Lorem Ipsum Generator

It gets the texture of prose wrong

Filler is a shuffled word pool, and a shuffled pool has a shape that real writing does not. Two differences matter for layout work.

The first is word length. The pool is genuine Latin, and Latin is made of long words with long inflections on the ends; only a small fraction of it is three letters or shorter, and the longest runs to fourteen characters. English prose is nothing like that — the commonest words in any paragraph are two, three and four letters, and every sentence is built around them. Fill a column with one and with the other and the lines break in different places: fewer words per line, a different rhythm down the right edge, and a block that measures the same in words and different in height.

The second is repetition. Real text repeats: a handful of function words carry most of the sentences, a subject noun comes back three or four times in a paragraph, and the repetition is what makes a block of type read as a block of type. A shuffled pool produces almost no repetition — most words appear once, and a word appearing twice in a row is a coincidence rather than a pattern. The result is a page that looks slightly scattered in a way nobody can name:

Real copy, first 20 words of a paragraph
        the  the  a  of  and  to  is  it  that  in
        the  you  that  a  and  of  the  to  is  the

      Filler, first 20 words of a generated paragraph
        27 distinct words among the 20 shown, mostly
        once each, with the commonest appearing twice
        by chance rather than because a sentence
        needed it there.

This is why filler is good at exactly one thing and poor at everything adjacent to it. Judging a type scale, a line length or the space between a heading and a paragraph needs shapes, not words, and a shuffled pool gives shapes honestly. Judging whether a block of copy feels dense, or whether two columns balance, is already a judgement about the words.

Where the test is about meaning

Anything that reads the text rather than drawing it needs real words. A screen reader announces every one of them, which turns a page a sighted reviewer skims in five seconds into a minute of nonsense for anyone listening, and the evaluation they were asked to do cannot be done. A heading level is announced as a heading and not as a phrase, so a page full of filler headings produces a document outline with eleven entries that say nothing — and the question of whether the outline is useful is unanswerable.

Translation is the other one, and it is the case filler fails hardest. The longest word in the target language is not the longest word in the source, and neither is the longest compound noun or the longest verb form. German compounds and Finnish case endings routinely produce words that overflow a button sized for English, and a layout tested in a dead language has been tested against no language at all. If a page is going to be translated, the layout has to be judged with the longest expected translation in it, which means getting that translation before the layout is signed off rather than after.

Search, sort and filter belong in the same group. Filler has no clusters, no duplicates worth collapsing and no meaningful order, so a list of it sorts into an order nobody has any reason to check. Real data is lumpy — a column where half the rows share one value, a name that appears eleven times, a field where the empty string turns up more often than anything else. Those shapes break interfaces, and a shuffled pool will never show you one.

The failure that ships

Every tool of this kind has one dangerous property: it works well enough to move on. A page built from filler lays out correctly, passes a link check, passes an accessibility scan, and can be deployed. Nothing in the pipeline objects, because there is nothing wrong with it — the text is real text, the markup is real markup, and the page is about nothing. What makes it a recurring problem rather than an occasional one is that the placeholder arrives early, before anybody has written the copy, and the last person to touch the page is never the person who put it there.

The discipline that prevents it is small. A section that has not been written yet is better left out than filled, because a shorter page is honest and a page of Latin is a promise that something is coming. Where a placeholder has to exist for a template or a content slot, it is worth making the placeholder say what it is — a comment in the markup, a distinctive string that a grep will find before a release — so the check is mechanical rather than a matter of somebody remembering. Where a layout genuinely needs text to be tested against, use the real copy, or the longest plausible version of it, and the Word Counter will tell you how far the sample sits from the length the real thing will be.