Ideas
Why Streaming Algorithms Show You the Same 20 Things
By Jordan A. · · Updated · 5 min read
You open the app, scroll four rows, see the same twenty titles you saw on Tuesday, and close it. This is not your imagination and it is not a bug. A streaming home page is doing several jobs at once, and only one of them is showing you something you have not seen.
Here is what is actually happening, mechanism by mechanism, with the sources where they exist and an honest note where they do not.
1. The home page is a sales floor, not a library
Start with the part people find hardest to believe: on some platforms, prime placement is literally for sale. Roku sells a home screen unit called the Marquee ad, described on its own advertising site as the first thing viewers see when they turn the television on. That is a device home screen rather than an in-app row, and the distinction matters. Inside Netflix or Max, rows are personalised rather than sold to outside buyers. But the platform still has its own commercial priorities, and a service that spent nine figures on an original has a reason to put it in front of you more than once.
The honest version of this claim: some of what you see is bought, and much of the rest is promoted internally. Neither is a recommendation.
2. Recommendations carry more weight than anyone expected
Netflix's own product leaders wrote the standard account of this in a 2015 paper for ACM Transactions on Management Information Systems, and the headline number, quoted constantly since, is that recommendations influence around 80 percent of the hours people stream. Read that from the platform's side. If four fifths of viewing comes from the rows, the rows are the product, and the safest thing a row can do is show you something with a proven completion rate. Novelty is a risk. Repetition is not.
3. The same title, wearing a different face
You are not always seeing the same catalog. Sometimes you are seeing the same catalog redecorated. Netflix's engineering team published a detailed write-up of artwork personalisation explaining how the thumbnail for a given title is selected per viewer from a set of candidates, using bandit algorithms that learn which image gets a click from which kind of account. That is why a film can feel like a romance to one person and a thriller to another before either has pressed play.
It also means a title you have skipped ten times can reappear with a new image and read as new. It is not new. You have already voted on it.
4. Cold start, and why your taste barely moves
Cold start is a real and well-named problem in recommender systems: a system with little data about a new user or a new item falls back on population-level popularity. The consequence for you is that a fresh profile gets the mainstream, and here is the part that stings, a profile that has been used for years by three people in one household also gets the mainstream, because the signal is mush. Two adults and a nine-year-old sharing a profile produce an average that describes nobody.
If your rows feel bland, the most useful thing you can do costs nothing: give every person in the house their own profile.
5. The catalog outgrew the interface
The scale is the part people underestimate. Nielsen's State of Play report found that the number of video titles available across the US, UK, Canada, Mexico and Germany grew from 1.9 million in July 2021 to 2.7 million by June 2023. The same research put the average decision at 10.5 minutes per session, with one in five viewers abandoning the session entirely.
No interface made of horizontal rows can represent 2.7 million things. A home page can show you maybe forty. So the surface you scroll is not a view of the catalog at all. It is a very small sample, chosen for you, refreshed slowly.
What actually works
Separate profiles, as above. It is the single highest-return change and most households never make it.
Second, use the search box with a specific title rather than browsing. Search reaches the catalog directly. Rows do not.
Third, finish your list or delete it. A watchlist with sixty items on it is a to-do list you are avoiding, and it makes the decision harder every time you open the app.
Fourth, decide the shape before the title. Ninety minutes or six hours, subtitles or no subtitles, something that ended or something still running. Every one of those cuts the field faster than scrolling does.
Fifth, take recommendations from people. A friend who knows what you liked last year is a better system than any row, because they know what you are like on a Wednesday.
The short version
Your home screen repeats itself because it is optimised for a safe click, funded in part by promotion, redrawn with new thumbnails rather than new titles, blunted by shared profiles, and far too small to represent what is actually available. None of that is a conspiracy. It is just a different goal from yours.
If you want a recommendation built from what you love rather than from what a row is promoting, BingeNext's TV recommendations take shows you name and return picks with a reason and where each one streams. A run takes about a minute, and the free account covers three a day.
Topics covered: algorithms, streaming, opinion
About the author
Jordan A., Founder, BingeNext
Jordan A. is the founder of BingeNext and is responsible for every guide published on this site. He built the recommendation engine after one too many nights lost to the streaming menu. Dates, season counts and where-to-watch details in these guides are checked against official sources before publishing, and corrections are welcome at Jordan@Bingenext.com.