The Finder starts with one anime you name. It combines catalogue similarities and community recommendations to propose other titles, then lets you narrow the results by genre, length, and what you have marked seen. It is a starting point for a choice, not a claim to know your entire taste.
The description here reflects the implementation reviewed on September 14, 2026. The underlying catalogue can change independently of this article. When the algorithm changes, this explanation should change with it.
What the numbers mean
| Signal | Source | How to use it |
|---|---|---|
| Catalogue score | AniList community rating, displayed on a ten-point scale | Broad reception on that platform |
| Match percentage | Fantera’s transformation of its similarity score | Relative fit to the selected starting title |
| Recommendation explanation | Shared attributes, curated copy, or an offline model-written rationale | A reason to inspect the suggestion, not verified proof of quality |
| Seen mark | Your browser’s saved list | Exclude something you already watched when hiding watched titles |
The ranking has two ingredients
The offline calculation combines 60 percent tag-and-genre similarity with 40 percent a transformed community-recommendation signal. Tags carry AniList relevance weights. The community component uses a logarithmic transformation so large raw counts do not grow linearly without restraint.
This is a description of the calculation, not an empirical claim that those weights are optimal. We have not published a controlled evaluation showing that this weighting outperforms every alternative. A recommendation can be sensible for one aspect of a show and poor for another.
The calculation also groups related franchise entries and limits repetition from the same franchise. This reduces the tendency to recommend a sequel as though it were an independent discovery. It is only as reliable as the underlying relationship data and grouping logic.
A strong match is a reason to look closer, not a promise that you will love it.
Why a high match can still feel wrong
Two shows can share fantasy, travel, and supernatural tags while feeling very different. One may be warm and episodic; another may be grim and violent. Similarity captures overlap, not everything about the emotional experience.
The displayed percentage is not calibrated as a probability that you will enjoy the show. A 90 percent match should not be read as a promise of nine successful recommendations out of ten. Nor does it mean the show is rated nine out of ten by viewers. The catalogue score is displayed separately.
Written explanations can help expose the supposed connection, but they can also simplify or mischaracterise it. Treat them as a prompt to compare the premise and tone. Our editorial comparison guides spend more space on those differences than a result card can.
What your controls actually do
- Choose an anchorSelect the exact title from autocomplete.
- Narrow the fieldUse genre, length, and seen controls.
- Read the reasonCheck whether the shared quality matters to you.
Selecting an autocomplete suggestion tells the Finder exactly which title you mean. A free-text query instead asks it to resolve the likely title. Genre and length filters narrow the candidate list; they do not create a new understanding of your viewing history.
A “short” result is based on the episode count of that catalogue entry. It does not establish complete-story length, and a film can be represented as a single entry. Use the viewing-commitment guide when that distinction matters.
Seen marks are stored in your browser. The current server cookie includes the most recent 500 IDs for exclusions. The Seen page lets you review the full local list and reverse a mark. This is not cross-device account synchronisation.
Where editorial judgement enters
The articles make explicit recommendations using a stated viewing need, sourced facts, and comparisons of fit and mismatch. They are not generated by copying the Finder’s top results into a numbered list. The two surfaces complement each other: the guide helps define your criteria, and the Finder helps continue exploring.
Start with a title you liked, read why a candidate was selected, and check one reason it might not work. If the recommendation misses, try an anchor that better represents the particular mood or relationship you want. One favourite cannot encode every preference you have.
Sources and review
Factual references for this guide are listed below. Recommendation comparisons are Fantera’s editorial interpretation. See how we research and review articles.
- Fantera: similarity calculation source
- Fantera: catalogue queries and ranking
- AniList API documentation
Published . Reviewed .