Slop Radar: AI writing labels

See which posts in your feed are AI slop before you read them. Slop Radar is an AI writing detector for your LinkedIn feed. As each post scrolls into view it gets a small cream tag: Human, Unclear, or Reads like AI. Hover or tap the tag for a popover with one meter (human on the left, slop on the right), both percentages, and the signals behind the call. The signals Pointing to slop: a generic hook, one-sentence-per-line "broetry" or emoji bullets, stock phrasing like "let that sink in", engagement bait like "Agree?". Pointing to human: concrete first-hand detail a template could not produce, such as named people or exact technical detail. A post is only tagged Human or Reads like AI when one side holds at least 60% of the probability. Anything in between is Unclear, and the popover says so. Posts are rated once they are mostly on screen, one request each, cached so scrolling back is free. The tags follow the site's dark theme, and a post is rated again when "...more" reveals more text. It judges style, not authorship No tool can prove a post was written by AI. A person writing in the platform's house style will be tagged too. Read the tags as a hint, never as a verdict on anyone. What gets sent There is no Slop Radar server and nothing reports back to me. What leaves your browser is the visible text of feed posts, sent with your own API key to the provider you picked: TypeSafe, or TypeSafe by way of Vercel AI Gateway. The author's name, profile, comments, messages and anything about your own account are never sent. Ratings are cached in this browser. It runs on www.linkedin.com and nowhere else. Full policy: github.com/dgr8akki/slop-radar/blob/main/PRIVACY.md You need Chrome 140 or newer, plus an API key of your own from TypeSafe (console.typesafe.ai/keys) or Vercel AI Gateway (vercel.com/docs/ai-gateway). Settings open when you install. Pick TypeSafe or Vercel, paste, and Connect tries the key once before keeping it, and after that you only see it masked. An evening on the feed costs under a cent, on your account. Where it fails The site obfuscates its markup, so Slop Radar relies on two stable hooks. If they change, tags stop appearing until an update. The questions are written in English; other languages are rated less reliably. Something wrong or missing? github.com/dgr8akki/slop-radar/issues The code is MIT-licensed: github.com/dgr8akki/slop-radar Not affiliated with or endorsed by LinkedIn.