Email verification accuracy: how to evaluate the evidence
An accuracy percentage is difficult to interpret without knowing the test set, the definition of a correct result and the treatment of unknowns. Evaluate a verifier against decisions in your own workflow, not just a single headline number.
Open the relevant free tool →Certainty differs by check
Define what “correct” means
Are you testing format errors, nonexistent domains, recipient rejection or future inbox delivery? These are different targets. A tool can be excellent at syntax and still have many unknown mailbox results. Decide which errors are costly for your product. Rejecting a legitimate customer at signup and retaining a stale marketing contact have different consequences.
Build a controlled sample
Use addresses you control or have permission to test, with known expected categories. Include invalid syntax, a nonexistent test domain, an explicit no-mail domain, a known working mailbox and a catch-all setup if you administer one. Do not use a random purchased list as ground truth: you may not know which addresses were valid when the comparison was made.
Keep unknowns in the report
A service that labels difficult cases unknown may be more honest than one that forces a binary answer. Report how many checks were conclusive, then assess correctness within relevant categories. Do not count unknown as automatically wrong or quietly remove it from the denominator. Explain the policy so a comparison can be repeated.
Compare like-for-like settings
Check whether SMTP and catch-all probes were enabled, whether tests happened at similar times and whether rate limits affected results. Record the response reason and duration. If one run uses only DNS and another attempts mailbox probes, their overall statuses are not directly interchangeable. A provider can also update behaviour between runs.
Evaluate the next action
Measure whether the result improves the workflow: fewer obvious input mistakes, clearer review segments or more useful diagnosis of failed deliveries. Keep actual sending outcomes separate from verification outcomes. Inbox placement depends on factors outside an address checker’s view. In EmailValidly, the displayed score is a heuristic summary of checks, not a calibrated probability that the next message will arrive.
Put it into practice
- Define the target outcome.
- Use a controlled, authorised sample.
- Report unknown coverage explicitly.
- Compare settings, timestamps and reasons.