How inbox filters actually sort cold email

If you sell to businesses, the Promotions tab is probably not your problem. The tab only exists for consumer Gmail accounts and for Google Workspace accounts where an admin has switched it on. Most Workspace customers run the default policy, a single unified inbox with no tabs at all. If your list is full of people at corporate domains, a large share of them never see a Promotions tab.

That matters, because "how do I get out of Promotions" is the wrong question. It sends people down the wrong path. They strip images out, remove the tracking pixel, rewrite the subject line, and nothing changes. The reply rate stays flat. They conclude email is broken.

Email is not broken. Something else is happening to the message.

When someone says their cold email did not get seen, one of three things actually happened. The message was delivered and ignored. It arrived in the inbox. The recipient looked at the first line, recognised a template, and archived it. No filter was involved. This is the most common case by a wide margin, and no technical work will fix it.

Or it was delivered to spam. The provider accepted the message but decided the recipient probably did not want it. This is a reputation and engagement problem.

Or it was rejected outright. The receiving server refused it. Since November 2025, Microsoft has been returning permanent 550 errors for non-compliant senders rather than quietly routing the mail to junk. The message does not reach the junk folder. It does not reach anything.

These have different causes and different fixes. Lumping them together is why so much deliverability advice is useless.

Before any filter thinks about your content, it checks whether your domain is authorised to send the message. SPF says which servers are allowed to send on your behalf. DKIM signs the message so it can be verified as unaltered. DMARC ties the two together and tells receivers what to do when a check fails.

This is not optional any more. Google, Yahoo and Microsoft all require it. Any domain sending 5,000 or more messages a day to Gmail addresses is permanently classified as a bulk sender, and that classification does not lapse if volume drops later. Most B2B cold campaigns will not hit that threshold, because the recipients are on corporate domains. The authentication requirements still apply.

Getting this right does not earn you anything. It only stops you being filtered for a reason that has nothing to do with your message. It is the cost of entry, and it takes an afternoon.

Once you are authenticated, providers score your domain on how recipients behave when your mail arrives. The number that matters most is the spam complaint rate. Gmail's hard ceiling is 0.3%. Cross it and you are in trouble. But 0.3% is the cliff edge, not the target. The working threshold most senders should hold is 0.1%, roughly one complaint per thousand messages.

Sit with that. One person in a thousand hitting the report button is enough to damage your sending domain. Not unsubscribing. Not ignoring you. Reporting you.

Bounce rate feeds the same score. If a meaningful share of your sends hit addresses that do not exist, you have told the provider you did not verify your list, and lists that are not verified are usually scraped. Volume patterns feed it too. A domain that has sent nothing for months and then sends four hundred near-identical messages in an afternoon looks exactly like what it looks like. Microsoft weighs IP reputation heavily alongside domain reputation. If you send through a shared IP that someone else has burned, your mail suffers even when your own authentication and behaviour are clean.

Authentication and reputation get you into the room. Engagement is what keeps you there.

Filters watch what recipients do. Opens are a weak signal and getting weaker. Replies are the strongest signal available. A message that gets answered is, by definition, wanted. Forwards, moving a message out of a spam folder, adding a sender to contacts, all tell the provider the same thing.

Here is the part most deliverability guides skirt around. Every layer above can be bought or configured. You can buy domains, warm them, configure DNS, rotate inboxes, manufacture a clean technical profile. Engagement cannot be configured. It is generated by the recipient, and the recipient only generates it if the message was worth their attention.

The filters are not really scoring your infrastructure. They are scoring whether people want your email. The infrastructure is just how they check you are not lying about who sent it.

The chain runs in one direction and it is unforgiving. A templated message gets deleted. Deletions produce no positive engagement. A few recipients report it rather than ignoring it. Complaints push you toward the threshold. Reputation drops. The next campaign gets filtered harder, so fewer people see it, so engagement drops further. The usual response is to send more volume to compensate, which accelerates every part of the cycle.

Most people diagnose this as a deliverability problem and go looking for a technical fix. It is not a technical problem. The infrastructure is doing its job. It has correctly identified that the recipients do not want the mail.

Do the technical work first, because it is cheap and it removes a variable. Set up SPF, DKIM and DMARC on your sending domain. Use a subdomain for outreach so problems do not touch your main domain. Warm new domains gradually. Verify every address before you send. Monitor your spam complaint rate through Postmaster Tools. Treat 0.1% as the line. Make it genuinely easy to opt out. An unsubscribe costs you nothing. A spam report costs you the domain.

Then accept that none of it addresses the real constraint.

If replies are the strongest signal a filter can read, then the highest-value deliverability work is not configuration. It is making the message worth replying to. That is not a slogan, it is arithmetic. A thousand templated messages producing a 1% reply rate generate ten positive signals and a trail of complaints that degrades every future send. A hundred researched messages producing a 15% reply rate generate fifteen positive signals, almost no complaints, and a domain that gets treated better next month than it did this month.

Lower volume, higher relevance. The filters were built to reward exactly that, and they have been getting better at spotting the alternative every year since 2024.

This is why we research every prospect before we write anything, and why the email we send is two sentences with a link to the work rather than a pitch. Not because it is a clever tactic. Because it is the only version of cold outreach that the infrastructure is designed to let through.