Since 2023, Netflix has published bi-annual engagement reports. Each report lists every title on the platform and how many hours subscribers spent watching it. The data is granular but raw — titles, hours, nothing more. As a long-time user and fan of Netflix, I’ve always wanted to dive into their reports for years, but never had the time to wrangle mountains of raw data. AI changed that, making it possible to finally crack open the numbers.
To see how Netflix shapes global viewing, I used Claude Code to build a metadata enrichment pipeline across six bi-annual reports (2023–2025), leveraging TMDB, IMDb, and LLMs to generate metadata tags for 28,000 titles and 560 billion hours of viewing, followed by Model Context Protocol (MCP) for exploratory analysis. What follows are the insights that emerged.
Netflix is fundamentally a television platform. In 2025, shows generated 74% of all viewing, gaining roughly a percentage point from movies over the last three years. Total platform hours grew modestly from 93.5 billion in H1 2023 to 96.2 billion in H2 2025, with series accounting for all of that growth.
Yet subscriber growth tells a different story. While paid subscribers surged 36% (from 238M to 325M), total viewing hours grew just 3%. That divergence means the average subscriber is watching meaningfully less than before, and a couple of factors appear to be driving it.
The first is the composition of the subscriber base. Netflix’s ad-supported tier now accounts for 50% of all new subscribers as of 2025, up from roughly 20% in 2023. Ad-supported subscribers watch approximately 41 hours per month on the platform, compared to an implied platform-wide average of around 49 hours, suggesting premium subscribers watch meaningfully more. As the ad-supported tier represents a growing share of the base, the per-subscriber average is pulled down.
The second is that competition for viewing time has intensified, particularly from short-form content. YouTube held 11.1% of total US TV viewing time in December 2024, according to Nielsen’s The Gauge, compared to Netflix’s 8.5%. By December 2025, YouTube had grown to 12.7% while Netflix reached 9.0%. YouTube gained 1.6 percentage points in a single year against Netflix’s 0.5. Short-form and free video is capturing attention at a faster rate than premium streaming is growing it.
Over the last three years, original content accounted for an average of 51% of all platform views, with shows at 65% and movies at 41%. For shows, original share dipped to 57% in H2 2024 before recovering to 68% by H2 2025. This dip maps directly to the 2023 Hollywood strikes. Production delays created a content vacuum that materialized on screen by H2 2024. Licensed catalog stepped in to fill it, and what it revealed in doing so is how much structural weight that catalog actually carries. For movies, the strike’s impact was equally visible: original movie share fell to 33% in H2 2024 before recovering to 43% in H1 2025 and reaching 52% in H2 2025, crossing above licensed movie share for the first time, driven significantly by KPop Demon Hunters generating 20% of all movie views in that period alone.
The sheer volume makes the performance gap stark. In 2024, Netflix added 1,090 licensed shows versus 367 originals — yet originals drove 55% of new show viewing. For movies, the ratio was even more lopsided: 1,640 licensed films to 104 originals, with originals still capturing 36% of new viewing from just 6% of the titles. Originals punch well above their weight regardless of content type.
The gap between original and licensed content performance is most visible when looking at new releases. In 2025, Netflix released 445 original show titles and 332 original movies. New licensed content added the same year was nearly twice the volume: 1,017 licensed shows and 979 licensed movies. Yet original content generated 72% of new show views and 72% of new movie views. Across both content types, originals represented one third of new titles but drove nearly three quarters of new content viewing.
Drama dominates, accounting for 35% of all show views. Notably, just five licensed medical dramas (Grey’s Anatomy, The Resident, House MD, New Amsterdam, and The Good Doctor) generated 8.4 billion hours — 6% of all show viewing over three years. That is more than any genre other than drama itself, from just five shows, none of which are Netflix Originals.
The mismatch between catalog volume and audience demand is stark: Action & Adventure make up 4% of show titles but 8% of views, while Crime captures 7.5% of views from just 5% of titles. Conversely, Documentary and Reality represent 18% of titles combined but only 9.6% of views. The pattern suggests that audiences want more action and crime than Netflix offers, and less documentary and reality than Netflix produces.
