Pick an email verification vendor and its homepage will show you an accuracy figure. It is almost always a high one, it is almost never accompanied by a method, and it answers the question every buyer actually has: will this tell me the truth about my list? Kawaa's site has no such figure, and this post explains why: the honest answer to “how accurate is it?” is not a number, and a vendor who prints one anyway is asking you to trust something nobody measured.
The short version
- There is no ground-truth list of real mailboxes to score a verifier against
- Catch-all domains have no knowable answer, so any single number hides the hardest cases
- The vendor chooses the test set, and a flattering one is always available
- What you can check instead: does it say unknown instead of guessing, does a verdict carry its reason, and can you test your own addresses free
The number that isn't there
Accuracy marketing works because the question is reasonable. You are about to trust a service with your sender reputation, the slow-to-build asset that decides whether your mail reaches the inbox, and you want to know how often it is right. A percentage feels like the answer.
But a percentage is a measurement, and a measurement needs a method: a test set, a correct answer for every row in it, and a rule for scoring the hard cases. Ask where a vendor's figure comes from and the answers get thin, because the thing the figure claims to measure cannot be assembled.
What accuracy would have to mean
To say “accurate” about an email verifier you need a list of addresses whose true state is known: which mailboxes exist right now, at the moment of the check. Nobody has that list. A deep check can ask the receiving server whether it would accept an address and close the conversation without delivering anything (Kawaa's does, when the server cooperates), but acceptance is not existence: the only final proof is a delivered message, and sending to a bad address is exactly the harm verification exists to prevent. A bounce is a fact, but it arrives after the send you were trying to protect.
Even the definition is slippery. Is a catch-all address “correct” when the verifier says deliverable? The domain accepts every address at the door, so there is no certain answer to check. Acceptance is not a mailbox, and the later send can still bounce or be silently discarded. Is a disposable address wrong because it is real, or right because it is worthless? Two vendors can score the same result opposite ways and both call themselves accurate.
Why it can't be measured honestly
No ground truth
A benchmark needs addresses with known outcomes. The known-bad addresses in any public test set are known because they bounced for someone, once. A mailbox that bounced last quarter may be closed, recycled or reprovisioned today. The set ages the day it is frozen.
The hardest cases have no answer
A catch-all domain accepts every address, so no check can confirm a specific mailbox there. A single accuracy figure either excludes those domains (flattering the number) or scores a guess as a result. Catch-all is where verification goes to be humble, and catch-all is a common setup on business domains.
The seller picks the test
A vendor measuring itself chooses the list, the moment and the scoring rule. A list heavy on old consumer addresses is easy; a list of fresh corporate catch-alls is not. None of this is visible in the number, and none of it is audited.
The answer moves
Mailboxes close, fill up, and get reprovisioned. A verifier that was right about an address in January can be wrong about the same address in March through no fault of its own. That is why a figure with no date and no method tells you nothing about your list next month.
What a high number does tell you
That the vendor is willing to print a figure it cannot support, on the page where it asks for your trust. Treat the presence of the number as a data point about the vendor, not about the product.
What to ask a verifier instead
The useful questions are the ones a vendor can answer with its product rather than its marketing:
- Does it ever say unknown? A service that returns a confident verdict on every address is guessing on the hard ones. Unknown is an honest answer; a forced one is not.
- Does a verdict carry its reason? “Invalid” because the domain does not exist and “invalid” because the mailbox was refused are different problems with different fixes. A reason code lets you act; a bare verdict does not.
- What does it do with a catch-all domain? Flagging it, and telling you the mailbox question is unanswerable there, is the honest ceiling. Anything more confident deserves a hard look.
- What does an inconclusive result cost you? Some services refund unknowns; others charge for them. On a large list the difference is real money.
- Can you test your own addresses for free? Your list is the only benchmark that matters to you. A vendor confident in its results lets you check a sample of your own data before you pay.
What Kawaa publishes instead
In place of a number, Kawaa shows its working on every result:
A status with its reason
A completed check carries a status, a quality score from 0 to 100, and (for most statuses) a reason code naming how the answer was reached: a refused mailbox, a domain that does not resolve, a server that would not talk. The reason is the part you can act on, and it is on the result rather than in a sales deck.
Unknown, said out loud
When a check cannot be completed (a provider that throttles or blocks the probe, a catch-all domain, a timeout), the result says so. Kawaa does not convert an inconclusive check into a confident verdict, because a guessed “valid” is how a list full of traps ends up labelled clean.
Catch-all flagged, not resolved
Catch-all domains are identified as catch-all, and when one is detected the result adds a confidence score from address patterns and AI analysis: a risk signal to weigh, presented as one. What it is never presented as is proof the mailbox exists.
A free way to check our work
The free email checker runs a real verification with no account, three a day, and shows the status, the score and the reason. A free plan adds 100 credits, enough to run a slice of your own list and compare the results against what you already know about it. That is the only accuracy measurement we trust: yours, on your data.
If a verifier's accuracy figure is the thing selling you, ask for the method behind it. If there is one, it will survive the questions above. If there isn't, you have learned something the number was there to hide.
