Why your streaming recommendations are bad, and how to fix them
Why your streaming recommendations are bad, and how to fix them
Streaming recommendations are mostly bad, and it's not because the algorithms are dumb. The algorithms are very good at their actual job, which is not the same as your job. Understanding the gap is the difference between a homepage that feels like it's reading your mind and one that feels like it's stuck on repeat.
The algorithm isn't optimizing for you
Every major streaming service optimizes for retention and time on platform. A recommendation that keeps you watching another episode is, from the platform's perspective, a perfect recommendation, even if it's the same comfortable show you've already seen three times. The system learns that you finish familiar things, so it shows you more familiar things. That's why your Netflix home screen slowly narrows over time until it's mostly sequels, comfort rewatches, and one genre you clicked once at 2am.
Your goal — watch something you'll actually love, maybe something new — is only loosely aligned with the platform's goal. The algorithm is not malicious; it's just working for the other side.
Your data is shallow
The other half of the problem is the input. Most platforms infer your taste from what you clicked, what you finished, and what you skipped after ten minutes. That's behavior, not preference. Finishing a show you had on while folding laundry is not the same signal as loving a show. Skipping a slow-opening movie is not the same as disliking it. The system can't tell the difference, so it averages you into a blurry middle.
What it's missing is explicit, rated taste: not "you watched this" but "you loved this, you liked this, this was fine, you disliked this." Four buckets, applied consistently, is dramatically more signal than a year of watch-time telemetry. It's also the thing platforms are worst at collecting, because rating takes effort and they've trained you not to bother.
Recency and rewatch distort the picture
Two quieter biases make it worse. Recency bias: the system weights what you watched lately, so a two-week true-crime bender can convince it you've become a true-crime person for months. Rewatch bias: because you reliably finish shows you already love, the engine keeps surfacing them, crowding out anything new. Both come from optimizing for "will they keep watching" rather than "will they be glad they watched."
How to actually fix it
The good news is you don't need a better algorithm; you need better input and a recommender that's on your side.
Rate, don't just watch. Even a loose loved / liked / meh / disliked on everything you finish gives a real taste profile that behavior alone never will. It takes a second per title and compounds fast.
Log the misses too. Marking something "not interested" or noting a disliked title is negative signal, and negative signal is what stops a recommender from surfacing the genre you're politely done with. Platforms collect almost no negative signal, which is why they keep suggesting the same things.
Keep your history in one place, not siloed per service. If your taste lives only inside Netflix, Netflix recommends like a Netflix machine. If it lives across everything you watch — Netflix, Max, the indie film you rented, the show a friend recommended — a recommender can actually see the whole you, and its suggestions stop being platform-shaped.
Use a recommender that isn't trying to keep you on one platform. This is the core fix. A recommender whose only job is to find you something you'll love, wherever it lives, has no incentive to narrow you. It can suggest the obscure thing, the older thing, the thing on a service you forgot you had.
That's the idea behind FleckFlix: your whole watch history and your ratings in one place, and recommendations tuned to what you'll love rather than to keeping you subscribed. Better input, aligned incentives, better picks.
