What Netflix’s patents reveal about the future of watching movies
I read through almost 500 Netflix patents to look at what the streaming has been tackling, building and disrupting.
Netflix’s patents offer a rare glimpse into the future that major streamers are building for film and television. While much of the industry debates box office figures and windowing, Netflix’s filings show where the technology behind viewing, making, and marketing films is heading next.
For filmmakers, these details reveal not just what audiences will expect, but how future workflows and releases will be shaped.
Almost half of Netflix’s patents are around how to store, retrieve and deliver the content. The second most patented category is that of content recommendations.
In order to give you a glimpse into what Netflix has discovered (and now controls), I have broken down the most interesting patents into four groups:
What we’re offered. The next generation of recommendation engines decides what viewers see and how films get discovered.
How it’s made. AI and automation are reshaping how films are edited, localised, and marketed.
How Netflix looks. The interface itself is a new creative layer, with personalisation and interactivity built into how films appear to each viewer.
Behind the scenes:. New infrastructure and delivery methods are changing expectations for quality, speed, and security.
Let’s look at each in turn:
1. What we’re offered
Netflix’s recommendation engine has been key to their growth right from the start.
By 2017, Wired was reporting that Netflix’s recommendation engine was driving over 80% of all viewing on the platform. The system learns from each user’s viewing habits, device use, and even the time of day, adapting suggestions in real time
Netflix has spent considerably on developing and patenting aspects of their recommendation engine, including:
New recommendation models are faster and more open
Netflix’s latest models update themselves in hours, not months, and their simpler maths makes results easier to check and test [CA3118083C].
Launching a new title with no history
Netflix uses metadata and similar shows to connect new releases with the most likely audience from day one [EP2813990A1].
Group recommendations, not just single picks
Netflix now recommends sets of titles likely to interest certain groups or demographics, making discovery feel more curated [US8903834B2] [US10698909B2].
Your skips and ignores are signals too
Ignoring a show or movie tells the system as much as picking one, so the next round of recommendations gets smarter [US10482519B1].
Netflix is clearly trying to tackle the recommendation bubble, in which you feel you’re only seeing a narrow selection of titles. This includes:
What you see on the homepage (and what plays next) is tailored
Netflix decides in real time which previews and rows of titles show up first, changing layouts based on your browsing and watching patterns [AU2012294442B2] [US10129596B2].
What you watch right now changes what shows up next
Netflix updates suggestions based on actions you take in a single session, adjusting picks on the spot to fit your mood [US20230388596A1] [WO2018009681A1].
Fighting the echo chamber
Netflix’s system can spot and reduce repetition in recommendations, working to keep your feed from going stale or narrowing over time [US20220180186A1] [WO2024249880A1] [US20240403713A1].
More than “Because You Watched”
Netflix’s system now looks for novelty, not just similarity. Recommendations push viewers toward new or unexpected titles instead of just more of the same [US11017024B2].
Another way they try to make good recommendations is to learn more about who’s watching:
Local tastes guide what’s recommended
Recommendations are tuned by country and region. A hit in the US will not automatically fill up homepages around the world [US20200133971A1].
No more guessing who is watching
The system figures out who is watching even if households share accounts, so personalised picks still reach the right viewer [AU2016202876B2].
The factors Netflix says it uses when calculating recommendations are:
Viewing history and watch time
Titles skipped, ignored, or abandoned
User interactions: clicks, hovers, scrolls, searches
Ratings and thumbs up/down
Similarity to other users’ preferences
Metadata tags: genre, cast, crew, themes, mood
Session context (i.e. time of day, device, current browsing behaviour)
Recent and trending titles
Regional and local viewing trends
Novelty preference and past openness to new content
Direct search intent and results
Household and profile identity (to personalise in shared accounts)
2. How it’s made
Netflix’s way of working has changed how shows are made, marketed, and released across the industry.
Decisions on which shows to renew are now based on how many viewers finish a season, not just how many start it. Netflix runs constant A/B tests on everything from artwork to episode order, including experiments like serving episodes of Love, Death & Robots in different sequences to see what keeps viewers watching longer.
But it goes way beyond just what to commission or the order in which to present it. Netflix has brought automation and data science into every step of content creation.
AI selects promotional artwork and generates trailers by identifying key scenes, reducing the need for manual editing. Tools for localisation, such as AI-powered dubbing and subtitles, allow for global releases that reach audiences as soon as a title drops, as seen with Squid Game and Money Heist. Data-driven marketing then refines artwork and campaigns in real time, with Netflix reporting up to 30% higher engagement for optimised versions.
Netflix is using AI and machine learning to create and alter existing content, such as:
AI-generated trailers and previews
Machine learning picks out the key moments from films and shows to automatically build trailers, highlights and promotional clips, cutting down on manual editing. [US11350169B2] [AU2020388552B2] [US11604935B2]
Automated editing via match cuts and scene analysis
Systems suggest possible match cuts and transitions by analysing video content and shot changes, helping make editing faster and more consistent. [US20230147904A1] [US10686969B2]
Automatic VFX and object tracking
Editing tools can track and isolate objects in video frames using splines, making VFX and background replacement much faster and less manual. [US20230064431A1] [EP4396776A1] [US20240202932A1]
Keep reading with a 7-day free trial
Subscribe to StephenFollows.com - Using data to explain the film industry to keep reading this post and get 7 days of free access to the full post archives.