Simon Willison turned AI clichés into something editors can mark

Simon Willison released LLM cliché highlighter, a tool that highlights 10 common patterns associated with LLM prose. It runs in the browser, shows match counts, lists hits, supports quick navigation and, according to the description, uses localStorage so text does not need to leave the browser.

LLM style becomes highlighted evidence, not a vibe

Willison's motive is practical: he was frustrated by an article crammed with phrases like no fluff, no filler, no jargon and had Fable 5 quickly build a highlighter. Small tool, useful signal.

AI writing often fails not because it is factually wrong, but because it sounds molded from the same template. The highlighter turns that into an editable problem instead of an aesthetic argument.

Editors get a linter for brand voice

For teams that publish a lot, this is more useful than another generic workshop about writing with AI. An editor can point to a sentence, a repeated pattern and the reason the copy feels cheap.

The same logic applies to product copy, documentation and support material. If every surface starts using the same LLM phrases, readers quickly notice where attention was saved.

A cliché list will not replace judgment

The limit is obvious. A phrase detector cannot replace an editor and may flag sentences that work perfectly well in context. Style is not regex, and good writing sometimes uses patterns that bad writing overuses.

So the tool is best treated as a linter, not a judge. It can show the smell, but it cannot decide whether a paragraph is true.

The useful next step is checking whole publishing systems

The interesting move is whether checks like this enter publishing pipelines. The real problem is not one phrase in one article, but thousands of similar sentences across docs, newsletters and onboarding emails.

If teams track clichés as seriously as typos, AI-assisted writing will improve faster than with another prompt about tone of voice.

Lilith's verdict

This is a red pencil placed next to the text generator. It does not ban AI from writing, but it reminds the author that readers can smell plastic in the first paragraph.