How to Fix a Domain That Resolves for You but Not for Everyone Else
Your domain loads for you but not for your client, or the other way round. Here is how to find where the stale answer lives and fix it.
You moved your site to a new host, or you changed an A record, and now your client in another city sees the new site while you still see the old one. Or the other way round: it works fine on your laptop and your customer swears the domain does not exist. I have been on both sides of that call. Two people can ask for the same domain in the same minute and get two different answers, and both answers can be correct from where each person is standing.
Work out which side is stale
Before you touch anything, decide who is seeing the old answer. The person looking at the old site is the one with a stale cache. That sounds obvious, but I have watched people roll back a perfectly good migration because they assumed their own screen was the truth. Your screen is one resolver's opinion. The new site being visible somewhere means the change went live, and the old site being visible somewhere means a cache has not expired yet. Both can be true for hours, and neither means anything is broken.
Check the source of truth, not your browser
Your browser asks a chain of caches, so it is the worst place to find out what your DNS actually says. Start with our DNS Propagation Checker instead. It asks resolvers in several countries the same question at the same moment and lines the answers up. If most of them return the new address, your change is live and you are watching caches expire. If every resolver still returns the old one, the change never went out. Then run a DNS Lookup against the authoritative nameserver itself. If that answer is wrong, the problem is at your provider, not in propagation.
A worked example, with real numbers
Here is a move I did recently, with the numbers left in. A small site pointed at 198.51.100.23. I changed its A record to 203.0.113.45, and the old record had a TTL of 3600 seconds, one hour. I made the change at 10:00. At 10:05 the propagation checker showed the new address in London, Frankfurt and Singapore, and the old address in two US resolvers and at my own ISP. By 11:00 every resolver agreed. The maths is simple once you see it. A resolver that last looked the record up at 09:30 holds the old answer until 10:30, because its one-hour countdown started at 09:30. So the wait is not one fixed delay for everyone. It is the old TTL, counted from the moment each resolver last asked. If your old TTL was 86400, a full day, that is how long the slowest resolver can take, and that is normal, not a fault.
The three caches between you and the answer
When the checker shows the new record everywhere but your screen still shows the old site, the stale copy is on your side of the connection, and there are three places it can hide. First, your browser keeps its own short DNS cache, about a minute in Chrome, so a fresh tab is sometimes enough. Second, your operating system caches lookups. On Windows, open a command prompt and run ipconfig /flushdns. On a Mac, run sudo dscacheutil -flushcache followed by sudo killall -HUP mDNSResponder. On Linux with systemd, systemd-resolve --flush-caches does the same job. Third, your router caches answers for your whole network, and restarting it clears that copy. If none of that helps, the cache belongs to your ISP, and you cannot flush it. Switch your device to a public resolver such as 1.1.1.1 or 8.8.8.8 for a quick test. If the site loads there and not on your ISP, you have your answer, and the only cure is waiting out the old TTL.
The mistake that wastes a whole day
The classic error is lowering the TTL after you make the change. It feels like the right fix, and it does nothing at all. A resolver that already holds your old record keeps the old TTL, so your new, shorter TTL only reaches resolvers that ask after the change. The right order is the other way round: a day or two before a planned move, drop the TTL to 300 seconds and wait for the old, long TTL to expire. Then make the change, and every cache in the world holds the new answer for at most five minutes. There is a second trap in the same family. If anyone looked up a record before it existed, the resolver can cache the "does not exist" answer as well. That is negative caching, and it is why a brand-new subdomain can look dead for a while even though the record is correct.
When waiting will not fix it
Propagation has a deadline, and past it, something else is wrong. If the checker still shows a split, or the old address everywhere, after the old TTL has fully expired, stop waiting and check the setup. First, look for authoritative nameservers that disagree with each other, which means the zone has not synced at your DNS host. Second, you may have edited the record at the wrong provider: your nameservers point at one company while you edited the zone at another, so the live record never changed. A WHOIS Lookup shows which nameservers the domain actually uses, and it settles that question in seconds. Third, you may have edited the wrong record type, an AAAA instead of an A, or a CNAME that still points at the old host. And if you changed the nameservers themselves, expect a longer window: nameserver changes go through the registry and can take up to 48 hours, sometimes 72.
What to do first, in order
Run the propagation checker and read the split. If most resolvers show the new address, the change is live, so flush your browser, operating system and router caches, then test on a public resolver. If the checker shows the old address everywhere, check the authoritative record and confirm you edited the zone at the provider your nameservers actually point to. Then look at the old TTL and give it that long. If the split is still there after the old TTL has expired, the fault is in the setup, not in propagation, and that is the point to contact your DNS host with the checker results in hand. If you want the background, our guide on DNS propagation explained covers the whole chain. Most of the time, though, the answer is simpler than it feels: one cache, one countdown, and a wait you can measure.
A new post lands here every day. If there is a topic you want covered, email ask@websmartstudios.com.