# Recommendation Engines Want You to Finish. Your Taste Wants to Wander

**Published:** July 5, 2026
**Author:** Achriom
**URL:** https://www.achriom.com/blog/the-recommendation-engine-wants-you-to-finish

> Streaming recommendations feel worse every year because they optimize for completion, not taste. Here is what the algorithm is really doing, and how to wander instead.

**Tags:** cross-media, ai, recommendations

---

Streaming recommendations feel worse every year for a simple reason: the engine is not trying to understand your taste. It is trying to keep you in the app. Those are different goals, and the gap between them is what you feel when the "For You" row shows you the same five things in a new order.

A recommendation engine wants you to finish the episode, start the next one, and stay until the autoplay countdown runs out. Your taste wants something else. It wants to wander. It wants to follow a mood out of a song and into a film, out of a film and into a book you would never have found by staying inside one platform.

This post is about that gap, why it exists, and how to recommend for yourself again.

## What a recommendation engine is actually optimizing for

Before you judge a recommendation, it helps to know what it was built to do. Almost every mainstream engine is measured on the same handful of numbers.

**Time on platform.** The core metric. A good recommendation, by the system's definition, is one that produces another minute of watching or listening. Whether you loved it barely registers.

**Completion rate.** Did you finish. Engines learn to feed you things you are statistically likely to complete, which pushes hard toward the familiar and the frictionless. Difficult, strange, or slow work gets buried because it lowers the average.

**The next slot, not the next chapter of you.** The engine fills a slot. It has no model of where your taste is heading over a year, only what keeps the current session alive.

**One library, one silo.** This is the quiet limit. Netflix cannot recommend a book. Spotify cannot recommend a film. Each engine only sees the catalog it sells, so it can never follow a thread that leaves its walls.

Hold those four in mind and the "bad recommendation" stops feeling like a bug. The engine is doing exactly what it was asked to do. You just wanted something it was never built to give.

## How each engine narrows you

The specifics differ by platform, but the direction is the same. Every one of them pulls you toward the middle.

### Netflix

Netflix optimizes for the next autoplay. It rewards shows that binge cleanly and titles that its own data says people finish. That is why the homepage drifts toward interchangeable thrillers and why a slow, singular film can vanish from your rows even after you rate it highly. Completion is the god metric, and completion loves the predictable.

### Spotify

Spotify is built to keep the audio running. Its recommendations lean on acoustic similarity and mass listening patterns, so it is excellent at finding you one more song that sounds like the last one and poor at the leap that changes how you hear music. Endless similarity is the point. The playlist never wants to end, so it rarely wants to surprise you.

### YouTube

YouTube optimizes watch time above nearly everything, which is why the sidebar keeps escalating: louder, longer, more extreme versions of whatever you clicked. It is a machine for the next click, not the right one. Follow it long enough and your feed narrows to a single loud lane.

### TikTok

TikTok's feed is the purest version of the pattern. It reads micro-signals (a half-second of hesitation, a rewatch) and serves more of exactly that. It is astonishingly good at holding attention and nearly useless for building taste, because it optimizes for the reflex, not the person having it.

### Goodreads and Amazon

Even books get the retail treatment. "Readers also bought" is a sales signal wearing a recommendation's clothes. It surfaces what sells alongside what you bought, which is why the suggestions cluster around bestsellers and series you have already finished. It tracks the market, not your mind.

None of these are broken. They are working. But not one of them can see the whole of what you love, and none of them is measured on whether you were glad afterward.

## A thread the algorithm cannot see

Here is the kind of connection no single-platform engine can make, because it crosses four media in one breath. Follow a feeling instead of a catalog and this is what wandering looks like.

### Blue: Joni Mitchell

![Blue (1971)](/blog/assets/the-recommendation-engine-wants-you-to-finish/blue-album.jpg)

Start with *Blue*. It is an album about leaving and the ache of freedom, sung so plainly it sounds like someone talking to you at a kitchen table. Spotify will hand you ten more folk records that sound like it. Useful, but that keeps you inside the sound. The interesting move is to leave the sound and keep the feeling.

### Paris, Texas: Wim Wenders

![Paris, Texas (1984)](/blog/assets/the-recommendation-engine-wants-you-to-finish/paris-texas-movie.jpg)

The feeling in *Blue* lives inside *Paris, Texas*: a man walking out of the desert with no words for where he has been, a story about distance and the people we leave. No music engine would ever send you here from an album, because it is a film, and the film lives in a different silo. But the emotional line is unbroken. Same longing, different medium.

### The Left Hand of Darkness: Ursula K. Le Guin

![The Left Hand of Darkness (1969)](/blog/assets/the-recommendation-engine-wants-you-to-finish/the-left-hand-of-darkness-book.jpg)

From the loneliness of that landscape, the thread runs to *The Left Hand of Darkness*, Le Guin's novel about a stranger crossing a frozen world, learning to trust someone genuinely unlike him. It is science fiction, so no film algorithm points you toward it, yet it is the same subject seen from another angle: what it takes to reach another person across a distance that feels total.

