Where Suggest Me Something Gets Its Data

Every recommendation on this site is a real entry in a real public catalog. Nothing is generated. Nothing is guessed. If a title comes back in your shortlist, it exists, it has a real release year, and somebody other than us has rated it.

That sounds like a low bar. In 2026 it isn't. So here is the whole tour — every source, what it's responsible for, and where its limits are.

The six catalogs

Movies and TV series come from TMDB, a community-maintained database that supplies posters, synopses, genres, runtimes and — crucially — per-country streaming availability. IMDb-style ratings are fetched separately from OMDb, because TMDB's own user scores skew high and cluster tightly, which makes them poor at separating a good film from a great one.

Anime comes from Jikan, the open API that reads MyAnimeList. Anime needs its own source because the metadata that matters — episode count, whether a series is finished, how much of it is filler, which season a cour belongs to — simply isn't in a general film database.

Songs come from Apple's public iTunes Search API, which returns tracks, artists, album art and a 30-second preview. Your answers about mood, tempo and vibe become search terms, and the results are then filtered against what you actually asked for. Each pick links out to Apple Music, plus a search link for Spotify so you can open it wherever you listen.

This one deserves a note, because the site used to say Spotify. The song quiz was originally built on Spotify's recommendations endpoint and its audio-feature parameters — valence, energy, tempo. Spotify closed those endpoints to new applications in late 2024, so the feature was rebuilt on Apple's catalog. The copy across the site lagged behind the code for a while; this page is now accurate, and so is everything else.

Books come from Google Books for titles, authors, descriptions and page counts, with covers and read-online links from Open Library. Page count matters more than it sounds — it's the single most reliable predictor of whether someone in a reading slump will actually finish the thing.

How a mood becomes a query

The quiz is a translation layer, nothing more. Each answer maps to a parameter the catalog already accepts. "I want something light" becomes a genre inclusion set and a keyword exclusion set. "Under ninety minutes" becomes a runtime ceiling. "Something I haven't heard of" becomes a popularity band — not the top 200, not the bottom of the tail.

Nothing about that is clever, and that's the point. The catalog already knows which films run under ninety minutes. It does not need to be asked in natural language, and it does not need a model to interpret the request. It needs the request expressed in its own terms.

Why not just use a language model?

Because language models make up titles. Not occasionally — structurally. Ask one for "short cosy anime with no filler" and you will get a list that is roughly eighty percent real, ten percent real-but-wrong-description, and ten percent confidently invented. The invented ones are the problem, because they look exactly like the real ones.

There's a subtler failure too. A model trained on the internet recommends what the internet talks about most. Ask ten different models for film recommendations and you will see the same forty titles. That's not taste — it's a popularity average wearing a conversational voice.

A deterministic query has an unglamorous advantage: it is reproducible. Answer the quiz the same way tomorrow and you get a coherent set again. If a pick is wrong, it's wrong for a reason we can find and fix, not because a sampler rolled differently.

What we don't do with any of it

We don't store your answers. There's no account, so there's nothing to attach them to. The catalog requests are made from our servers rather than your browser, which means those services never see your IP address. Each session re-queries the source, so picks stay current with the catalogs themselves rather than with a stale copy we took months ago.

Where the limits are

Being honest about the sources means being honest about their gaps. TMDB's streaming availability is good but not perfect, and it lags for smaller regions. OMDb's ratings are IMDb-derived, which carries IMDb's well-known skew toward English-language and male-voted titles. MyAnimeList scores are generous at the top end. Google Books metadata for older or self-published titles can be thin or wrong.

None of that is fixable from our end. What we can do is not pretend otherwise — and not paper over a gap by inventing something to fill it.