Important email keeps landing in Promotions.

2026-08-26 · guide · Gmail tabs and categories

Gmail's tabs are a classifier, and like any classifier it misses. When an invoice, a booking confirmation or a human being lands under Promotions, the cost isn't annoyance — it's that Promotions is where mail goes to be ignored. Four fixes, from fastest to most permanent, and the reason the problem keeps coming back.

1. Drag it, and answer the prompt

Drag the message from Promotions to Primary (on mobile: Move to → Primary). Gmail asks whether to do this for future messages from that sender — say yes. This is the built-in training loop, and it works tolerably for senders that email you often.

Its limit: the training is per-sender, invisible, and decays. Senders that rotate their from-address (most transactional platforms) reset the lesson each time.

2. A filter that forces the category

The permanent version of fix 1:

1. Open the message → ⋮ → Filter messages like this (or Search options on the web).
2. Match on the sender or domain (from:(@yourbank.com)).
3. Create the filter with Categorize as: Primary.

Filters outrank the category model, so this never regresses. It also never generalises: every new sender that matters is another filter you haven't written yet.

3. Star or mark important

Starring a message and marking a sender important (the ⌄ arrow / importance marker) both feed Gmail's importance model, which influences — but does not override — the tab decision. Cheap to do, weakest of the four.

4. Turn tabs off entirely

Settings → Inbox → Categories, untick everything but Primary. No more misfiled mail, because there is no more filing — you have traded a wrong classifier for no classifier, and everything competes for your attention again. For plenty of people this is honestly the right call.

Why this keeps happening

Every fix above is a correction with no explanation attached. Gmail never tells you why a message was judged promotional — bulk headers? sender reputation? an unsubscribe link in the footer? — so you can't predict the next miss, only react to it. And your corrections train a model you can't inspect, which is why the same class of mistake returns with each new sender.

This is the specific gap Klorn was built for: every message gets exactly one lane with the reason attached ("clear marketing", "you reply to this sender"), moving a misfiled row is itself the training signal, and two identical corrections make that sender's lane deterministic — the model is skipped entirely from then on. Nothing learned can ever silently hide a sender: learned rules may promote mail to visible lanes, never bury it. Misclassification doesn't disappear; it becomes visible, explainable and correctable instead.