Either way, it's impossible to argue the data collection was accidental. You don't send a van out running software without having RTFM and testing it out in some trial runs.
Not impossible at all. Kismet provides data in various different formats. And even then, if what you're doing is extracting particular pieces of data from the traffic capture but not paying much attention to everything else, it isn't unreasonable to not really notice what else you've captured.
I used to occasionally run Kismet during my commute. I was curious about what access points I could see during my route and what state of configuration they were in (with the expectation to scoff at all the default unsecured - actually surprised that those numbers had fallen out in the real world). After doing this for a few months, I was going back through my directory to clean up. Just for giggles I decided to actually look at the caps I had collected and see if there was anything interesting in the packet payloads. Most of it was junk; driving around isn't a particularly good way to snoop on a network. But I did find one email password from a slice of captured POP traffic. So I did end up with someone's sensitive data sitting on my drive for possibly several months despite the fact that I wasn't particularly interested in it or being aware of it.
I suspect this is more or less what happened with Google. Scanning through the Google van captures might have turned up nothing. But Google was doing this on a larger scale so the odds were in the favor of something turning up due to the sheer amount of unsecured traffic out there.