How to test Gmail inbox placement properly

A 99% delivery rate tells you almost nothing. It means Gmail accepted the message at the SMTP layer — not that a person will see it. The number you want is where the message landed: Primary, Promotions, Updates, or Spam. Testing that correctly takes more care than most tools admit.

Three numbers people confuse

If your campaign reports 98% delivered and gets no replies, the gap lives between row one and row three.

Why label-based checks overstate Gmail placement

Gmail applies categories as additional labels on top of INBOX. A message sorted into Promotions is labelled INBOX and CATEGORY_PROMOTIONS. So a test that connects to a seed mailbox over the Gmail API and asks "does this message have the INBOX label?" returns yes for Promotions-tab mail.

That is the single most common source of inflated placement figures. It is not a bug in anyone's arithmetic — it is a definition problem. If a report gives you one Gmail inbox percentage with no tab breakdown, read it as "not in Spam", which is a much weaker claim.

A correct Gmail check reads the category labels as well, and reports Primary separately:

labels: ["INBOX","CATEGORY_PROMOTIONS","UNREAD"]   -> Promotions, not Primary
labels: ["INBOX","UNREAD"]                          -> Primary
labels: ["SPAM"]                                    -> Spam

Build a test that reflects your real sending

A placement test is only as good as its realism. Four things have to match production:

1. The same sending path. Same mailbox or same SMTP credentials, same IP pool, same signing domain. A test sent from your ESP's web UI when production goes through an API is a different sender as far as Gmail is concerned.
2. The same message. Real subject, real body length, real links, real signature, real tracking setup. Stripping the tracking domain out of the test removes one of the things most likely to be scored.
3. Seed mailboxes that behave like strangers. A seed account you personally email every day is not cold. Placement to a mailbox with prior positive interaction is the best case, not the typical case.
4. Enough seeds per provider. One Gmail seed gives you a coin flip, because filtering is partly per-recipient. Several seeds give you a distribution.

Run it as a series, not an event

Placement moves. A single test is a snapshot of one moment in your reputation, which means it cannot tell you whether a change you made helped. Useful practice:

• Baseline before you change anything.
• Test on a fixed cadence — the same time of day, the same seeds — so the series is comparable.
• Test again after each deliberate change: new subdomain, new ESP, new template family, higher volume.
• Keep the raw per-seed results, not just the average. Three seeds in Spam and seven in Primary is a different story from ten in Promotions.

Cross-check with the signals Google still publishes

Seed testing tells you where mail lands. Postmaster Tools v2 tells you what Google thinks of your traffic in aggregate: spam rate, authentication pass rates, delivery errors and compliance status, plus the Deliverability analysis report added in June 2026. Note what is not there: the domain and IP reputation dashboards were retired with v1 — senders were redirected from 30 September 2025 and v1 was gone by the end of 2025 — so there is no reputation grade to compare against. Spam rate is the number to watch, against Google's 0.30% threshold and a self-imposed 0.10% target.

Read the two together. Seeds in Promotions with a spam rate near zero is a content and engagement problem. Seeds in Spam with a rising spam rate is a reputation problem, and the fix is volume discipline and list quality, not a new template.

What not to do

• Do not treat "not in Spam" as inbox placement. On Gmail that conflates Primary with three category tabs.
• Do not test only Gmail. Microsoft and Yahoo filter on different signals; Yahoo and AOL share one stack, so they move together, but neither tracks Gmail.
• Do not use a single seed address. Per-recipient variation will make your results look like noise, because they are.
• Do not send the test to seeds that are also in your warm-up pool. Warmed seeds have interaction history, which is precisely the variable you are trying to hold constant.
• Do not compare a test from last quarter to one from today if your volume, template or sending domain changed in between. You will attribute the difference to the wrong cause.
• Do not rely on open tracking as a placement proxy. Image proxying and prefetching make opens a poor substitute, and tracking pixels themselves can affect placement.

How MailPilot helps

MailPilot's seed-inbox network measures placement per provider and reports Gmail's tabs separately, so a Promotions delivery is never counted as a Primary-tab hit. It runs on a schedule rather than on demand, so you get a series you can reason about, and it tracks each mailbox's reputation daily alongside SPF, DKIM and DMARC checks on the sending domain.