I've only seen it once or twice (in a week). So, it's pretty intermittent.
And, of course, one doesn't think about what they DID immediately prior to seeing it so there are few "clues" that could be offered.
IMnsHO, the IDEAL use for "Google Glass" would be for developers to be able to replay actions THAT THEY MAY HAVE OBSERVED (but weren't prepared to document BEFORE they happened). Imagine being able to answer the question: "What were you doing when this happened?"
[Can you even BUY Google Glass, anymore? Or, does it rely on something server side (as most Google products do -- so they can have their fingers in your pie) that may have been "turned off"?]