How to run an inbox placement test that means something

A seed-based placement test answers exactly one question: when this message, from this sending path, hit these mailboxes, which folder did it land in at each provider. That is narrow, and it is the only direct measurement of folder placement available to a sender — no provider reports folder decisions back to you. The value is entirely in how you scope the test.

Scope it per stream, not per account

Placement is a property of a sending path, not of a company. Each of these is a different path and needs its own test:

• Each sending domain and subdomain. mail.example.com and example.com have separate reputations.
• Each ESP or mailbox provider you send through. The same copy from a Google Workspace mailbox and from SendGrid will land differently.
• Each IP pool, if you run more than one. On SendGrid, pools are configured under *Settings → IP Addresses*; on SES, the equivalent split is managed vs standard dedicated IPs or the shared pool.
• Each message stream. Postmark separates transactional and broadcast streams explicitly, and they are scored separately by receivers too.

A single "our inbox rate is 87%" number averaged across all of those is not actionable. The useful output is a grid.

The row that is failing tells you where to look. In that example, nothing is wrong with the domain — something about the marketing stream's content, volume or complaint history is wrong at Microsoft specifically.

Test the real message

The most common way to waste a test is to send a simplified version. Keep:

• The real subject, preheader and body, with merge fields rendered as a real recipient would see them.
• The real links, including any click-tracking redirect domain. A tracking domain with no reputation is a frequent cause of filtering, and stripping it from the test hides the cause.
• The real sending path, authenticated the same way. If production sends via SMTP on your ESP, the test must too.
• The real From name and address, and the real unsubscribe header for bulk mail.

Change one variable per test, run before and after, and keep the seed set constant. Two tests with different seed sets are not comparable.

How much to trust the number

Seed sets are small — tens of mailboxes, not thousands. Treat the result as directional:

• Binary signals are reliable. "Microsoft put this in Junk" is a fact about that path and that message.
• Percentages are noisy. The difference between 84% and 91% on a small seed set is usually nothing.
• Seeds have no engagement history. They never reply, star or move mail out of Spam, so filters that weight recipient-specific engagement will treat them more harshly than a real engaged recipient — and more kindly than a recipient who has ignored you for a year.
• Authentication results in the test headers are facts, and are often the most valuable part of the output. Read Authentication-Results for spf=, dkim= and dmarc= and confirm header.from alignment.

The Gmail tab caveat

Gmail's category tabs are not folders. A message sorted into Promotions still carries the INBOX label, so any tool reading placement via IMAP or the Gmail API counts it as inbox — including ours. That is technically correct and practically optimistic: a Promotions placement for cold outreach behaves like a miss, because the recipient may never look at that tab the same day.

What to do about it: do not expect a placement tool to report Promotions reliably, and judge tab placement by sending to a mailbox you control and looking with your own eyes. The content signals that push mail into Promotions — image-heavy HTML, multiple links, bulk-style footers, marketing phrasing — are the same ones to strip from outreach anyway.

A cadence that stays useful

1. Baseline each stream before a change, with the production message.
2. During a warm-up or ramp, test weekly per stream, so a degrading trend shows before the campaign does.
3. After any DNS, ESP or template change, test immediately. Those changes are the ones that break placement silently.
4. Monthly for stable streams, as a regression check.
5. On every complaint spike, test and read the headers rather than guessing at content.

What not to do

• Do not use placement tests as a KPI. They are a diagnostic. Replies and conversions are the outcome.
• Do not send seeds from a path you do not use in production.
• Do not run one test and quote it for a quarter. Reputation moves.
• Do not confuse a spam-word content score with placement. Content scores are generated locally and have no knowledge of your reputation.

MailPilot's seed-inbox network measures placement per provider on the mailbox you actually send from — OAuth-connected Gmail and Outlook, or SMTP and IMAP for SendGrid, Amazon SES, Mailgun, Postmark and custom hosts — alongside daily per-mailbox reputation and SPF, DKIM and DMARC checks, so a placement result arrives next to the reputation trend that explains it.