What do you mean by conflated? The tracking and profile building is how they correlate results with ad impressions. That this data is then also used for training the models and identifying the user coefficients is just a different use of the same correlation data (a website visit is just another type of ad conversion, after all).
We shouldn’t identify the user. We should identify the content. “I want to be shown next to car pictures, because maybe the guy is missing carpets for his car” instead of “We want to be shown to guys 24-30 with interest in cars.” First is tracking the content, second is tracking the user.
