Related reading
Netflix knows things about me that I've never told anyone.
It knows I once watched three seasons of a show about cakes in a single weekend. It knows I've started the same movie four times and never made it past the first twenty minutes. It knows what I watch when I can't sleep (British baking shows) and what I watch when I'm sad (anything with Keanu Reeves, apparently). It's been collecting data on me for years.
So why does my "Recommended For You" section look like it belongs to a completely different person?
I'm staring at it right now. True crime documentary. Gritty thriller. Prestige drama everyone's talking about. Korean action movie. None of this is for me. I know it's not for me. Netflix should know it's not for me. And yet here we are.
I finally got curious enough to actually think about why this happens. And the answer, I'm sorry to say, is kind of annoying.
Why it can feel like the algorithm wants you to keep browsing
I can't know Netflix's internal objectives, but recommendation systems are usually optimized for measurable signals: clicks, watch time, starting new titles, browsing.
Those signals often correlate with satisfaction, but they aren't the same thing. "Browsing" can show up as engagement in the metrics even when you're annoyed and about to give up.
To be fair, sometimes these systems do help you find something great quickly. But the incentives can still tilt toward showing you more options, more often, instead of helping you commit to one.
"Because you watched X" can be misleading
The basic idea behind recommendations seems reasonable: you watched movie A, so you might like movie B because they're similar. Makes sense, right?
The problem is "similar" is doing a lot of heavy lifting there.
The algorithm often matches on stuff that's easy to measure. Same genre. Same actors. Same decade. Same rating. These are categories a computer can sort into buckets. But they're often not the things that actually made you enjoy something.
I loved The Social Network. Not because it's a "drama" or because it's "David Fincher" or because it has Jesse Eisenberg in it. I loved it because of this specific energy it has, this sense that building a website is life or death, this tension between people who used to be friends. The dialogue felt like a knife fight. The score made me anxious in a good way.
How do you categorize "makes building a website feel like a heist" in a database? You don't. So the algorithm looks at surface features and recommends other dramas, other David Fincher movies, other Jesse Eisenberg vehicles. Some of those might work. A lot of them won't. Because they're not matching on what actually mattered.
There's also a popularity problem. Algorithms recommend popular stuff because they have more data on popular stuff. If a million people watched something, Netflix knows a lot about who liked it. If only ten thousand people watched something, there's not enough data to work with.
So your "personalized" recommendations can end up being... popular things, slightly filtered. Look at your homepage right now. How much of it feels genuinely personalized versus just "trending this week"?
They don't know what you want right now
Even if the algorithm perfectly understood my taste in general, that wouldn't tell it what I want tonight.
What I want on a Friday with my girlfriend is different from what I want on a Tuesday alone. What I want when I'm exhausted is different from what I want when I'm restless. Sometimes I want to think. Sometimes I actively don't want to think. Sometimes I want to cry. Sometimes I'd rather die than watch something sad.
Mood matters. Energy matters. Context matters.
The algorithm knows my history. It doesn't know my state. So it just averages everything out and shows me "stuff you usually watch." Which might be exactly wrong for how I'm feeling right now.
A streaming service can "know" you love thrillers but keep recommending them when what you actually want is something comforting and easy. The algorithm sees your past. It doesn't see your present.
What would actually help
I've thought about this probably too much. A recommendation system that actually wanted to help people find movies they'd love would need to do a few things differently:
| What Algorithms Do | What Would Actually Help |
|---|---|
| Track what you watched | Track what you loved |
| Match surface features (genre, actors) | Understand why something worked for you |
| Show the same recommendations anytime | Ask what you're in the mood for right now |
| "Because you watched X" | Explain the specific qualities being matched |
Care about what you loved, not just what you watched. I've finished plenty of movies I didn't like. I've rewatched my favorites dozens of times. A system that treats both the same is missing the point. What you loved matters. What you merely consumed doesn't.
Understand why you loved it. "I liked Inception" could mean a lot of things. Mind-bending plots. Christopher Nolan specifically. Leonardo DiCaprio. Heist movies. Movies that look expensive. A good recommendation system needs to figure out which layers actually matter to you. Otherwise you just get surface-level "more of the same."
Ask about right now. Are you watching alone or with someone? Do you want something intense or something chill? Are you in the mood to laugh or think or feel? The best recommendation for you is different depending on these answers.
Explain the reasoning. "Because you watched Stranger Things" tells me nothing. I watch lots of things for lots of reasons. But "you loved 80s nostalgia with genuine scares and ensemble casts, and this has all three" tells me something I can actually evaluate.
So what do you do
Netflix may change over time, but engagement signals are deeply baked into streaming products. They're measurable, they move business metrics, and they don't always line up perfectly with "I loved this."
But you don't have to accept their recommendations as the only option.
You could ask friends. You could read reviews. You could browse with more intention instead of just scrolling and hoping.
Or you could try something built with a different goal: helping you decide faster. Different goals lead to different recommendations.
The algorithm knows what you've watched. It doesn't know what you've loved, why you loved it, or what you're in the mood for tonight.
Until something actually solves those problems, "Recommended For You" is always going to feel a little bit like it's meant for someone else.
Because, honestly, it kind of is.
A 15-minute recommendation audit
If recommendations keep missing, run this once and you will usually see why:
- Open one recommendation row and choose three titles you are likely to skip.
- For each title, write one reason it was shown (actor, franchise, genre, popularity, recent watch).
- Compare those reasons to what you actually want tonight.
When those two lists differ, stop browsing and switch to intent-based search (for example: "slow thriller under two hours," "light comedy with no tragedy ending," or "rewatchable comfort drama").
