How it works
Unvail (say it like “unveil”) lives in your browser and stays out of the way until you ask it something. Here is exactly what that looks like, for each kind of media — and for checking the claims a piece is making.
Checking an image
Hover any image and a badge appears in its corner. One click, and it settles on a verdict with a confidence score.
Checking an image
Hover any image on a page. A small circular badge appears in the corner — it does nothing until you click it.
Click the badge. It spins while we check, then settles on a colour: green for likely human, indigo for likely AI.
The badge shows a percentage. Hover it for the detail — what was checked and how confident we are.
Checking a video
On YouTube, look for the Unvail button in the player controls, next to like and share.
Click it. If the video is paused it starts playing, because frames can only be read while it runs.
We sample frames as it plays — up to twelve — and update as we go.
When it finishes you get a summary card and a tick on the badge. Hover the tick: it tells you which stretch of the video was actually checked, because that is rarely the whole thing.
Checking music or a voice
Open the Unvail panel on a video or track and choose Voice or Music.
It records a short sample — thirty seconds — and analyses that.
The result carries a tick when it is done, and hovering says exactly which seconds it covered.
Checking the claims in an article, a passage or a video
On a news article, click the Fact-check button in the bottom corner, or right-click the page and choose “Unvail: fact-check this page”. It reads the page as you see it, so an article you can read behind a paywall can be checked too.
To check just one paragraph, highlight it, right-click and choose “Unvail: fact-check this passage”. Only what you selected is checked, so the answer is about that passage alone.
On YouTube, open the menu from the Unvail badge on the player and choose Fact check, then the whole video or just a part of it. We read the transcript when there is one, and watch the video when there is not.
It pulls out the statements being made and searches for reporting on each one, opening the pages it finds to look for the sentence itself.
The card opens with a rating out of four for how well the piece held up, then the headline, then each claim with its sources and what each one said. On a video every claim carries the moment it was said: click the time to jump there.
Read the sources. A claim we could not confirm is not a claim we found to be false — it often means the reporting is behind a paywall, or that only the original outlet has carried it. The card says which.
No extension, or on a phone? Paste a YouTube link or an article link into the Fact check tab of the web app. For a video, choose the whole thing or just the minutes you care about; an article that will not open from there (a paywall or a sign-in) can be pasted as text instead.
Text works a little differently
There is no badge to hover. Select a passage of 50 words or more and two small buttons appear beside it: “Unvail it” checks it now, and “Add to queue” adds it to a batch so several passages are scanned together. The same two actions are on the right-click menu under “Unvail Text AI”, and you can always paste text straight into the web app. Text is the hardest kind to judge, so treat those results with more caution than the rest.
Common questions
Does it watch everything I browse?
No. Nothing is captured or sent anywhere until you click to scan something. We also block scanning entirely on private conversations — WhatsApp Web, Gmail, direct messages — because the other person in that conversation never agreed to it.
Why does it say "likely" instead of yes or no?
Because a yes or no would be a claim we cannot support. Detection is probabilistic. We show a likelihood and cap the display at 99%, so we are never asserting certainty we do not have.
Do I need an account?
Yes, to run a scan. It is free, and it is how we keep usage fair while the beta is small.
Does a scan cover the whole video?
No, and this matters. We sample a stretch of frames, not the entire runtime. The finished-scan tooltip tells you exactly which seconds were checked — you can scan a different part separately.
Does a claim check cover the whole video?
When the video has a transcript, we read all of it unless it is very long; without one we watch a stretch of it. The card always says exactly what was read or watched, and you can choose a part of the video yourself.