What is the difference between genre and mood in recommendations?

Genre is a production taxonomy — it describes how a work was made, marketed and shelved. Mood is an experience taxonomy, describing what encountering the work is actually like: how much attention it demands, how heavy it is, how long it takes and what kind of company it is. Most recommendation systems use the first while human choices depend on the second.

Why do most recommendation engines still use genre?

Because it is cheap. Genre already exists in the metadata for every title, and assigning it requires nobody to have watched anything. Experience attributes are judgements that vary by viewer, so platforms infer them indirectly from behaviour — which reintroduces popularity bias and feedback loops.

What four questions work across all media?

How much attention does it demand, how heavy is it willing to be, how long is the commitment, and what kind of company is it. The units change per medium — pace and density for books, tempo and energy for music, runtime and tone for film — but the structure holds.

Mood, Not Genre: A Better Mental Model

This is the general version of an argument that shows up all over this site. The medium-specific cases are elsewhere — for music and for books. This is the underlying idea, which turns out to be the same in every medium.

Two kinds of taxonomy

A production taxonomy classifies a work by how it was made, marketed and distributed. Genre is one of these. So is "A24 film," "prestige drama," "shōnen," "literary fiction," "indie." These labels are real and carry information, but the information is about the object's origin — its budget, its intended audience, its shelf.

An experience taxonomy classifies a work by what it is like to encounter it. How much attention it demands. How it leaves you. How long it takes. Whether it is company or confrontation.

Nearly every recommendation system uses the first and nearly every human choice depends on the second. That mismatch is the whole problem. When a recommendation feels technically correct but wrong — yes, it's a thriller, and no, I didn't want this — you're feeling the gap between the two taxonomies.

The axes that actually generalise

Across every medium, four questions do most of the work. The vocabulary changes; the structure doesn't.

Demand. How much of you does it want? Subtitles, unreliable narrators, dense prose, non-linear structure all raise it. This has to match your available attention or nothing else matters.
Weight. How dark is it willing to go, and where does it leave you? The most-skipped question and the most consequential one.
Duration. Ninety minutes, six hours, six weeks. Not a quality judgement — a commitment you're either in a position to make or not.
Company. What kind of presence it is. Warm, cold, funny, clinical, furious. You're choosing who to spend the time with.

Translated per medium: pace and density for books, tempo and energy for music, episode count and whether it's finished for series, runtime and tone for film. Same four axes, different units.

Why the good version of this is hard to build

If experience axes are so much better, why does everything still use genre? Because genre is cheap. It's already in the metadata for every title ever released. Nobody has to watch anything to assign it.

Experience attributes are expensive. "How much attention does this demand" is a judgement, not a field, and it varies by viewer. Large platforms approach it indirectly, inferring it from behaviour — which works at scale but reintroduces all the biases of behavioural data.

There's a genuine tension worth admitting: mood-first requires the user to do a little work. You have to answer questions about yourself. Genre-first requires nothing. For a lot of people, most of the time, nothing is the right price.

The one thing genre can never know

Even a perfect production taxonomy has a hole in it: your evening isn't a property of the work. The same film is a different film on a Friday you're celebrating and a Tuesday you're surviving. No amount of metadata about the object can capture a variable that lives entirely in the viewer, and no amount of watch history can either — because your history records what you watched, not what state you were in.

That's the case for simply being asked. Eight questions is a small price for the one input nothing else in the system has.

Using it without a tool

You don't need a quiz for this. Before you open anything, say out loud: how much attention do I have, how heavy can it be, how long have I got, and what kind of company do I want. Then go looking. You'll be right more often, and — more usefully — you'll know why you were wrong when you're wrong.

If you'd rather have it asked for you, every guide and quiz here is built on those four axes.