Slop Meter

Marks each paragraph human-ish, machine-ish or mixed, with the odds. Scoring runs on your device.

Slop Meter marks each paragraph on the page as human-ish, machine-ish, mixed or can't tell, and shows the odds behind the mark. How it works - A small model (56 KB) runs on your device. Text is scored locally and nothing you read is uploaded. - Most paragraphs get "can't tell". It only speaks when it is sure: it is right on 97% of its calls and flags about 1 in 1,000 paragraphs written by people. - Click a mark to see which writing habits drove it. Each one links to a rulebook of 38 tells, with an example and a fix. - Select any text and choose "Score with Slop Meter", or paste text into the popup. - Per-site switches. It stays off on mail, documents and banking. Sharper reading (optional) Turn it on and a small open language model (SmolLM2, about 120 MB, downloaded once) reads each paragraph on your device using WebGPU. It gives a verdict about twice as often. Opt-in only - Rewrite sends the one paragraph you choose to slop-meter.com and returns a plainer version. - Shared corrections send numbers about a paragraph and your label, never the text. Open source under AGPL-3.0: github.com/zaidmukaddam/slop-meter