Email Verifier Accuracy: What '99% Accurate' Actually Means

"99% accurate" on an email verifier usually means the vendor checked its results against some set of addresses and most agreed, without telling you which addresses, what counted as correct, or whether "unknown" and catch-all results were left out. There is no independent ground truth. Measure coverage, false-valid rate and false-invalid rate on your own data instead.

This is the longer companion to how to tell if your email verifier is lying, which gives a six-row test that catches the worst failure in a minute. That test tells you whether a tool is guessing. This post is about what to do once a tool passes it, when you want to know how good it actually is.

What vendors claim, as of October 2026

Several verifiers publish a single accuracy percentage, and none of the pages we read defined it. We fetched these homepages on 1 October 2026 and recorded every percentage claim about accuracy or results. These are the vendors' own claims, not audited figures.

Vendor	Claim on its own site	Method published on that page?
ZeroBounce	"99.6% accuracy with real time email validation"; "Lowest 'unknown' Rates: 1.75%"	No
Clearout	"99.73% accuracy"	No
EmailListVerify	"99% Verification accuracy"	No
DeBounce	"97% Deliverability Guarantee"	Linked separately
Bouncer	"<2% Unknown Results"; refund if a deliverable address bounces within 72 hours	Guarantee terms on a separate page
NeverBounce	Not readable: the page returned an error to our fetch	n/a
Kickbox	Not readable: our fetch was blocked	n/a
Hunter	No percentage accuracy claim found on the verifier page	n/a

Sources: ZeroBounce, Clearout, EmailListVerify, DeBounce, Bouncer and its guarantee page, Hunter.

SimpleVerifier does not publish an accuracy percentage, for the reasons below.

Why there is no ground truth

To score a verifier you need to know the right answer for each address, and for a large share of real addresses nobody does. Accuracy is a comparison against truth. In email verification, the truth is only knowable in some cases:

Case	Is the true answer knowable?
Domain does not exist	Yes, from DNS
Mailbox rejected at the RCPT TO step	Yes, the server said so
Mailbox accepted, and a made-up address on the same domain rejected	Yes, as of that moment
Catch-all domain	No. The server accepts everything
Provider that rejects only after the message is sent	No, without sending. See why Yahoo addresses are hard to verify
Server timed out or greylisted	Not at that moment
Mailbox closes between the check and the send	It changes after the fact

The cases in the bottom half are exactly where verifiers differ from each other, and they are often the cases a vendor's accuracy test does not include. A test built from addresses with clear answers will produce a high score for almost any competent tool, and says little about the hard cases you are paying for.

That last row matters more than it looks. A result is a snapshot. Clearout's documentation, for example, makes its "safe to send" status conditional on sending "within 24 hours of the verification time" (Clearout docs), and Bouncer's guarantee has a 72-hour window. Those windows are an honest acknowledgement that a correct answer on Monday can be wrong by the following week.

How a high accuracy number can hide a weak tool

The denominator decides the score, and excluding undecided results inflates it. Here is an illustration with made-up numbers, to show the arithmetic rather than describe any real vendor.

A tool checks 10,000 addresses and returns:

Result	Count	Later found to be wrong
Valid	6,000	60 hard bounced
Invalid	2,000	40 were real people
Catch-all or unknown	2,000	(no claim made)

Counting only the 8,000 addresses it gave a definite answer for, it was wrong 100 times: 98.75% accurate. That is a true statement. It is also a tool that declined to answer for one address in five.

A second tool, on the same list, gives definite answers for 9,500 addresses and is wrong 190 times. Its accuracy on decided addresses is 98.0%, lower. It also leaves you with 1,500 fewer addresses in limbo.

Which is better depends entirely on what you do with the undecided ones. A single percentage cannot tell you, because it has collapsed two different things, how often the tool answers and how often its answers are right, into one number.

The three numbers to ask for instead

Ask any vendor, or measure yourself, for coverage, false-valid rate and false-invalid rate.

Metric	Definition	Why it matters
Coverage	Share of addresses given a definite valid or invalid result	Low coverage means you are left deciding the hard cases yourself
False-valid rate	Of addresses the tool called valid, the share that hard bounce	This is the number that damages your sender reputation
False-invalid rate	Of addresses the tool called invalid, the share that are real	This is the number that throws away customers and leads

A tool can trade these against each other. Calling more addresses valid raises coverage and the false-valid rate together. Calling borderline addresses risky lowers the false-valid rate and lowers coverage. Vendors that advertise both very low "unknown" rates and resolving catch-all addresses are claiming to push coverage up; the question to ask is what that does to the false-valid rate, and on what data it was measured.

For most senders the false-valid rate is the one to protect, because a false valid becomes a hard bounce and a false invalid is only a missed send.

How to test a vendor on your own data

Build a small labelled set from your own sending history, run it through each tool, and count. This takes an afternoon and produces a better answer than any homepage.

1. Build the known-bad set

Export addresses that hard bounced from your ESP or sending tool in the last few weeks, with a 5.x.x code such as 550 5.1.1. These are your known-invalid addresses. Recent matters: an address that bounced two years ago tells you about two years ago.

2. Build the known-good set

Export addresses that replied to you or clicked in the last few weeks. A reply proves a person read the mailbox. Opens are weaker evidence, because privacy features in some mail apps load images automatically.

3. Add the six-row sanity check

Include the fabricated addresses from the six-row test, so a tool that guesses is caught immediately.

4. Run every tool on the same file, on the same day

Results drift as mailboxes change, so run the comparison within a short window.

5. Score it

	Tool said valid	Tool said invalid	Tool said risky / unknown
Known good	Correct	False invalid	Not covered
Known bad	False valid	Correct	Not covered

Then compute coverage, false-valid rate and false-invalid rate for each tool.

Sample size: why 20 addresses is not enough

Small tests produce numbers that look precise and are not. We calculated 95% Wilson score intervals for a few plausible results:

Known-bad addresses tested	Wrongly called valid	True false-valid rate is probably between
100	2	0.6% and 7.0%
200	0	0% and 1.9%
200	4	0.8% and 5.0%
200	12	3.5% and 10.2%

With 200 known-bad addresses you can tell a tool that calls 1 in 50 dead addresses valid from one that calls 1 in 15 valid. With 20, you cannot tell much at all. Aim for a few hundred in each set.

What this test cannot tell you

Your known-bad set comes from addresses that bounced, so it excludes catch-all domains and accept-then-bounce providers by construction. That is fine: it measures the tool where truth exists. For catch-all addresses, the only real test is sending and measuring, covered in should you email catch-all addresses.

Guarantees are more useful than percentages

A guarantee with a bounce threshold and a time window is testable, which an accuracy figure is not. Bouncer's guarantee page, for example, offers money back "If we say deliverable and it bounces within 72 hours after verification (for an objective reason, not just spam)". You can check that against your own bounce log. If you are comparing vendors, read the guarantee terms, note the window, and send within it.

For a broader buyer's checklist, see how to choose an email verification service.

Practical takeaway

Ignore the headline percentage. Ask what was in the test set, whether undecided results were excluded, and what the false-valid rate is. Better still, spend an afternoon pulling a few hundred recent hard bounces and a few hundred recent repliers from your own history, run them through the tools you are considering on the same day, and compare coverage, false-valid and false-invalid rates. For a quick first screen on any tool, the free single email checker and a fabricated Gmail address will do.