Movies reveal a starker version of the same imbalance. Action movies represent 11% of the movie catalog but 20% of movie views. Animated movies account for 5% of titles but 11% of views. Drama and Comedy movies together represent 46% of the entire movie catalog but generate only 28% of movie views. Nearly half the movie library is concentrated in the two genres audiences engage with least proportionally.
For movies, the titles that break through to become the platform’s most watched each period are almost exclusively Netflix Originals. Every single top-performing movie in every period I analyzed was produced by Netflix. Their IMDB ratings are consistently modest, clustering between 5.9 and 7.4. Critical quality rating has almost nothing to do with which movies reach the largest audiences. What drives top movie performance appears to be event status: marketing, star power and cultural timing. KPop Demon Hunters, released in June 2025, sits in a category of its own. As the father of a 6-year-old girl who has watched it well over 10 times, owns a small mountain of merchandise, and forced me to listen to its soundtrack for hundreds of hours, I didn’t need data to tell me this movie was massive, but the numbers are staggering anyway. The animated musical generated 20% of all movie views in H2 2025, the largest single-period movie performance in the dataset by a substantial margin. It won the Oscar, Golden Globe, Grammy, Critics Choice and Annie Award simultaneously. The formula it found, family animation built around KPop as its cultural engine, drew audiences across age groups and language barriers in a way that conventional animated films rarely achieve.
One format trend worth noting is the steady rise of limited series. Total titles increased by 52% from 609 in H1 2023 to 923 in H2 2025, compared to a 9% decline in regular titles from 3,660 to 3,327 over the same period. Their share of platform views grew from 5.6% in H2 2023 to 7.1% in H2 2025, driven by Korean dramas and original titles including Baby Reindeer, Adolescence, Monsters and Fool Me Once, which demonstrate that compact self-contained formats can generate substantial viewing when the storytelling is strong.
Three patterns in the data point to content decisions that appear more deliberate than they might look from the outside.
The new season effect
When Stranger Things released its fifth and final season in H2 2025, it generated 879 million hours of new viewing. Prior seasons generated 1.2 billion hours in the same period, more than Season 5 itself. When Wednesday’s second season arrived, prior season viewing jumped 254%. When Squid Game released Seasons 2 and 3 in consecutive periods, prior season viewing outpaced the new seasons by more than two to one.
Tentpole new seasons act as a triple catalyst. They generate algorithmic buzz for new subscribers starting from Season 1, nudge hesitant viewers off the fence, and act as a direct retention driver for loyal fans. This extends a new season’s value well beyond its fresh episodes, turning it into a powerhouse engine for back-catalog viewing.
Licensed content decays and the return rate Is low
Major licensed shows generate an immediate surge, then decay fast. Within a year of arriving in H2 2024, Prison Break lost 87% of its peak audience. Lost followed a near-identical trajectory. Suits peaked at 0.60% of platform views in H2 2023 and retained just 0.18% two years later, a 70% decline. The pattern is consistent: licensed content arrivals generate strong initial engagement that normalizes quickly, and most subscribers who have watched a show do not return to it.
The picture looks different for shows that were already on Netflix before our dataset begins. Narcos, The Walking Dead, The Blacklist and Breaking Bad all show stable catalog engagement with no meaningful decay, hovering at consistent view levels across all six periods. These shows have already found their settled catalog audience. They are not growing, but they are not declining either. The most likely explanation is a combination of new subscribers discovering these shows for the first time, driven by strong reputations, algorithmic recommendation and Netflix’s continued expansion into new markets, alongside a smaller proportion of existing subscribers returning to rewatch. For Netflix, this long-tail catalog stability represents a quiet but durable source of engagement that requires no new investment to maintain.
IMDB rating matters for licensed shows, not for licensed movies, and not at all for kids
As someone who tracks IMDB ratings closely, I’ve developed a personal heuristic over the years. For movies, anything above a 7 is usually worth your time. For series, the bar sits higher at an 8. When testing that against Netflix’s data, it turns out the platform agrees, at least for shows. For licensed English-language series, quality is a reliable compass, following a clear staircase upward. Viewership climbs steadily from around 4 million views for shows in the 6.0 range, crossing 13 million for the 8.0 tier, and more than doubling again for elite shows rated 9.0 and above. Higher-rated licensed shows clearly draw significantly larger audiences.
