistics page to the older-statistics section: the mobile line switched from “almost 40% of watch time” to “half of views”, so “more than half of watch time” is on no version.
• Oct 25, 2026: Added the March-August 2012 version of YouTube’s statistics page, “over 3 billion hours of video are watched each month”.
• Oct 26, 2026: Credited Alan Levine’s 2014 CogDogBlog search for the “60,000 times faster” line, including the 1986 study author’s reply that his research had nothing to do with processing speed.
• Oct 26, 2026: Moved the earliest known “60,000 times faster” to a June 1982 Business Week advertising section quoting Philip Cooper, found through Alan Levine’s CogDogBlog search.
• Oct 29, 2026: Traced "YouTube is the second most visited website" to Alexa’s top-sites list, closed since May 1, 2022, and added Similarweb’s and Semrush’s September 2026 rankings as current sources.
• Oct 31, 2026: Added when we read YouTube’s press page (October 2026) to the sentence on the totals it no longer gives.
• Nov 9, 2026: The '70% of YouTube watch time comes from recommendations' origin now cites Annotote's September 2017 verbatim quote of The Verge's feature, instead of the Verge article we had read through the Internet Archive although theverge.com bars AI agents.