URL Slug Generator

How to use the URL slug generator

Paste a title and the slug appears immediately, lowercased, accent-stripped, punctuation-free, and joined with single hyphens. How to Make the Perfect Cafe au Lait at Home becomes how-to-make-the-perfect-cafe-au-lait-at-home, 44 characters long. Four options sit alongside it: hyphen or underscore as the separator, lowercase on or off, stop-word stripping on or off, and a maximum length. With stop words stripped that same title becomes how-make-perfect-cafe-au-lait-home, and with a 30-character cap it becomes how-to-make-the-perfect-cafe, truncated at a word boundary rather than mid-word.

What separates a real slug generator from replacing spaces with hyphens is what happens to letters outside plain ASCII. The first step is Unicode normalisation to NFD, which decomposes an accented letter into its base letter plus a separate combining mark. The accent can then be removed and the letter kept, so cafe with an acute accent becomes cafe rather than caf.

Skip that step and the punctuation strip deletes the accented character outright: a naive replace turns resume with two acutes into nave-rsum and Reykjavik into reykjavk, which is not a slug anybody would recognise as the article they clicked on.

That is why this tool returns grosse, ol, lodz, and thor where the mapless pipeline returns gro-e, l, odz, and or, and the reference table below shows the two pipelines side by side on the same titles.

Which letters can normalisation not save?

Normalisation alone is not enough, and this is the part almost every home-grown slug function gets wrong. Some letters are not accented forms of anything, they are their own letters in their own alphabets, so decomposition leaves them untouched and the punctuation strip then deletes them.

The German eszett, the Nordic o with stroke and the ash ligature, the Polish l with stroke, and the Icelandic thorn all behave this way. Without an explicit map, grosse written with an eszett slugs to gro-e, ol written with a stroked o slugs to just l, and Lodz slugs to odz. Those are silently wrong, and wrong in exactly the languages whose readers would notice.

The fix is a transliteration table applied before normalisation, mapping the eszett to ss, the stroked o to o, ash to ae, the stroked l to l, thorn to th, eth to d, and the rest of that set to their conventional Latin equivalents.

The formula, worked line by line

A slug is produced by a fixed pipeline, and the order of the steps is what makes it correct. Transliteration comes first because normalisation cannot help the letters it handles. Normalisation comes second so accents become separable. Only then is punctuation removed, because removing it earlier would take the accented letters with it.

The last two steps are what keeps the output clean. Splitting on runs of whitespace, hyphens, and underscores rather than replacing them one for one means no slug can ever contain a doubled separator or end with one, whatever punctuation soup went in. Truncation, when a maximum length is set, cuts back to the last separator so no word is left half-written.

1. transliterate   eszett to ss, stroked o to o, ash to ae, stroked l to l, thorn to th
2. normalise       NFD, then drop the combining marks in U+0300 to U+036F
3. strip           replace anything not a letter, digit, underscore or space with a space
4. split and join  split on runs of space, underscore and hyphen; join with the separator
5. truncate        if over the limit, cut to the last separator so no word is broken

The transliteration map is the part worth arguing for, because its absence is a real bug that a test in this repository exists to catch. NFD decomposes a letter only when the letter is defined as a base plus a mark.

The German eszett, the Nordic stroked o and ash, the Polish stroked l, the Icelandic thorn and eth, and the dotless Turkish i are none of them accented forms; they are independent letters, so decomposition returns them unchanged and the punctuation strip in step 3 then deletes them.

The result is grosse becoming gro-e, ol becoming l, Lodz becoming odz, and Thor becoming or. Mapping them explicitly to ss, o, ae, l, th, d, and i produces grosse, ol, lodz, and thor instead.

The pipeline is also idempotent by construction: feeding a finished slug back in returns the same string, because every step is already satisfied. That matters when a slug is regenerated on every save, which most content systems do.

A slug is a naming decision, not a ranking lever: keep it short, keep it readable, use hyphens rather than underscores because search engines have long treated hyphens as word separators and underscores as joiners, and once a URL is published leave it alone or serve a permanent redirect, because every link pointing at the old address is an asset you would be throwing away.

Questions people ask