Your sending platform reports 99.4% delivered. That means a receiving server answered 250 OK and took the message. It does not mean a person saw it. Between acceptance and a human eye sits a filter that can drop your mail into spam, into a tab nobody opens, or into nothing. Inbox placement testing tells you which — before the campaign goes out, not three days later.
Accepted by the server is not visible in the inbox
Four things can happen after that 250. The message lands in the inbox, as you assumed. It is filed silently into spam — the transaction succeeded, so your logs look perfect. It lands under Gmail's Promotions or Updates tab, where response rates for a transactional message collapse. At a Microsoft 365 tenant it is pulled into quarantine, visible only in a daily digest or to an admin. Or it is accepted and silently discarded — rarer, but real, and no bounce is generated. None of that produces feedback you can see, which is why seed testing exists.
Building a seed list that matches your audience
A seed list is a set of real mailboxes you control at the providers your recipients use: Gmail, Outlook.com, Yahoo, iCloud, plus one corporate Microsoft 365 tenant and one Google Workspace account, since business filtering is set by an admin and behaves nothing like the consumer product. Add the regional providers that matter in your market — GMX in Germany, Free.fr in France, Mail.ru for Russian speakers. Weight the list from your own subscriber data, not market-share articles.
A disposable inbox from our temporary email service covers the throwaway half of this work: confirming that a signup or password-reset mail leaves your system at all, that the template renders, that every link resolves. Be honest about the limit — it tells you the message was sent, not how Gmail would have foldered it, because it is not Gmail.
Verify the addresses before you test anything
An invalid seed address is not a neutral zero — it is a hard bounce, generated by your own test, against the identity you are measuring. Run every address through an email verification tool first and drop whatever fails syntax, has no MX record, or is refused at the recipient stage. Mailboxes from a previous round are the usual culprit, since Yahoo and Outlook deactivate unused accounts.
Send the test the way you send production
The test only means something if every variable a filter looks at matches the real send: same domain and subdomain, same IP or pool at the same ESP, same From name and address, same DKIM key, same template and links, and a real subject line, not "test 4". Do not hand-send from your own mailbox or use your ESP's test-mode flag — both measure an identity that does not exist in production.
What to record on every run
For each provider, record four things: placement (inbox, spam, which tab, quarantine, or missing), time to delivery, whether images loaded without a click, and the authentication result. One run is a snapshot; the value is in the series. A sheet with one row per run and one column per provider, kept for a year, shows the week your Outlook placement started sliding.
Read the headers, not just the folder
Open the raw source of each message that arrived. Authentication-Results is the receiving side telling you what it concluded: spf=pass means the connecting IP was authorised by the envelope sender domain, dkim=pass means the signature verified and the body was not altered in transit, and dmarc=pass means one of those aligned with your visible From domain. A DKIM pass with a DMARC fail is the classic alignment mistake — you signed with your ESP's domain, not your own. Read the Received chain from the bottom up to see where time was lost, then find the score the provider stamped on: X-Spam-Status on many servers, or X-Microsoft-Antispam with SCL and BCL values at Microsoft 365, where SCL 5 or above means the tenant called it spam. If those records are new to you, our explanation of SPF, DKIM and DMARC covers what each does.
Reading the pattern across providers
The shape of the result tells you where to look:
- One provider bad, the rest fine: reputation or filter policy there. Check its postmaster tools and your complaint rate.
- Everything bad at once: authentication, content, or domain and IP reputation. Start with the headers; a DMARC failure explains a lot.
- Intermittent from run to run: volume and ramp, or a noisy neighbour on a shared IP.
- Inbox, but in Promotions: content classification, not reputation. Fewer tracked links, more plain text.
Content and link checks worth running
Some things push mail toward spam regardless of reputation. An image-only email gives filters nothing to read and matches a classic evasion technique. URL shorteners are heavily abused. Link domains that do not match the sending domain weaken the tie between sender and content. A missing plain-text part is a cheap fix, and a hidden unsubscribe raises complaints.
Bounces contaminate the whole test
If your seed run rides along with a broader send to a list you have not cleaned recently, the numbers describe your list, not your placement. Lists decay by roughly a fifth a year, and a hard bounce rate above about 2% depresses placement for everything in that send, seeds included. Clean the segment with the email verifier first, suppress every past hard bounce, then test placement.
When to re-test
Before every significant campaign. After any DNS change — a new SPF include, a rotated DKIM key, a tightened DMARC policy. After switching ESP or adding a sending IP. After a template redesign. And monthly even when nothing changed on your side, because the filters change on theirs.
What seed testing cannot tell you
It measures a sample. Gmail applies no single verdict to your domain; placement varies per recipient, based on how that person has treated your mail before. Your seed mailbox has no engagement history, so it sits at the pessimistic end — a subscriber who replies may see the inbox on a run where your seed saw spam. Commercial panels are larger, but their mailboxes are strangers to you too. Every placement figure is directional.
Mistakes that waste the run
Testing only to your own address proves the pipeline works and nothing more. Logging "delivered" when the message landed in Promotions hides the real problem. Staging measures an identity your recipients never encounter. Declaring victory after one clean run ignores that placement drifts — two clean runs a week apart is the minimum honest standard. If your tests keep coming back spam, our companion piece on why your emails go to spam works through the causes in the order worth checking them.
