A common instinct when publishing is to include every term a reader might use, on the reasonable theory that more terms means more chances to be found. In practice this produces titles and tag lists that read as machine output, and both the systems and the readers respond badly to it.
The instinct is not wrong; the execution usually is. There are places where listing variants genuinely helps, and places where it costs more than it returns, and the two are easy to tell apart once the distinction is named.
Why stacking fails with systems
Search and recommendation systems have handled keyword padding for a very long time. Repeated near-synonyms are a recognised pattern, and pages carrying them are commonly demoted rather than promoted, so the technique achieves the opposite of its intention.
Platform tag systems apply their own limits, penalising long lists or ignoring everything after a certain number. Adding a twentieth tag frequently means the twentieth is discarded, and occasionally means the whole listing is treated as low quality, which affects the good tags too.
Why stacking fails with readers
Readers assess a listing in about a second, and a title made of stacked terms reads as automated. That impression transfers to the work itself, and readers who assume something was produced carelessly do not open it, however good it is.
It also fails to communicate. A title that lists eight terms tells a reader nothing about what makes this particular work distinct, which is the memorability problem described in the piece on titling. Being findable is worthless if nobody can tell what they found.
Where repetition genuinely helps
Genuine spelling and romanisation variants are a real case, since exact-match systems treat them as unrelated words. Listing them once, in a tag field or at the end of a description, is a reasonable accommodation rather than padding, as described in the piece on spelling variants.
The same applies to community terms that differ from plain descriptive language. Including both, once each, in the appropriate field serves two audiences without addressing either badly, which is the compromise recommended in the piece on reclaimed words.
Consistency beats coverage
One term applied consistently across an entire body of work builds an association that no amount of one-off coverage produces. Systems learn what a publisher is about from repeated signals, and a catalogue tagged differently every time provides no repeated signal at all.
This is the same argument as the tagging consistency in the piece on how tags route readers. Choosing a vocabulary and keeping it is worth more than trying to occupy every possible term once.
The fields have different jobs
A title is read by humans deciding whether to open something. A description is read by humans who have already decided to look. Tags are read almost entirely by machines. Treating all three as the same field is what produces titles that look like tag lists.
Write the title for a person, the description for a person who is interested, and the tags for a system. Each then does its own job well, and the work is both findable and legible, which is what everybody wanted from the padding in the first place.
Testing a title
Read it aloud to somebody and ask what they think the work is. If they cannot say, the title has failed regardless of how many terms it contains. If they can, and it was short, the terms you left out were probably not doing much.
This is a five-second test and almost nobody performs it. It catches the specific failure of a title that satisfies an imagined algorithm and communicates nothing to the reader standing in front of it.
Writing titles that work for both
Keep the title human, put variants in tags once, choose a consistent vocabulary, use each field for its own purpose, and test by reading aloud. That covers discovery and legibility together, which stacking never does.
Our own approach to naming and description is described in the editorial principles.