Another factor is that the numbers are almost always "reports" - because there are so many reports generated. People doing this content moderation are very much

Another factor is that the numbers are almost always "reports" - because there are so many reports generated. People doing this content moderation are very much operating on a "better safe than sorry" approach (for obvious reasons). The result of that is that statistics I've seen put it at somewhere between 70-90% of these reports don't actually correctly identify CSAM and are discarded at the first stage. But it's always the much larger, initial number that is used to illustrate the apparent size of the problem.