The causes are specific and each is a decision by the receiving server. Yahoo and AOL accept every recipient to defeat harvesting. Microsoft throttles repeated connections. Gmail and M365 block probes from cloud infrastructure. Greylisting returns a temporary rejection by design. A connection is refused or times out.
None of those is information about your address. They are facts about the server's posture, and reporting them as a verdict about the mailbox would be inventing data.
How a vendor treats this bucket is the single most revealing thing about them. Report it honestly and the coverage figure drops while the accuracy figure means something. Resolve it to valid and the published accuracy rises, because the hard cases left the denominator.
So the useful question when comparing services is not what accuracy do you claim. It is what share of my list will you decline to judge. A vendor reporting 2% unknown on a business list has either extraordinary infrastructure or a habit of guessing, and the second is far more common.
For your own sends, treat unknown as a separate segment rather than a deletion. Mail it apart from your verified addresses, watch what bounces, and you have converted an unanswerable question into a measured one.
A pile of 1,500 unknowns, sorted by cause
Say you check 25,000 addresses and 1,500 come back unknown, which is 6%. Sort them by what the server did before you decide anything. Suppose 600 were temporary: timeouts and 4xx replies from busy or greylisting servers. Another 500 got a 252 from hosts that accept everything by policy. The last 400 were blocked outright with a 530.
Each of the three piles needs a different move. The 600 temporary ones are worth a second run tomorrow, and since unknowns aren't billed with us, the retry costs nothing unless it produces a verdict. Say 350 resolve. The 252 and 530 piles won't change however often you ask, because the refusal is that server's policy and nothing about the moment.
You're left with 1,150 addresses to mail as a small, separate send, and a rule for next time.
Don't let a spreadsheet filter turn unknown into invalid
This is where most real contacts get lost, and it takes one click. Someone opens the results file, filters the verdict column to valid, copies those rows into the email tool and deletes the rest. Every unknown address goes out with the invalid ones, and nobody decided that it should.
Set the export up so that can't happen. Split the results into separate tabs or files: valid, invalid, catch-all and unknown. Name the unknown file with the date, because its contents go stale the way any verdict does.
In your CRM, store the verdict as text and don't squeeze it into a yes-or-no field. A checkbox has no way to say nobody knows. Add the date of the check next to it. Six months from now, the person planning a campaign can see that these 1,150 contacts were never judged. That's a very different thing from being judged bad.
Related questions
Should I delete unknown addresses?
No. Many are real. Send to them as a separate batch so any bounces stay out of your main campaign.
What share of a list comes back unknown?
It depends entirely on the domain mix. Lists heavy in Yahoo, AOL and throttling corporate hosts produce far more than a consumer Gmail list.
Do you charge for unknown results?
No. We reached no verdict, so there is nothing to bill for.
Is unknown the same as catch-all?
No. Catch-all means the domain accepts everything, which is a specific finding. Unknown means the server would not answer at all.