Your Streaming Algorithm Doesn't Know You Like That Guy at Blockbuster Did

Somewhere in America right now, a person is scrolling through Netflix for forty-five minutes and then giving up and watching The Office again. This is not a personal failing. This is a design outcome. The algorithm worked exactly as intended — it kept you on the platform, kept you comfortable, and successfully prevented you from discovering anything that might challenge your existing preferences or, God forbid, require subtitles.

We were promised that the internet would make finding great obscure movies easy. We were lied to. And the proof is in how much effort it takes to stumble across something genuinely unexpected in 2024 compared to a random Tuesday at your local video store in 1998.

The Rotten Tomatoes Problem Nobody Wants to Admit

Let's start with review aggregation, because it's the place where good intentions went most spectacularly sideways.

Rotten Tomatoes was supposed to democratize film criticism. Instead of trusting one powerful critic at the New York Times, you'd get a consensus score from dozens of voices. More data, better decisions. Sounds reasonable.

Here's what actually happened: studios started engineering their releases around the score. Films get a small preview screening, the score comes in, and if it's below a certain threshold, the marketing budget evaporates. The score becomes a self-fulfilling prophecy — low-scoring films don't get seen, don't get word-of-mouth, don't get the cultural oxygen they need to find their audience. Films that are genuinely strange and polarizing — the kind that score 58% because half the critics love them and half are baffled — get buried, even though a 58% on Rotten Tomatoes might actually indicate something far more interesting than a safe, crowd-pleasing 85%.

The algorithm doesn't know what to do with a film that divides people. It's optimized for consensus. And consensus, in art, is frequently the enemy of the remarkable.

What the Video Store Clerk Actually Did

Nostalgia is a liar, and we want to be upfront about that. Video stores were also full of bad recommendations, empty shelves, and late fees that felt like personal attacks. We are not here to pretend the past was perfect.

But here's what a good video store clerk actually provided that no algorithm has successfully replicated: contextual human judgment applied to a specific person standing in front of them.

When you walked into a good independent video store and told the person behind the counter that you'd just watched Blood Simple and your mind was broken in the best way, they didn't run a collaborative filtering model on your viewing history. They looked at you — a specific human being with a specific mood on a specific night — and said something like, "Have you seen Red Rock West? It's in the thriller section, second shelf from the bottom, and it'll ruin your week in the same way."

That recommendation came from a person who watched movies compulsively, who had opinions, who was wrong sometimes and spectacularly right other times, and who was present in the conversation. The interaction had stakes. It had personality. It could surprise you in directions the data didn't predict because human taste is not actually a pattern that resolves cleanly into a dataset.

How Streaming Platforms Trapped Us in Our Own Preferences

Every major streaming platform uses collaborative filtering as its recommendation backbone. You watched Parasite, so the algorithm shows you other Korean films, or other films about class tension, or other films that people who watched Parasite also watched. This sounds helpful. In practice, it creates a recommendation loop that gradually narrows rather than expands.

The problem is compounded by the sheer volume of content available. Netflix alone has thousands of titles at any given moment. When you can't find anything, you don't dig deeper — you retreat to the familiar. The algorithm reads this retreat as a preference signal and doubles down on familiar content. The loop tightens.

Compare this to a video store with 2,000 titles. The constraint was actually useful. You'd already seen the obvious stuff, so you'd wander into the foreign film section out of desperation and accidentally discover Kieslowski. Limitation forced exploration. Abundance enabled paralysis.

The Aggregator Trap and the Films It's Killing

Here's a specific, concrete example of how this plays out in practice.

When Uncut Gems came out in 2019, it opened with a 92% on Rotten Tomatoes and a B+ CinemaScore — which, for a film that genuinely makes you feel like you're having a panic attack for two hours, is remarkable. It found its audience. But think about how many films like Uncut Gems — formally aggressive, emotionally uncomfortable, genuinely unlike anything else — don't get that score and simply disappear.

A film like Possessor by Brandon Cronenberg scored well with critics but got almost no algorithmic push on streaming platforms because it didn't fit neat genre categories and its audience crossover with mainstream viewers was low. The algorithm looked at the data and said: niche. And so it stayed niche, even though the people who found it were evangelical about it in exactly the way that used to create cult classics.

The video store clerk would have hand-sold Possessor to every person who rented The Fly. The algorithm put it in a subcategory and moved on.

So What Do We Actually Do About It?

The video store isn't coming back. We've made our peace with that.

But the human curation model isn't dead — it's just migrated. The film Twitter/Letterboxd ecosystem, at its best, replicates the video store clerk dynamic. You follow people whose taste overlaps with yours in interesting ways, you see what they're excited about, you take the leap on something you'd never have found otherwise. Letterboxd lists are the new Staff Picks shelf.

Independent film publications — including, we will immodestly note, this one — exist precisely to do what the algorithm won't: tell you that a film is worth your time even if the data says it's not for you. Especially then, actually.

The discovery problem is real. The solution is the same one it's always been: find someone with good taste and listen to them. The medium has changed. The principle hasn't.