You're right. My guess is most "anonymized" data is lazy--stripping out obvious identifying information. There are quite a few cases where you can pinpoint users from correlating to publicly available data. CGPGrey posted a video today showing a layman's example with Social Security Numbers[1].

One cool thing I heard Apple talk about[2] was injecting noise or subsampling to help mask individuals. Although, I don't see that mentioned in their public page on privacy[3] or their whitepaper[4].

[1] https://youtu.be/Erp8IAUouus

[2] https://www.wired.com/2016/06/apples-differential-privacy-co...

[3] http://www.apple.com/privacy/approach-to-privacy/

[4] http://images.apple.com/business/docs/iOS_Security_Guide.pdf