Phishing-resistant describes an authentication method that a fake login page cannot capture and replay, usually because it is bound to the real site's domain. With such a method, the proof of identity is created for one specific website, so a lookalike site gets nothing it can use on the real one.
It matters because many common methods can be defeated by a convincing fake page. Passwords and codes typed by the user, and approvals of push notifications, can be handed to an attacker, who then uses them on the real site. Methods such as security keys and passkeys avoid this by checking the site's address.
The term describes a property of the login step, not complete safety. An attacker can still try other routes, such as stealing a session after sign-in, tricking users into approving a malicious application or targeting the account recovery process. Whether a method counts as phishing-resistant depends on how it is implemented and configured.
To use it, prefer such methods for administrators and high-risk accounts first, close off weaker fallback options that would undo the protection and protect the recovery process too.