For licensed English-language movies, the picture is strikingly different. Average views are essentially flat at 3.6–4.2M across all bands from 5.0 to 8.4. Only at 8.5 and above does a meaningful breakout occur, driven by genuine cinema landmarks like *Interstellar, Gladiator* and *Back to the Future*, alongside *Spider-Man: Across the Spider-Verse* which at 39.2M views is the standout performer in that band.
For kids content, IMDB rating has no predictive value whatsoever, and the data reveals why. *CoComelon*, rated 3.4 on IMDB, is one of the most-watched shows in the Pre-School category. *The Boss Baby*, rated 6.3, generated 125.3M views. *Despicable Me*, rated 6.2, generated 107.1M. *Peppa Pig*, rated 6.2, and *PAW Patrol*, rated 6.1, are two of the most consistently watched Pre-School shows in the dataset despite ratings that would place them firmly in the middle of the general audience distribution. *Shrek*, rated 7.9, generated 92M views, barely more than *Woody Woodpecker Goes to Camp*, rated 4.5, which generated 80.4M. And *KPop Demon Hunters*, winner of the Oscar, Grammy and Golden Globe and the biggest film in the dataset by views, carries a 7.4 IMDB rating. A cultural phenomenon that broke every Netflix film record is rated lower than thousands of adult titles that generated a fraction of its audience. The chart for kids content is not a staircase. It is a random walk.
The reason is structural. IMDB ratings reflect adult opinions. The actual audience for Pre-School and 6–12 content, children, cannot rate what they watch. Parents who do rate kids content apply adult sensibilities to programming designed for four-year-olds, systematically underrating material their children may watch dozens of times. For kids content, engagement metrics are the only reliable signal. IMDB is measuring the wrong audience entirely.
Content produced in 126 countries across 86 languages appeared in Netflix’s engagement data between H1 2023 and H2 2025. South Korea’s 1,423 series, representing approximately 4% of the total show catalog, generated 8.9% of all platform viewing across the three years, comparable in scale to the entire output of the United Kingdom. Korean content topped the original show rankings in three of the six periods analyzed. In H1 2025, Squid Game’s back-to-back Season 2 and Season 3 releases drove Korean content above 10% of platform viewing, briefly making it the second largest content category by origin, behind only the United States.
Non-English content overall grew from 31% of platform viewing in H1 2023 to a peak of 35% in H1 2025, before settling at 33% in H2 2025. Netflix management has itself confirmed that more than a third of viewing comes from non-English programming, a figure the engagement data validates.
Japanese content tells a quieter story: anime grew its share of viewing in every single period, rising from 4.7% in H1 2023 to 6.3% in H2 2025 — the only language with uninterrupted growth. Meanwhile, Spanish content fluctuated with breakout hits, and Hindi content hovered around 1% but climbed steadily with India’s subscriber base.
The depth of engagement with Korean content reveals something about who is actually watching it. Korean shows generate significantly more hours per view than English shows, consistent with viewers outside Korea making the same long-form commitment to 16-episode drama seasons that Korean domestic audiences do. These are not casual samplers of foreign content but viewers for whom K-drama has become a preferred format.
KPop Demon Hunters adds an interesting dimension to this story. Produced in the United States and classified as English-language content, its cultural engine is entirely Korean. It is the clearest example in the dataset of Korean cultural influence operating through an English-language vehicle, and perhaps a preview of how global content increasingly resists simple language categorization.
What the data reflects is a deliberate bet that paid off. Netflix identified the strength of Korea’s mature television industry, its storytelling traditions, production quality and passionate domestic audience, and gave it global distribution. The result is a structural shift in how entertainment travels across language and cultural boundaries. No traditional broadcaster or previous streaming service had achieved anything comparable: content produced in one language, in one cultural context, generating nearly 9% of global viewing hours on a platform with 325 million subscribers. The same playbook is now unfolding, more slowly, in Japanese anime and potentially Hindi and Turkish content. Whether those markets follow the same trajectory will be one of the defining questions for the platform.
