Verdah

Scores LinkedIn posts for AI-written text as you scroll. When it flags a post, it has been right every time so far.

NEW IN 1.4.1 Detection retuned against a 100-post labelled set: correct AI flags rose from 8 to 21, and false accusations fell from 1 in 50 to zero. The old Aggressive setting wrongly flagged 10 people in 50, so it is gone. Free tier is now 25 posts a day. Full audit at verdah.com/accuracy.html Half of LinkedIn reads like it came out of a machine. Verdah tells you which half, and it will not accuse a real person of being a bot. It scores posts as you scroll and puts a small badge on each one: AI, Human, or Mixed, with how confident it is. You keep scrolling. It just tells you what you're reading. WHAT IT DOES • Puts a verdict badge on every post: AI, Human, or Mixed • Shows how confident it is, and gets quieter when it isn't sure • Summarises how much of the feed you're looking at reads as machine-written • Lets you disagree with any call and mark the right one • You choose how strict it is, and every setting shows its own measured numbers Pro adds: scanning on profiles, company pages, groups and post search, unlimited scans, filtering a feed by verdict, an author check showing what share of someone's posts scored as AI, the exact phrases that drove each score, and right-click checking on any website. HOW ACCURATE IS IT Partially accurate, and we'd rather say so than round up. Measured against 100 posts labelled by hand, 50 written by a model and 50 by people: • Every post it flagged as AI really was AI. 21 out of 21. • Not one human writer was wrongly flagged. 0 out of 50. • It also let 28 machine-written posts through. That is the trade we chose. Verdah misses machine-written posts. What it does not do is tell you a real person is a bot. The full result, including the confusion matrix and everything it got wrong, is published at verdah.com/accuracy.html. Most detectors in this category quote a percentage and never show you the set they measured it on. This one shows you. No AI detector is reliable enough to be a verdict on a person. Treat every badge as a reason to read more closely, never as proof. WRITING IN A SECOND LANGUAGE IS NOT AI This is the part most detectors get wrong, and it matters. AI detectors have a documented bias against people writing in English as a second language. Missing articles, formal textbook phrasing, idioms translated a little too literally. These get read as machine-like when they are the opposite: evidence a real person wrote it. Verdah is told to score on structure and specificity, never on how fluent the English sounds. Across a set of ten human-written posts using Spanish, Hindi, German, French, Chinese, Arabic, Polish and Japanese language patterns, it flagged none of them as AI. LinkedIn is global. Getting this wrong means accusing people of faking their own writing, and that is not acceptable. YOU CHOOSE HOW STRICT IT IS A recruiter screening candidates wants zero false accusations. Someone clearing out a feed will trade one for a bit more detection. Three settings, each showing what it scored on the same 100 posts: • Cautious — 17 of 50 AI posts caught, zero people wrongly flagged • Balanced — 21 of 50 caught, zero people wrongly flagged. The default. • Aggressive — 22 of 50 caught, one person in 50 wrongly flagged Change it any time. Verdicts on screen update immediately. PRIVACY • It reads the text of posts on LinkedIn, and text you select yourself elsewhere. That's all. • It does not read your messages, your connections, or your account details. • It does not track you across the web or sell anything to anyone. • Your feed history is stored on your own device and never leaves it. • Free users can supply their own API key. Scoring then goes straight from your browser to the provider and never touches our servers. Full policy: https://verdah.com/privacy.html FREE AND PRO Free — your own home feed, 25 posts a day, full verdicts and confidence, correct any verdict, and the sensitivity setting. Pro — every profile, company page, group and post search, unlimited scans, the exact phrases behind each score, right-click checking on any website, filter a feed by verdict, author check on profiles, feed composition over time, and a shareable summary card. $19/month or $149/year. Teams of 2 or more get per-seat pricing at verdah.com. Verdah is an independent tool. It is not affiliated with, endorsed by, or connected to LinkedIn.