Current data, if stored, can become a problem in the future if data analysis techniques improve dramatically. (Brute force, technological progress, etc.) Which seems... likely.
In a hypothetical scenario where Nazis come into power in the USA, who is safer, the Jewish person with their real name tied to their Amazon account, who orders menorahs with their voice over Alexa, or the Jewish person who does not own an Alexa, only buys religious items from a physical shop, pays with cash, and offers a fake name to the seller?
Everything that you do or don't do produces data in some sense, but the odds that what you're not doing is being explicitly logged are almost certainly lower than the odds of what you are doing being explicitly logged. Besides, what you're not doing is data point. What you are doing is information. Data does not deterministically produce information, it is merely possible to extract information from data, and it's possible to extract the wrong information from data.
Ergo, I posit that the optimal amount of data to voluntarily hand over under the premise of "I'm doing nothing wrong, so I have nothing to hide" is zero, because voluntarily handing over your data is offering more concrete information than not doing so offers, and because other people, including those who achieve positions of power, can define "doing something wrong" extremely differently than you do.
Who gives their name to a physical shop when paying cash?
Not taking a phone along has been used against people, for example (not a US case).
By the way, buying something in a shop that people can observe is zero protection against actors like Nazis.
Besides the Germans are now on the same team there isn’t much danger.