Here's a good starting point www.simplypsychology.org/loftus-palmer.html
Basically, if an eyewitness is exposed to new information between observing an event a recalling it, they can assimilate that information into their memory of the event.
The new information can be in the form of leading questions.
Most of Loftus' research has been based on videos of car crashes. As well as the one mentioned in the article, they found that participants were more likely to recall a smashed headlight if they had been asked "did you see the smashed headlight?" As opposed to "did you see a smashed headlight?". Participants were also more likely to recall seeing broken glass if they were asked questions about the cars smashing into each other as opposed to bumping into each other.
In the context of the OP, it means that it is possible that someone who was there who then watched a video/read comments/are asked about whether they saw anyone 'taking the kids away' could then write a Facebook account of it, involving a social worker trying to take the children away. It would then snowball- the more people writing about 'taking the kids away' the more people would 'remember' it.
It's important to note that the eyewitnesses are being entirely truthful- their actual memory has been distorted.
I seem to recall another piece of research- can't remember who by, may have been Loftus again but maybe not- that showed that people remember witnessing high profile events, such as a shooting, yet they weren't actually there!