Unvail (Beta) — AI detector & fact-checker

Unvail tells you how likely it is that something you're looking at was generated by AI - and, on articles and videos, whether the claims it makes hold up. It works on the pages you already use. Nothing is sent anywhere until you click. WHAT IT CHECKS Images - hover any picture and a small badge appears in the corner. Click it and you get a likelihood with the reasoning behind it. Video - on YouTube, a button sits in the player controls. It samples frames as the video plays and tells you which stretch looks generated. Hover the result and it says exactly which part was actually checked, because that is rarely the whole thing. Music and voice - analyse a thirty-second sample of what's playing. Text - select any passage for a check, or use the right-click menu. Text is the hardest kind to judge and we say so in the result. Claims - point it at a news article or a video and it pulls out the statements being made, searches for reporting on each one, and shows you what came back: which outlets agree, which disagree, and which claims nobody independent has covered. You get a rating out of four and every claim with its sources, so you can read them yourself. WHAT IT WILL NOT DO We return a likelihood, never a verdict. You will never see 0% or 100% - certainty isn't something a detector can offer. AI detection is imperfect in both directions. Accuracy drops on edited, cropped, compressed or shortened content, and on output from generators newer than the engines checking them. A result is one signal. It is not evidence of how something was made, and it should never be the sole basis for a decision about a person. A claim check reports what our search found. It is not a ruling on what is true, and a claim we could not confirm is not a claim we found to be false - often it just means the reporting sits behind a paywall. PRIVACY It does not watch you browse. No page history, no activity log, nothing transmitted until you click to check something. Media checking is blocked outright on messaging and email services, because the other person in that conversation never agreed to it. That block is enforced inside the extension and fails closed: an address it cannot identify is treated as private. A free account is required so we can keep usage fair while the beta is small.