We Verified 2,482 Scraped Emails of Known Age. Here Is How Fast They Die.

The setup

Our enrichment pipeline records when it first saw each email on a business website. That gave us something rare in this industry: a pool of scraped addresses with honest timestamps. We sampled 2,491 unique addresses across six monthly cohorts, February through July 2026, and verified every one through MillionVerifier SMTP checks on July 30, 2026. Nine lookups failed on network errors, leaving 2,482 measured addresses.

Every address is a business email scraped from the company's own website, found via Google Maps sourced prospecting. No pattern-guessed addresses are in this sample. Each check returns one of four states: verified (the mailbox exists), invalid (the mailbox does not exist, mail bounces), catch-all (the domain accepts everything, so the check proves nothing), or unknown (the server would not answer decisively).

The results, cohort by cohort

The dead rate below is calculated on decisive results only: invalid divided by verified plus invalid. Catch-all and unknown addresses are excluded from that ratio and reported separately, because counting them either way is how vendors cook their numbers.

Three findings stand out.

First: even fresh lists are leaky. Addresses scraped within the last month were 9.6 percent dead among checkable results, consistent with the roughly 1 in 12 we measured in an earlier, separate sample. A scraped list starts life with bounces already in it, because websites list addresses that stopped working long before we ever saw them.

Second: older cohorts trend worse. The three-to-four-month cohorts ran 13 to 17 percent dead, roughly one checkable address in seven. If you scraped a list in spring and are sending to it now, a meaningful slice of it has died while it sat in your spreadsheet.

Third, and this is the finding we did not expect: age is not the whole story. Our June cohort (only about six weeks old) came back 17 percent dead, worse than the older April cohort, while the small May cohort was just 4.4 percent. These cohorts come from different search batches: different industries, cities, and site types. Composition moved the dead rate as much as age did. Any vendor quoting one universal accuracy number is averaging away exactly the variation that determines whether your campaign bounces.

The third of the list nobody talks about

The single largest bucket after verified was not invalid. It was catch-all: 842 addresses, 33.9 percent of everything we measured. On those domains the mail server accepts any address, real or fake, so no verification service on earth can prove the specific mailbox exists. Another 7.7 percent returned unknown.

Hold that against the marketing you see in this category. If a tool claims a 95 or 99 percent accuracy rate and does not say how it counted catch-all domains, the claim is unfalsifiable by construction. We wrote about this pattern in detail in what "95 percent accuracy" actually means. Our own approach is to label the four states honestly and let you filter on them, because a risky address labeled risky is useful, and a risky address labeled 99 percent accurate is a bounce with good branding.

What to do with this

The operational lessons are short.

Verify at send time, not at scrape time. A verification result from months ago has drifted with the list. Inside Lyre Leads, verification runs against the live mailbox when you need it, and every address carries its state: verified, catch-all, unknown, or rejected, plus its provenance, scraped versus pattern-generated and then confirmed. We never blend those into one optimistic number.

Distrust undated accuracy claims. The age of the data behind an accuracy number changes it by half or more, and cohort composition changes it again. A number without a methodology date and a catch-all policy is decoration.

And budget for decay. If a list matters to you in October, do not verify it in July and trust it. Re-verification of a previously checked address is cheap; a damaged sending domain is not. That trade is the entire economics of choosing a platform over raw scraped rows.

Methodology notes for anyone citing this: sample is 2,491 unique business emails scraped from company websites (deterministically sampled per cohort from our production data), 2,482 successfully checked via MillionVerifier SMTP verification on July 30, 2026. Dead rate is invalid over (verified plus invalid). February and May cohorts are small (17 and 45 decisive results); treat those rows as indicative. Cite freely with a link.