Support articles love the phrase “allow 24–48 hours for DNS propagation.” That number is a blunt instrument. A recursive resolver that learned www.example.com was 192.0.2.10 with a TTL of 3600 will keep saying so for up to an hour, even if you already pointed the name at a new host. A resolver that never saw the old record will get the new one on the next query. Two users on two ISPs can therefore disagree, and both can be “right.”
The lever you actually have
If you know a move is coming, lower TTL to 300 (five minutes) a day in advance — after the previous TTL has had a chance to die. Make the change. Watch a lookup from a resolver that is not your laptop. Raise TTL again once traffic looks healthy so you are not paying extra lookup latency forever.
Places that ignore your wishes
- Some CPEs and corporate caches clamp very low TTLs upward.
- Browsers and apps have their own short caches.
- CDNs may cache the target of a CNAME independently of your apex TTL.
- NS changes at the parent can take longer because TTLs on delegation are often high.
How to tell “not propagated” from “wrong zone”
If NS records still name the old provider, you did not change what the world uses. If NS is correct but A is old, you are waiting on TTL or you edited the wrong hostname (@ vs www). If one region is new and another is old, that is cache, not a geography filter unless you use traffic steering.