Have you considered how a combination of innocuous data points, such as "browser + city + top 3 popular sites" can make a person uniquely identifiable?
Or any other of the billions of combinations of your browsing patterns or seemingly random daily activities. Your entropy fingerprint, if you will.
Check out "differential privacy" to learn more [0].
We've built a product to help companies identify the more obvious "private data" cases (https://pii-tools.com), but we're not fooling ourselves that we've solved "personal data". Or that the task is even solvable. A dedicated person or algorithm can identify people from surprisingly little information (in the extreme, think Sherlock Holmes). Identification is a matter of degree, rather than a binary "name, IP, email" thing.