Every vendor in this category publishes an accuracy percentage. ZeroBounce says 99.6%. NeverBounce says 99.9%. Clearout says 99.73%. Reoon says over 99%. Read those next to each other and the market looks solved.
Then Hunter ran a benchmark across 40,000 verifications and measured the field at 63% to 70%. Hunter is an interested party and they published the method anyway, which is why the number stuck.
Both sets of figures are arithmetically true. They are dividing by different things. The vendor number asks: of the addresses where we returned a definitive verdict, how often were we right? The addresses where the server refused to answer never enter that calculation.
That exclusion is not a rounding error. It is where the entire difficulty lives. Catch-all domains, Yahoo's anti-harvesting policy, Microsoft throttling, greylisting, servers that drop the connection: these are the addresses a verifier cannot resolve, and removing them from the denominator is what makes 99.6% possible.
What a second number prevents
Coverage is the share of a list we returned a definitive verdict on. Accuracy is how often those verdicts were right. Published together, they cannot be gamed in the direction that hurts you.
Consider a verifier that resolves 70% of a list at 99% accuracy, and one that resolves 95% at 92%. The second sounds worse and is probably more useful, because it made a call on far more of your list. With one number you cannot tell those two apart, and the incentive runs toward the first.
Worse, a verifier can move from the first shape to a 99% headline by labeling difficult addresses as valid. Accuracy barely moves, coverage looks unchanged because nobody reports it, and your bounce rate rises. The number improved and the product got worse.
Where this argument costs us something
The short version
- Ask any verifier what share of your list came back unresolved. That figure is more informative than their accuracy claim.
- Run the same sample through two vendors and compare the unresolved counts rather than the marketing pages.
- Treat a single accuracy percentage with no denominator as marketing, whoever is publishing it.
Questions people ask
Is a 99% accuracy claim dishonest?
Not arithmetically. It is calculated on the addresses the verifier chose to judge, which is a real number about a smaller set. It becomes misleading when it is presented as a statement about your whole list.
Why does Hunter's benchmark show such lower numbers?
Because it measured against a full dataset including the addresses nobody can resolve. Different denominator, same underlying reality.