You change an A record at 9:02, reload your site at 9:03 and get the old page. Reload at 9:10: still the old page. A colleague on another network says the new version loads fine for her. Nothing is broken, but it certainly feels that way.
That gap between "I changed it" and "everyone can see it" is what people call DNS propagation. The word is a bit misleading, and once you know what is really going on it becomes much easier to predict how long you'll be waiting.
Nothing is actually being pushed anywhere
When you edit a record, your DNS provider updates its own authoritative servers almost immediately. Those servers are correct from that moment on. What takes time is everyone else's memory.
Every time a resolver looks up your domain, it keeps the answer for a while so it doesn't have to ask again for the next visitor. How long is set by the record's TTL, a number of seconds attached to the answer. Until that countdown hits zero, the resolver will keep handing out the old value, and it has no idea you changed anything.
You can watch the countdown yourself. We asked Cloudflare's public resolver for www.wikipedia.org twice, a few seconds apart, from our server. The final A record came back with a TTL of 91 seconds the first time and 19 the second. A drop of 72 seconds in a few seconds of real time is what you would expect if the two questions were answered by different caches behind the same IP address, which is exactly how large public resolvers work.
Where your old answer is hiding
Between your keyboard and the authoritative server there are usually four or five places that can hold a copy:
- Your browser, which keeps its own short-lived host cache.
- Your operating system, which caches answers for every app on the machine.
- Your router, which often runs a tiny forwarding cache of its own.
- Your ISP's or public resolver, the big shared cache, and usually the one that matters most.
Each layer expires independently. That is why two people can look at the same domain a minute apart and disagree.
How long does it actually take?
It depends on what you changed, because the cache lifetime depends on the record.
- A, AAAA, CNAME, MX and TXT records: roughly as long as the old record's TTL. If it was 300 seconds, most of the world sees the change within about five minutes. If your host's default was 14400 (four hours), plan for four hours.
- Nameserver (NS) changes: up to two days. We queried a .com registry server for
example.comand the delegation came back like this:
example.com. 172800 IN NS hera.ns.cloudflare.com. example.com. 172800 IN NS elliott.ns.cloudflare.com.
172800 seconds is 48 hours, and it is set by the registry, not by you. Resolvers that cached that delegation will keep going to your old nameservers until it expires.
- Brand-new records: usually fast, with one catch. If someone looked up the name before you created it, their resolver may have cached "does not exist". That negative answer is held for the zone's SOA minimum value. For
example.comthat last number in the SOA record is 1800, so about half an hour.
Why a checker shows a patchwork
Run a lookup against resolvers in several countries mid-change and you'll see a mix: some on the new value, some on the old, a few in between. That's normal, and it's useful. Each row is a separate cache with its own expiry time. Our DNS propagation checker shows the TTL next to each answer, so you can see roughly when a stubborn location will catch up.
A small number of resolvers also apply their own minimum or maximum TTL regardless of what you set. It's rare, but it explains the odd location that lags well past the deadline.
What you can do to speed it up
Most of the useful work happens before you make the change.
- Lower the TTL first. Drop it to 300 seconds at least as long ahead of time as the old TTL. If it was 3600, change the TTL an hour or two before the real edit, not five minutes before.
- Make the real change while the short TTL is in effect.
- Keep the old server running. Visitors on slow caches will still land there for a while. Switching it off early is how a harmless wait becomes an outage.
- Raise the TTL again once every location agrees. Longer TTLs are kinder to your DNS provider and slightly faster for visitors.
If you're moving nameservers rather than a single record, our guide to changing nameservers without downtime covers the extra steps.
Checking progress without guessing
First confirm the source of truth. Ask your authoritative nameserver directly, bypassing every cache:
dig +short example.com A @ns1.yourprovider.com
If that already shows the new value, your change is live and the rest is waiting. If it doesn't, nothing needs to propagate yet: the edit didn't save, or you edited the wrong zone.
Then run the lookup from many places at once, and enter the new value as the expected answer so every non-matching location stands out. Finally, clear your own machine's cache so you aren't fooled by it; the steps are in our guide to flushing your DNS cache.
Rule of thumb: wait for one full old-TTL after the change before you start worrying, and two days after a nameserver switch.
When it isn't propagation at all
If the authoritative server is right but a place is still wrong after the old TTL plus a generous margin, look at other causes: a leftover hosts file entry, a stale record at a second DNS provider, a DNSSEC mismatch that makes validating resolvers reject the answer, or a CDN caching the old origin. Our troubleshooting guide on DNS errors walks through how to tell them apart.
Understanding how TTL works takes most of the stress out of DNS changes. Set it sensibly before you need it.