Content length checker

The useful additions are the ones that could only come from you. What you tried that did not work and why. The number you measured rather than the range everyone quotes. The objection a knowledgeable reader would raise, answered honestly. The case where the advice does not apply. A screenshot of the actual thing, with a sentence explaining what to look at.

The additions that make a page worse are equally predictable. A three-paragraph introduction before the subject appears. A definition of a term the reader clearly already knows. The same point restated under a new subheading. A conclusion summarising what was just read. Each of these adds to the count and subtracts from the page, and enough of them turns a short useful page into a long useless one.

There is a simple test before you expand anything. Write down what someone would be able to do after reading the page that they could not do before. If that sentence is easy to write, the page needs more of what it already is. If it is hard, more words will not rescue it and merging it into something larger is the better move.

A site average is the most flattering number available and the least informative. One 4,000-word guide lifts the mean enough to hide thirty stubs, and the mean is not what anyone assessing the site experiences. They click three pages, and what matters is which three.

The distribution is the real signal, in both directions. A healthy content site has a wide spread: short notes, medium articles, a few long pieces, because subjects differ in what they need. A site where every page lands within a few percent of the same length is not varied writing, and the uniformity heuristic reads that spread as a fingerprint of templated production rather than as consistency.

So the per-page report is what you work from. Sort by count, look at the bottom of the list first, and decide page by page whether each one should be deepened, merged or deleted. That list is also usually the same set of URLs showing up in the near-duplicate pairs, which is a good sign you are looking at one problem rather than two.

A worked example, because the difference between depth and padding is easier to show than to define. The page is How to descale a kettle, currently 240 words: an introduction restating the title, four numbered steps that amount to fill with vinegar, wait, empty, rinse, and a closing sentence. It answers the question in the technical sense and fails everyone with a follow-up question, which is most readers.

The additions that earn their words are all specifics. The actual ratio, one part white vinegar to one part water. The soak time you tested, 45 minutes for light scale and overnight for a kettle that rattles, rather than a while. Citric acid as the alternative, a tablespoon in a full kettle, and when it is the better choice because the smell rinses out faster. The failure case: if flakes keep appearing after two treatments, the scale is under the element plate and no soak reaches it. A photograph of the scale line with one sentence saying what to look at. How often to repeat it in a hard-water area. That page is around 700 words, and every added sentence is something a reader might have come back to search for.

The padded version of the same page also reaches 700 words: a history of the electric kettle, a section titled Why descaling matters explaining that scale is bad, a definition of limescale for a reader who typed the word themselves. Identical count, and the page got worse, because the answer is now buried in the middle of material nobody wanted. Both versions clear this check. Only one of them deserves to, which is why the reviewer pass judges pages on what they say rather than how long they are.

After a rewrite like that, close the loop. Confirm the page cleared the thin-page flag, and update the title and description to promise what the page now actually delivers, because the old 55-character title was written for the stub and undersells the rewrite. The pages worth this treatment are the ones with search impressions and nothing else; a stub nobody is searching for is a merge candidate, not a rewrite candidate.