Kids and family content, titles targeted at Pre-School and 6–12 audiences, represented 21% of all platform views in H2 2023. By H1 2025, that share had fallen to 19%. The apparent recovery to 20% in H2 2025 is almost entirely attributable to KPop Demon Hunters, which added two percentage points to the kids category share on its own. Removing it, kids content share in H2 2025 would have been 18%, the lowest in the dataset.
The franchise-level data reveals steady lifecycle decay as core audiences age out. For instance, CoComelon (the largest Pre-school franchise in 2023 at 1.1 billion hours) dropped 57% to 468 million hours by 2025 despite continuous new releases. Masha and the Bear, Barbie, The Boss Baby, LEGO, Teen Titans and Sonic the Hedgehog all follow a similar arc, losing between 40% and 65% of their 2023 viewing levels by 2025.
The exceptions are instructive. The franchises that grew or held stable across the period did so almost exclusively because of specific new content releases, not organic platform momentum. Despicable Me more than doubled its 2023 hours by 2025, driven by Despicable Me 4 arriving on Netflix. SpongeBob SquarePants grew 78% on the back of Saving Bikini Bottom. KPop Demon Hunters generated 864 million hours across H1 and H2 2025 from a standing start, but it is a single film, not a franchise. PAW Patrol is the rare exception that grew organically, sustained by genuine broad appeal across Pre-School viewers and family co-viewing rather than dependence on a single release.
The implication is that kids content on Netflix is not in structural decline so much as it is in content lifecycle decline. The franchises that dominated in 2023 are aging. The platform has not yet developed the next generation of replacement IP at comparable scale. KPop Demon Hunters is what a genuine breakthrough looks like, a film that transcended the kids category entirely, drawing family audiences across age groups simultaneously.
The data makes a strong case for Netflix’s two-engine content model, and the writers strike of 2023 functioned as an unintentional stress test of it.
Originals are the ignition. They create the cultural moments that drive subscriber acquisition, generate disproportionate viewing relative to their volume, and win every top movie slot in the dataset. In 2024, original shows represented 25% of new show titles but generated 55% of new show viewing. No licensed show has topped the platform charts in any of the six periods analyzed.
Licensed content is the fuel. Existing licensed catalog consistently accounted for the majority of platform viewing in every period, the single largest bucket, period after period, regardless of what was new. When original content dried up in H2 2024 due to the strike, licensed catalog did not merely fill the gap. It revealed how load-bearing it actually is. A platform running on originals alone would have experienced a significant engagement decline during the strike. It did not, because licensed content absorbed the shortfall.
Neither engine alone is sufficient. Originals without a deep licensed catalog produce cultural moments with nothing sustaining engagement in between. Licensed content without originals produces a comfortable but culturally invisible platform, fine for retention, insufficient for growth.
The language diversification story, measured purely by engagement outcomes, represents one of the more remarkable shifts in entertainment consumption in recent memory. Content produced in 126 countries reaching a genuinely global audience, through repeated, deep viewing of Korean dramas and Japanese anime, suggests the platform has enabled storytelling traditions to find audiences that had no meaningful route to them before streaming.
The question the data raises most directly is about the next generation of viewers. Kids content is losing share on every measure across every period, without the pipeline of new franchise IP that would suggest a reversal is near. If that decline reflects a generation forming its entertainment habits on short-form video rather than long-form streaming, no amount of catalog management addresses it. Will Netflix’s next frontier be short-form content?
Data sourced from Netflix’s bi-annual engagement reports (H1 2023 to H2 2025), enriched with metadata from TMDB, IMDB ratings data, and Claude AI for genre, age segment, franchise, and content classification. Subscriber figure of 325 million paid memberships from Netflix’s Q4 2025 shareholder letter. Content metadata is AI-derived and directional in nature. Nielsen data from The Gauge, December 2024 and December 2025. Ad-supported tier viewing hours from Netflix upfront presentation, May 2025. Platform-wide average of 49 hours per subscriber per month calculated from Netflix engagement reports and subscriber disclosures. All analysis is independent and not affiliated with Netflix.
Inside Netflix’s Content Machine: Three Years of Engagement Data Decoded was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.