Every claim, checked against live search

The failure mode of watching YouTube to learn is not that videos lie to you. It is that a confident voice, good editing, and a number on screen are indistinguishable from a confident voice, good editing, and a correct number on screen.

Briefed pulls the checkable claims out of a video and checks them.

How a claim gets a verdict

First, claims get separated from everything else. A video is mostly not claims — it is framing, opinion, narrative, jokes, and speculation about the future. None of that is checkable, and pretending otherwise produces noise. What remains are statements that could be true or false: a statistic, a date, an attribution, an assertion that one thing caused another.

Each of those is then checked against live Google Search results — current sources, not a snapshot of what a model happened to memorise during training. That distinction matters most for exactly the claims people most want checked: recent events, current figures, anything where the answer moved in the last year.

Each claim comes back with one of four verdicts:

• True — the sources support it.
• False — the sources contradict it.
• Misleading — technically defensible but framed to imply something the evidence does not support. This is where most of the interesting results land.
• Unverifiable — checkable in principle, but the sources do not settle it.

Why “misleading” and “unverifiable” earn their place

A two-way true/false split would be cleaner and considerably less honest.

Misleading is where most real-world persuasion lives. A statistic can be accurate and still be doing dishonest work — the right number over the wrong time window, a real study with a sample of nine, a correlation narrated as a cause. Marking that “true” because the digits check out would be technically correct and practically useless.

Unverifiable exists so the tool can admit a limit. If nothing reliable addresses a claim, the honest output is “I could not confirm this”, not a confident verdict in either direction. A fact checker that never says I don’t know is telling you less than it appears to.

What to do with the results

Treat the verdicts as triage, not judgment. They tell you which two minutes of a 40-minute video deserve your scepticism — which is exactly the thing that is expensive to work out on your own.

That is also why fact checking pairs naturally with the clickbait detector. One asks whether the video’s claims hold up. The other asks whether the video delivered what its title promised. A video can pass one and fail the other badly, and knowing which is which tells you a lot about whether to trust the channel next time.

For anyone doing this at volume — researchers sifting sources, students checking a lecture against the literature — the verdicts are saved alongside the summary and stay searchable across every video in a project.