How to Build a Tag Taxonomy for Your Library

A tag taxonomy isn’t a long list of tags — it’s a small set of axes you tag along. Get the axes right and 20 tags will index thousands of images; get them wrong and 200 tags will still leave you searching blind. This is about designing the axes. If you want the week-by-week process of applying them, that’s the tagging system for screenshots; here we design the structure it runs on.

Facets, not a flat pile

The mistake most people make is treating tags as one undifferentiated bag, so pricing, calm, and linear all sit at the same level with no relationship. A taxonomy gives tags roles. Borrowed from faceted classification, a visual library usually needs just three or four facets:

  • Type — what the thing is: pricing, onboarding, logo, poster, packaging, interior. The noun you’d point at.
  • Quality — why you kept it: minimal, playful, data-dense, warm-light, retro. The adjective future-you will search by.
  • Source / project — where it belongs, only when you’d filter by it: linear, client-x, instagram.
  • Medium (optional) — the craft dimension: type, color, layout, motion, illustration. Useful for designers who study one axis at a time.

Every item gets roughly one tag per facet — which is exactly how you land at the 2–4 tags per item that keeps search sharp.

Designing your axes (30 minutes, once)

Don’t invent the taxonomy in the abstract — derive it from what you already save. Pull up your last 50 saved images and do this:

  1. For each image, finish two sentences: “This is a ___” (Type) and “I kept it because it’s ___” (Quality). Write the words down as you go.
  2. Tally the words. The ones that recur are your starter vocabulary; the one-offs are noise — leave them to search.
  3. Group into facets. Sort the survivors under Type / Quality / Source / Medium. If a word won’t sit in any facet, it probably isn’t a tag.
  4. Cap each facet. Aim for 5–8 tags per axis, ~20–30 total. A facet with 30 values isn’t a facet; it’s a folder tree wearing a costume.

The output is one short list, grouped by axis. That’s the taxonomy.

The rules that keep it consistent

A taxonomy dies from drift, not from bad design. Four conventions prevent it:

  • One spelling, decided up front — lowercase, single word, plural: illustrations, never also Illustration.
  • No synonyms across the set — cta or buttons, pick one and kill the other, so search never splits.
  • Tags must narrow — if a tag would land on more than ~40% of the library, it distinguishes nothing; split it or drop it.
  • New tags are events — adding to the taxonomy is a quarterly decision, not an in-the-moment keystroke. This is the single habit that keeps it small.

Let it evolve — deliberately

A taxonomy is living, but it should change on a schedule, not by accident. Once a quarter: merge synonyms that crept in, split any tag that’s grown to hold half your library along the axis you actually search, and retire orphans with two items after three months. Ten minutes, four times a year.

Where Kelu fits

Kelu is designed for a faceted, few-tags-per-item approach: real tags (not folders), as many per item as the axes call for, and search across iPhone, Mac, and the web for everything the taxonomy doesn’t name. First 2,500 items free — enough to prove your axes hold at scale.

Once the axes are set, applying them is the easy part — and for design work specifically, see how to tag design inspiration for a worked example of these facets in use.