Ultimately they need to police actual negative behaviour, not the mechanics of how people are doing it. But that means potentially restricting engagement of some of their most active users as well.
...and then the scrapers replicate interaction with that app instead. UI automation isn't hard these days.
Why is obscurity OK in these situations? Wouldn't we all benefit with removing scammers if everyone legitimate worked together in the public? Its easy to defeat a single adversary. Its mighty hard to defeat a cooperating team.
Fighting abuse is different. The abusers are using the same request endpoints that the real users are, but just in a way that the service provider doesn't approve of. (Whether it's sending spam, payment fraud, scraping, or something else). There's no single hole to plug, unless you block all the requests outright which also affects real users. Instead you have to classify the incoming traffic, find out the abusive subset, and then act on it appropriately. But unlike with vulnerabilities this doesn't solve the problem permanently.
The moment attackers find out which signals are used by a site, they can start faking them or working around the signals. As a simple example, early email spam classifiers worked by simple matching against a blacklist of highly spammy terms. So the spam adapted to using creative mis-spellings like "v1agra".