This suggests that Google/Facebook/Twitter will still be able to track you, assuming you use their websites regularly, but advertising companies that don't have pages frequented by the average internet user won't.

This is going to be a pretty interesting case study in deploying ML in adversarial contexts :)

If I understand this correctly, the obvious counter-measure is for all links on example-recipes.com to go through example-tracker.com, which then immediately redirects to the original website with the linked-to content. Sort of like the weird link URLs in Google's SERPs.