Every viral claim has a genealogy: someone said something, on camera, in context — then a clip cropped it, a caption sharpened it, an aggregator restated it, and by the fifth generation the claim circulating bears a family resemblance to the original at best. Your newsroom now has to decide: report the circulating version, debunk it, or ignore it. All three decisions require the one thing nobody in the chain has: the original.

Tracing genealogy manually means archaeology across reposts and reaction videos, each pointing vaguely at the last. But the original is almost always a public video — a full interview, a hearing, a keynote — and the distortion is almost always measurable: what was actually said, minus what's being shared.

deep_query runs the genealogy hunt: give it the circulating claim and ask for the original source and full context. It researches across the spoken record — including the long-form originals that reaction clips crop from — and reports back with citations: who appears to have said it first, where, and what the surrounding conversation actually was.

get_youtube_transcript (after a quick import) delivers the forensic layer: the verbatim passage, the sentence before, the qualifier after, the question that prompted it. Now the story writes itself with receipts — "the viral clip omits the first half of the sentence, which reads..." — and the correction carries a timestamp instead of a counter-assertion. In the credibility economy, being the desk whose checks land this way compounds.