What signals do AI models analyze (content, engagement, headers, volume, etc.)?

You craft a great email, hit send, and it lands in spam anyway. What happened? Somewhere between your outbox and the inbox, an AI model ran your email through a stack of signals and decided it didn't belong there. Understanding what those signals actually are is the first step to fixing it.

Here's a breakdown of the main categories filters evaluate.

Content signals cover everything inside the email itself. Word patterns, phrase combinations, HTML structure, image-to-text ratio, number of links, where those links point, and whether the link domains match your sending domain. A well-written subject line paired with clean HTML and a single clear link looks very different to a filter than a wall of images with five redirect links buried inside.

Header signals are about how the email was built and routed. Filters check whether your SPF, DKIM, and DMARC all passed, whether the header formatting looks normal, and whether the routing path through mail servers makes sense. A header that looks hand-assembled or shows unusual relay hops raises flags before anyone reads a single word.

Engagement signals are probably the most weighted category at modern mailbox providers like Gmail and Outlook. Do your recipients open your emails? Do they delete them immediately without opening? Do they click spam? Do they scroll through and click a link? Over time, these behaviors build a reputation signal specific to your domain, your IP, and even individual recipient relationships.

Volume and velocity signals track how your sending behaves over time. A sudden spike from 500 emails a day to 50,000 looks suspicious even if every individual email is clean. Consistent sending patterns, steady growth, and predictable schedules all read as healthier than erratic bursts.

No single signal flips the verdict on its own. The model combines everything, with weights it learned from millions of classified examples, and scores the whole picture. That's actually useful news for senders. It means you don't need to be perfect on every dimension. But it also means you can't optimize one thing and ignore everything else. Ignoring engagement while having perfect authentication still gets you filtered.

The practical takeaway is this: send to people who want your email, keep your authentication clean, and let your volume grow steadily. Those three things move more of these signals in your favor than any content trick ever will.

But if you want to check whether your authentication signals are passing cleanly, try our free Email Header Analyzer. It reads the full header and shows you exactly what a filter sees when your email arrives.