I don't think that would suffice, though it's certainly a reasonable weighing consideration. And what they specified as the technique wouldn't suffice, either, though it would reduce the fake news considerably. And invite a "fake news" arms race.
At some point verification requires somebody you reasonably trust actually going and checking. I suppose a video might count, but you need to consider that every video is going to provide you with a biased view, selected by the angles from which observation happens. And fake videos aren't unknown, so you need to start including ways to detect whether the video was altered...which invites another arms race.
And you'll never be really sure. You shouldn't be sure of even things that you personally saw, because memory is fallible.
So consider this program as something that improves the signal-to-noise ratio. If that's the goal there are lots of things that can be done, and the early versions should be succeeded by more advanced versions for a long time. And even the early things are useful.
For comparison, consider the progress in e-mail filters. And the twin problems of false positives and false negatives. Now imagine trying to do without ANY e-mail filter.