The video keeps talking.Claimek keeps checking.
Captions move faster than anyone can check them. An AI model reads along, picks out the claims worth checking, and shows you what fact-checking groups have already said, right on the player.
Misinformation doesn't announce itself.It sounds like everything else.
A false claim comes with the same confident voice, the same good editing, and the same familiar face as a true one. By the time you stop to wonder about it, the video has already moved on.
It reads every line, so you don't have to
You can't open a new tab for every claim you hear. An AI model reads along with the captions and picks out the handful of sentences that are actually checkable, while the words are still on screen. It does not get bored on hour three.
It stays quiet when things check out
Most claims are fine, and Claimek says nothing about them. It only speaks up when something is rated misleading or worse. A tool that warns you all day gets switched off. One that warns you rarely gets believed.
Built for how Filipinos actually watch
Most tools break on code-switching. The model behind Claimek reads English, Filipino, and Taglish in the same sentence, because that is how our videos actually sound. Health myths, viral "alam mo ba" facts, political lines, all read in the language they were said in.
Three steps,all while the video plays
It reads the captions you already see
Claimek reads the caption text on the player, the same words already on your screen. No microphone, no recording, no camera. If a video has no captions, there is nothing for Claimek to read.
It picks out checkable claims
Opinions, jokes, and questions are left alone. Only statements of fact that can be checked, like "X causes Y" or "the law says Z", are sent to our server.
It shows its work
Each flag shows how serious the problem is, a short plain-English summary, and how sure the check is. If a fact-checking group has already written about the claim, you get a View Source link so you can read their article yourself.
Five levels.You only ever see three.
Every claim we check gets a level from 1 to 5. Levels 1 and 2 are dropped on our server before they ever reach your browser, so Claimek never interrupts you just to say something was fine.
Nothing to raise. You never see this one.
Nothing solid either way. Also never shown.
True in part, but said in a way that misleads.
Mostly wrong, with only a small bit of truth in it.
Goes against what the evidence shows.
What the flag actually says depends on where the answer came from, and Claimek always tells you which of the two it was.
You see their rating, in their words
- Claimek shows the rating the group published, names the group, and links to their article.
- You can read the original and decide for yourself.
- Claimek is passing on their work, not adding its own opinion.
You see an automated read, worded as doubt
- Claimek says the claim looks shaky and why. It does not declare the claim false.
- Every flag like this is marked Automated check, with no source link, because there is no article behind it.
- Treat it as a reason to look further, not as an answer.
Automated checks can be wrong, which is why we show how sure the check is instead of hiding it, and why every flag carries a Report this flag link. Tell us one is wrong and we look at it.
The modelis the easy part
Calling an AI model takes an afternoon. Keeping up with a playing video, without re-asking a question already answered, is where the work went. Four gates, in order, cheapest first.
Filter in the browser
Captions are de-duplicated, buffered, and screened for complete thoughts on-device. Fragments, questions and repeats never leave the machine.
Check what is already known
A two-tier cache, in memory and in Postgres, answers any claim that has been seen before without a fresh lookup.
Ask the fact-checkers first
Published human ratings are queried before any model runs. A rating a person wrote beats a generated one, so it is tried first, not second.
Then, and only then, the model
Whatever survives all three gates gets a structured-output call, run concurrently and streamed back so the first flag appears without waiting for the slowest.
The next viral "fact"is already being uploaded
Add Claimek before you watch it. Free, no account, ready in under a minute.
Add to Chrome — it's free →