Slug Generator

How to Use This Tool

1. Paste a titleA headline, a product name, a page title — anything destined for a URL.
2. Set the shapeHyphen or underscore, lowercase or preserved case, an optional length cap, and optional removal of short words.
3. Copy, or do a batchCopy the single slug, or paste a list of titles to slug them all with collisions resolved.

How It Works

Most of the work is done by Unicode normalization rather than a translation table. NFD decomposition splits an accented character into a base letter plus a separate combining mark, so removing the combining-mark range leaves the plain letter behind — that single step handles é, ñ, ü, å and every other accented Latin character at once. What normalization cannot do is handle letters that are not decorated versions of anything: German ß, Nordic ø and æ, Polish ł and Turkish ı decompose to themselves, so those go through an explicit map applied first, since several expand to two characters. Everything remaining is lowercased, non-alphanumeric runs become the separator, and repeated separators are collapsed.

Worked Examples

What Makes a Good Slug

A slug is read by three audiences and they want slightly different things. Search engines use it as a weak relevance signal, so the main keyword belongs in it. People read it in a link before deciding whether to click, so it should be legible and roughly describe the page. And systems parse it, so it should avoid characters that need percent-encoding, which is why the safe set is lowercase letters, digits and a single separator.

Hyphens rather than underscores, for one specific reason: search engines have long treated a hyphen as a word separator and an underscore as a word joiner, so first_post can be read as a single token where first-post is two words. Underscores remain a perfectly reasonable choice for internal identifiers where no search engine is involved.

Length is a matter of judgment rather than a rule. Very long slugs get truncated in search results and are awkward to share, while very short ones lose the keyword that justified having a slug at all. Somewhere around 50 to 60 characters covers most titles without cutting anything meaningful — and when this tool truncates, it breaks at a separator rather than mid-word, because a slug ending in a half-word looks like a bug.

Accents, Normalization, and the Limits of Transliteration

Unicode normalization is the part of this that feels like a trick. A character such as é can be stored as one code point or as an e followed by a combining acute accent, and NFD normalization converts to the second form. Once decomposed, deleting the combining-mark range leaves an ordinary e — so a single normalization step and one deletion handles every accented Latin character there is, with no per-language table to maintain.

The letters that need an explicit map are the ones that are not accented at all. German ß is its own letter and conventionally romanizes to ss. Nordic ø and æ, Polish ł, Icelandic þ and ð, Turkish dotless ı: each is a distinct letter whose decomposition is itself. These are handled before normalization, because several expand into two characters and doing it afterwards would be too late.

Then there is the honest limit. Chinese, Japanese, Arabic, Hebrew, Korean, Cyrillic and Greek have no Latin decomposition, and romanizing them properly requires knowing both the language and which system you want — Pinyin or Wade-Giles, Hepburn or Kunrei, and so on for every script. Choosing one silently would be worse than not trying. So those characters are dropped and reported, and if the result is empty the tool says so instead of handing back an empty string that becomes a broken URL.

Collisions, and Why Slugs Should Not Change

Two different titles regularly produce the same slug — two posts called "Update", or two products differing only in a symbol the slug dropped. The convention every CMS settles on is a numeric suffix starting at 2, because the unsuffixed slug is conceptually the first one. The batch mode here does exactly that, so pasting a list of titles shows you the collisions before they reach your database rather than after.

The more consequential rule is that a slug should stop changing once a URL is published. Regenerating slugs from titles on every save is a tempting feature and a quietly destructive one: every edit to a headline breaks inbound links, discards whatever ranking the old URL accumulated, and invalidates anything anyone bookmarked. Generate the slug once at creation, store it, and let the two drift apart.

When a slug genuinely must change — a wrong product name, a rebrand — the old URL needs a permanent redirect to the new one rather than being left to return a 404. That preserves the links and the ranking. A slug that simply disappears takes both with it.

Where Slugs Fit in the Rest of the Work

Slugs usually appear during a migration or a bulk import, which is why the batch box exists. Content coming out of an old system arrives as a table of titles far more often than as anything structured: converting that CSV to JSON gives you records to slug and load, and the reverse trip back to CSV is how the mapping usually gets handed to whoever is configuring the redirects.

When the content itself is being migrated rather than just the titles, converting the pages to Markdown and slugging their headings tends to be the same afternoon’s work — the slug becomes the filename, so the two steps are really one. Headings arriving from several authors are also rarely consistent, and normalizing their capitalization before slugging avoids a set of files that differ only in case.

A slug is also the wrong tool for one job it gets used for. It is derived from the title, so it is guessable by design — which is fine for a public page and wrong for a private share link, where the whole point is that nobody can find it by trying. Those need a random token, not a slug. Where a bulk slug mapping has to be handed to whoever configures the redirects, it usually travels as JSON, and validating that file before sending it saves a round trip.

Two smaller notes. If a slug is going into a query string rather than a path, or the source title contained characters you chose to keep, percent-encoding it is the step that stops a link breaking. And when you are designing a URL structure before real content exists, placeholder titles give you something realistic to slug and see how the paths look at length. And when a record id travels in the URL beside the slug, it is frequently packed as Base64 rather than printed, which is worth decoding before assuming a path is malformed.