We're talking generalized metrics. I have no idea which ones - I wasn't claiming to have solved the problem. The point is that if you can identify a general characteristic that is being used in a way which disproportionately contributes to a particular outcome then you can filter on that.
Estimated user age is an example of a metric largely unrelated to concerns regarding free speech. I doubt it has much relevance to the problem we're taking about here but hopefully you can imagine that prohibiting the targeting of ads or the curation of an algorithmic feed based on that metric would not be expected to unduly disadvantage any particular sort of speech.
Estimated user age is an example of a metric largely unrelated to concerns regarding free speech. I doubt it has much relevance to the problem we're taking about here but hopefully you can imagine that prohibiting the targeting of ads or the curation of an algorithmic feed based on that metric would not be expected to unduly disadvantage any particular sort of speech.