### The Leftovers

![The Leftovers (2014)](/blog/assets/the-recommendation-engine-wants-you-to-finish/the-leftovers-tv.jpg)

And the thread closes on *The Leftovers*, a series about people living inside an unexplained loss and finding, slowly, that connection is the only thing that answers it. It rhymes with everything before it: the leaving in *Blue*, the wordless distance in *Paris, Texas*, the crossing in Le Guin.

Four works, four media, one continuous feeling. No streaming engine could have drawn that line, because no streaming engine can see across the wall between music and film and books and television. The line was always there in your taste. The tools just could not follow it.

## What it takes to recommend across the wander

If the problem is that every engine is trapped in one silo and pointed at retention, the fix has a clear shape. A recommendation worth trusting needs three things a streaming feed does not have.

It needs to see everything you love in one place, not one catalog at a time. It needs a reason to suggest something other than keeping you engaged. And it needs to reason about meaning, not just similarity, so it can jump from an album to a novel when the feeling connects.

That is the machine Achriom is built to be. Your books, films, music, TV, and anime live in one library, and the AI librarian reasons across all of it. It has no watch-time number to protect and no ad slot to fill, so when it suggests the next thing, the suggestion follows your taste rather than a retention curve. You can ask it why a film connects to an album you loved, and it will trace the thread instead of handing you ten lookalikes.

<div class="blog-inline-cta">
<p><strong>Want all of it in one place?</strong> Achriom tracks what you wander toward alongside your books, films, music, TV, and anime, with an AI librarian that finds the threads between them. That is the part no single-format tracker can do.</p>
<a href="https://app.achriom.com" data-cta="blog-inline-cross-media">Try Achriom free →</a>
</div>

## Engine by engine: what it gives, what it costs

| Engine | Optimizes for | What you get | What it cannot do |
|--------|---------------|--------------|-------------------|
| Netflix | Completion, autoplay | A safe next watch | Recommend across media, surface the strange |
| Spotify | Continuous listening | One more similar song | Leap out of a sound into an idea |
| YouTube | Watch time | An endless escalating feed | Point you toward the right thing, not the loud one |
| TikTok | Reflex attention | Uncanny short-term holds | Build taste over time |
| Goodreads / Amazon | Sales signals | Bestsellers near your purchases | Read your mind instead of the market |
| An AI librarian | Understanding your library | A next pick that fits and a reason why | Nothing it cannot see, which is why it holds everything |

## When to let the algorithm drive, and when to wander

The engines are not the enemy. They are tools with a narrow, honest job, and sometimes that job is exactly what you want.

Let the algorithm drive when you want background and low stakes: a show to half-watch while you cook, a playlist for the drive, a quick pick when decision fatigue has won. In those moments, "good enough and immediate" is the right answer, and the feed delivers it well.

Wander when you want the thing you will still be thinking about next month. When you want to be changed a little. When a feeling from one book is still with you and you want to know where else it lives. That is the work the feed cannot do, because it requires seeing your whole taste at once and caring about the fit more than the next minute.

The trap is letting the first mode quietly become the only mode. Convenience compounds. Wander on purpose, or the algorithm wanders for you, and it always wanders toward the middle.

## The honest answer

Streaming recommendations feel worse because they are optimized for retention inside a single app, and that goal has almost nothing to do with your taste. Each engine can only see the catalog it sells, so none of them can follow a thread that crosses from music to film to a novel, which is exactly where the best recommendations live.

You get that range back by holding everything you love in one place and following the feeling rather than the feed. Sometimes that is you, paying attention to the connections in your own library. Sometimes it is an AI librarian built to reason across all of it instead of keeping you scrolling. Either way, the point is the same: recommend for the person, not the metric, and let your taste wander where the algorithms cannot go.

## Common questions

**Do recommendation engines actually get worse over time, or does it just feel that way?**
Both can be true. As a platform grows and leans harder on optimization, it converges on what works for the largest number of users, which flattens the experience for anyone with specific taste. Your taste is also deepening as you use it, so the gap between what you want and what the average-tuned feed offers genuinely widens.

**Can I train an algorithm to understand me better?**
Only up to the ceiling of what it is measured on. You can nudge Netflix or Spotify with ratings and skips, and they will refine within their silo, but you cannot make a music engine recommend a film or make any of them value your delight over your watch time. The limit is structural, not a matter of more data.

**Why does cross-media recommendation matter so much?**
Because taste does not respect format. The reason a film moves you is often the same reason a certain album or novel does, and following that reason across media is where discovery gets genuinely surprising. Single-platform engines are blind to it by design, so the whole richest category of recommendation is simply unavailable to them.

**Is an AI librarian just another algorithm?**
It is software making suggestions, so in the plainest sense yes. The difference is what it optimizes for and what it can see. Pointed at your own library with no retention metric to defend, it can recommend because something fits rather than because it fills a slot, and it can reason across every medium at once.

**How do I start wandering without a tool?**
Pick something you loved recently and name the feeling underneath it, not the genre. Then look for that feeling in a different medium: if a novel's loneliness stayed with you, find a film or an album that carries the same weight. Keep the thread, change the format. That single habit breaks the loop the feed keeps you in.

**Does wandering mean I should ignore recommendations entirely?**
No. Use the feed for what it is good at and wander for what it cannot do. The goal is not to reject the algorithm but to stop letting it be the only voice, so that convenience fills the small choices and your own attention drives the ones that matter.