Domain validation: clicking „Verify“ too early makes it slower
Fabian Blechschmidt
Every service that wants proof you own a domain does it more or less the same way: it hands you a TXT record, you put it in your zone, it looks the record up. Let’s Encrypt via DNS-01, Google Search Console, Microsoft 365, your mail provider, your CDN, the shiny SaaS tool someone in marketing signed up for last week. Same dance every time.
And every time, someone does it in this order:
Open the service, get the record.
Click Verify.
„Record not found.“
Open the DNS provider, actually create the record.
Click Verify again.
„Record not found.“
Click Verify eleven more times.
Delete the record, create it again, differently this time.
Complain about the DNS provider in the team chat.
Step 2 is the expensive one. Not because it fails — because of what it leaves behind.
What a failed lookup leaves behind
DNS caches negative answers. If a resolver asks for a name that doesn’t exist yet and gets NXDOMAIN (or NODATA, when the name exists but has no TXT record), it doesn’t just shrug and forget. It writes that down and answers every subsequent question from memory until the entry expires.
This is RFC 2308, and it is not optional behaviour someone bolted on. It is how the system is supposed to work — otherwise every typo’d hostname on the internet would hammer the authoritative servers forever.
So when you click Verify before the record exists, you’re not doing nothing. You’re teaching a resolver that the record is not there, and it will keep believing that for a while. Your record can be live and perfect thirty seconds later and the answer stays „not found“, because nobody is asking your nameservers anymore.
How long is „a while“? Ask the SOA
The negative TTL isn’t a fixed value and it isn’t set on the missing record — it can’t be, the record doesn’t exist. It comes from the zone’s SOA record: the negative caching time is the smaller of the SOA’s MINIMUM field and the TTL of the SOA record itself.
That last number is the one that decides how long your mistake sticks around.
MINIMUM
You wait
300
5 minutes, fine
900
15 minutes, annoying
3600
an hour, painful
86400
a full day, and now you understand why it „doesn’t work“
Some public resolvers cap negative caching at their own ceiling regardless of what your zone says, so you may get lucky. Plan for the number in your zone, not for luck.
If your zone ships with a MINIMUM of 86400 — and plenty of defaults do — lower it before you start onboarding services. It costs nothing and it turns a bad day into a coffee break.
The retry schedule is nobody’s standard
The other half of the question people ask us: how often does the service check again on its own?
There is no answer, because there is no rule. Every provider invents its own. What you see in practice is some kind of backoff — a few attempts close together, then progressively longer gaps, then a give-up after somewhere between a day and three days, after which the whole verification has to be restarted.
Two things follow from that:
You cannot rely on the automatic retry to save you. It might come back in 60 seconds, it might come back in an hour. If there’s a manual „check again“ button, that button is almost always faster than the schedule.
Every early click can push you further down the backoff. Some services count attempts, not time. Spamming the button is the one thing guaranteed not to help.
The actual recipe
Boring, and it works:
Create the record. All of it, exactly as given.
Wait a few minutes. Not for mystical „propagation“ — for your provider’s own nameserver fleet to be consistent. On a managed provider that’s seconds to a minute; on a self-hosted primary/secondary setup it’s your SOA REFRESH, unless you have NOTIFY working, which you should.
Verify it yourself with dig. See below. This is the step everyone skips and it’s the only one that tells you the truth.
Only then click Verify. Once.
The whole point is that step 4 should be a formality. If you already know the answer, the click can’t fail.
Verifying it yourself
Ask your own nameservers first — they have no cache and no excuses:
The gap between those two results is the whole diagnosis:
Authoritative yes, public resolvers no → negative cache. Wait it out. Nothing to fix.
Authoritative no → your record is broken. Fix it, then start the clock.
Want to see exactly how much longer you’re stuck? The TTL in the answer section counts down:
dig TXT _acme-challenge.example.org @8.8.8.8
Run it twice, thirty seconds apart. If the number is falling, you’re just waiting.
The four ways the record is actually broken
When step 3 comes back empty from your own nameservers, it’s nearly always one of these:
The domain got appended twice. Most provider UIs want the hostname relative to the zone. You paste _acme-challenge.example.org into a field that already means „….example.org„, and you end up with _acme-challenge.example.org.example.org. Check the record’s full name in the UI after saving, not what you typed into the field.
The quotes are doubled. TXT values are quoted in zone file syntax. Some UIs add the quotes for you, some expect them, some strip them. If your value shows up as ""abc123"", that’s a different string than "abc123", and the validator will notice.
Something else already answers that name. A leftover CNAME on the same label wins over your TXT, or an old TXT record from a previous attempt is still sitting there and the service reads the wrong one. Multiple TXT records at one name are legal — but a CNAME next to anything else is not, and resolvers handle it inconsistently.
A wildcard is covering for you. If the zone has *.example.org, a lookup for a name that doesn’t exist returns the wildcard’s data instead of NXDOMAIN. You get an answer, it just isn’t yours.
Do not delete and recreate
The reflex, when verification fails, is to delete the record and make a new one. Don’t.
Deleting it recreates exactly the condition you’re trying to escape: for as long as it’s gone, every lookup produces a fresh negative answer with a fresh negative TTL. You reset the clock you were waiting out. And on services that count attempts, you’ve spent tries you don’t get back.
If the record is right, leave it alone and wait. If the record is wrong, edit it. There is no case where deleting helps.
TL;DR
Create the record. Wait a few minutes. Confirm it yourself with dig — authoritative server first, then a public resolver. Then click Verify, once.
Clicking Verify before the record exists doesn’t just fail, it plants a negative answer in the resolver cache that outlives your fix. How long depends on your zone’s SOA MINIMUM, which you should go look at right now — and lower, if it’s still at some vendor default measured in days.
The fastest domain validation is the one where you waited five minutes first.
Running into this during a shop migration or a mail setup? We do this dance often enough that we’ve made the mistakes for you — drop us a line.