Did they really say that?
Paste the quote and a link to where it’s from: a podcast, an interview, a hearing, an article or a PDF. Yank reads the source and shows you whether the words are in it, and where.
A check publishes nothing. It says whether the words are in the source, never whether they are true.
The five answers
- Found
- The exact words are in the source, and Yank shows you where: the second of a video or podcast, the passage of an article, the page of a PDF.
- Close match
- Nearly the exact words. The words that differ are printed, never hidden behind a similarity score.
- Distilled match
- Not the source’s literal words, but one passage shares most of them. The passage is shown, so you can judge whether it says the same thing.
- Not found
- Yank read the whole source and the words are not in it. That is a claim about what Yank read, the captions, the transcript, the article text, not proof that nobody ever said them.
- Can’t establish
- Yank couldn’t read enough of the source to say either way: a paywall, a login, or audio that would have to be transcribed first. It says which.
What it reads
A YouTube video with captions, an article anyone can open, a public PDF, and any source Yank has already yanked are checked straight away. A podcast or a video without captions has to be transcribed first, so the check asks you to yank it once; after that, checks against it are immediate.
No model decides a verdict. Yank compares the words against the source it read, with the same guard that checks every line it publishes.
For agents
The same check is a tool: check_quote on the Yank MCP server, or POST /api/check with a statement and a source. AI writes the answer; Yank provides the receipt.