Writing guide · 6 min read
The engine matches your prose against 150 worn phrases and reports the rate per thousand words. Here is what that catches, what it misses, and when a cliché is the right call.
By Alyssa Glasco, Founder · Published
Cliché Density matches your text against a list of 150 phrases and reports how many it found per thousand words. That is the whole mechanism. Knowing its size is the difference between using this number well and trusting it too far.
These are ceilings, not targets. There is no floor, because no writing needs a minimum number of stock phrases. Poetry, screenwriting, and technical writing never see this reading at all.
Three marketing phrases get a single free use: world-class, state-of-the-art, and cutting-edge. One occurrence is ignored entirely. Two or more and every instance counts. One may be the ordinary word for the thing. A habit is a habit.
The score is scaled per thousand words, which makes it spiky on anything short. One cliché in a hundred-word passage reads as a density of 10, well past every ceiling on the list. The phrase is not worse for appearing in a short piece. Check the raw count before reacting to the rate.
It is exact string matching against a finite list, and both halves of that limit matter.
A phrase you invented can be doing tired work while scoring zero. Fresh words in a familiar shape are the more common failure in practice, and no lexicon catches them. Meanwhile a cliché rewritten just enough to dodge the match escapes entirely, which rewards the wrong edit: the goal is a truer phrase, not a phrase that evades a pattern.
Two cases the engine cannot distinguish and you should.
The first is dialogue. People speak in worn phrases. A character who says at the end of the day is telling you something about themselves that a fresher line would conceal. Stock language in a character’s mouth is characterization, and the metric counts it identically to the same phrase in narration.
The second is knowing use. A familiar phrase placed deliberately, turned, or undercut is doing work. The engine sees the string and nothing else.
Treat a score over your ceiling as a prompt to reread, not a defect list. Open the piece in the editor, run Check with the Meridian, and look at which phrases it actually found. Decide one at a time whether each is the true thing or the near thing. That decision is not one a word list can make, and the number is only useful as the thing that sends you back to look.