How Many Tags Should You Use? A Practical Guide

Here’s the answer up front, because most articles on tagging refuse to give one: two to four tags per item, from a total vocabulary of roughly fifteen to thirty tags. If you’re consistently above either number, your tags are quietly turning into noise.

Now the reasoning, because the numbers only stick if you know why.

Why more tags make search worse

Tagging feels like insurance: one more tag is one more way to find the thing. But retrieval doesn’t work that way. Every tag you add beyond the essential few does two kinds of damage:

It dilutes the tag itself. A tag earns its keep by narrowing. If you stamp inspiration on 800 of your 1,000 items, searching it returns a pile — you’ve built a tag that’s just your library with extra steps. The value of a tag is inversely related to how freely you hand it out.

It erodes the vocabulary. Generous tagging is how you end up with ui, UIs, ui-design, and interface coexisting — four spellings of one idea, each holding a quarter of the results. Search for one, miss three quarters. This is the failure mode that kills tag systems, and it’s caused by volume, not carelessness: the more tags you mint, the more near-duplicates you mint.

The old tagging-practice essays (Karl Voit’s How to Use Tags is the best of them) converge on the same counterintuitive rule: use as few tags as possible. Not as few as you can suffer — as few as still answer the questions future-you will ask.

The per-item rule: 2–4

A good tag set for one item covers, at most, three dimensions:

What it is — the pattern or subject: pricing, empty-state, anatomy.
Why you saved it — the quality: playful, data-dense, warm-light.
Where it belongs — a source or project, when relevant: linear, client-x.

One tag from each dimension is usually enough; four is the ceiling. If you’re reaching for a fifth, you’re describing the item instead of indexing it — description is what the image itself is for.

The vocabulary rule: ~15–30, fixed

Per-item discipline follows from vocabulary discipline. Decide your tag set deliberately, write it down, and treat a new tag as a small event, not a keystroke. The working heuristics:

A tag must be reusable. If you can’t imagine ten future items wearing it, don’t mint it.
No overlaps. buttons or cta — pick one, kill the other.
One convention. Lowercase, single words, plural: illustrations, not Illustration or nice illustrations.
Omit the obvious. Tags shared by everything distinguish nothing.
Prune quarterly. Any tag with two items after three months was a bad mint; merge it into a neighbor.

Fifteen to thirty well-chosen tags will carry a library of thousands of images. That’s not a compromise — a vocabulary you can hold in your head is the feature, because you tag consistently only when you remember what your options are.

The escape valve: search does the rest

The anxiety behind over-tagging is “what if I need to find it by something I didn’t tag?” That’s what search is for. Tags carve the big, recurring seams in your library; full-text and visual search cover the one-off queries. You don’t need a green tag if your tool can find green things — which is exactly the division of labor Kelu is built around: a few intentional tags per item, search for everything else, the same library on iPhone, Mac, and the web.

For the fuller argument on why this beats folder trees, read tags vs folders; for the screenshot-specific version, see why tags beat folders for screenshots.