Your Subscribers Aren't Reading Your Emails.

Gmail's AI reduces time spent on "unimportant mail" by 13%. Apple Intelligence rewrites your preview text with AI-generated summaries. Microsoft Copilot assigns High/Normal/Low priority scores before your subscriber even opens their inbox.

Every email you send now passes through an AI filter that decides if it's worth showing to a human.

But many brands have no idea how these AI systems make filter their emails. They write for human readers while AI gatekeepers sort their messages into promotional purgatory.

The new inbox challenge is writing emails for both AI systems and human readers.

The Math Behind the Gatekeepers

LLMs don't read emails the way people do. They convert your text into mathematical vectors and compare these against patterns they've learned mean "important."

The core architecture seems to come from a 2017 Google Brain paper called "Attention Is All You Need." It introduced something called self-attention, which lets the model relate any two parts of your email regardless of how far apart they are.

When the AI reads "Can you review the proposal by Friday?", the self-attention mechanism calculates which words matter most. It assigns higher weight to "review," "proposal," and "Friday" because these signal action and urgency. Words like "the" and "by" get lower weight.

This happens through something called Scaled Dot-Product Attention. The formula computes how much each part of your email should "attend to" every other part. Modern systems run 8 parallel attention heads that look at your content from different angles simultaneously.

Context windows determine how much information the AI can hold in working memory. GPT-4 handles 128,000 tokens (about 300 pages). Claude processes 200,000 tokens standard. Gemini 1.5 Pro can digest 2 million tokens—enough for your entire email thread history, sender relationship patterns, and contextual signals all at once.

The Weight Hierarchy That Decides Your Fate

Gmail's Priority Inbox uses a prediction formula that achieves 80% accuracy in determining which emails users will act on. It trains on hundreds of features across four categories: who sent it, what it says, how the thread behaves, and how you organize your mail.

Microsoft's Focused Inbox adds organizational hierarchy to the mix. Emails from your direct manager get highest weight, followed by frequent correspondents, then organizational position.

Both systems weight corrections heavily. When you manually mark an email as important or move it between tabs, that signal carries more confidence than passive behaviors like opens.

The weighting works in tiers.

Tactical Strategy #1: Exploit High-Weight Signals—Actionable Intent and Urgency

Many AI systems prioritize "unresolved loops." Emails that require action get flagged as high-priority tasks.

Use Actionable Verbs in Your First 20 Words.

From our internal testing with AI summaries, it seems like the words at the beginning of the email are weighted much more heavily than the end. So learning to front-load everything and put your call to action in the first 20 words leads the AI to prioritize that email and CTA.

Instead of: "We wanted to share our latest insights on email deliverability..."

Write: "This will help you identify the three email deliverability fixes that will get you to 95% inbox placement."

Instead of: "Our new product update helps you manage your campaigns more effectively..."

Write: "Click here to enable automated list cleaning before your next send."

Instead of: "This week we're discussing strategies for improving engagement..."

Write: "Answer this one question about your current open rates, and we'll show you exactly where you're losing subscribers."

The AI detects interrogatives and imperatives as tasks. Questions and commands signal that the recipient needs to do something.

Embed Specific Time-Sensitive Language

The key here is to be specific. In our internal testing, we found AI summaries pulled in better and more relevent information using specific time blocks - “11:59pm” over generic time blocks - “midnight”.

Instead of: "Sale ending soon"

Write: "Your 40% discount expires at 11:59 PM EST tonight—December 28th"

Instead of: "Limited spots available"

Write: "We have 12 seats remaining for the January 15th cohort. Registration closes January 3rd at midnight."

Instead of: "Act fast"

Write: "This is your notification of your invite to our training on Friday, January 5th, to learn about how to get more emails to the inbox. "

The AI extracts time entities (dates, times, deadlines) and context (expiration, deadlines). It may pin the email or include it in the user's "Today" view because it has high recency/expiry weight.

Create Pattern Breaks Through Inconsistent Action Items

AI systems detect patterns and flag deviations as high-priority events.

If you typically send weekly newsletters with educational content, occasionally break the pattern with something that signals urgency or change:

Instead of: "Weekly Newsletter #42: Five Tips for Better Email Copy"

Write: "Important change to how Gmail processes your emails starting January 2025"

If your typical email discusses product updates, occasionally shift to something that creates semantic inconsistency:

Instead of: [NEW] feature announcement

Write: "We noticed a 40% drop in your deliverability last week. Here's what happened."

The AI detects scope creep and semantic shifts. When a conversation about features suddenly mentions problems, refunds, or legal issues, the system ranks it higher because it perceives risk. CHANGE UP YOUR EMAIL STRATEGY!

Tactical Strategy #2: Build Entity Relationship Weight Through Reply Patterns

The AI weights emails based on sender-recipient relationship history.

Trigger Reply Behavior in Your Confirmation and Welcome Emails

In your welcome sequence, ask a simple question that encourages a one-sentence response:

"Quick question: What's the biggest email challenge you're facing right now? Just hit reply and let me know."

"We want to make sure these emails are valuable to you. Reply with one word: 'deliverability,' 'copy,' or 'automation' depending on what you want to learn about first."

"Before I send you the full guide, I need to know: Are you on Gmail, Outlook, or another provider?"

When a subscriber replies to you at least once, the AI's interaction history weight for your email address spikes. You're no longer "external marketing." You're a "frequent correspondent." And the more often you can get a reply - the more “weight” the AI system will give you to as a sender.

Use Name-Specific Personalization Beyond the First Name

Instead of: "Hi {FirstName},"

Write: "Tyler, I pulled your account data and noticed you haven't enabled authentication yet. That's likely why 30% of your emails are landing in spam."

Instead of: Generic content that's identical for everyone

Write: Reference their specific behavior, purchase history, or engagement patterns in the email body.

When your name appears in the body of an email (not just the greeting), the AI "boosts" that content because it implies you specifically are needed. You can often insert information like this using custom fields and merging that info into the email body copy. Similar to merging in their first name, just on a larger scale.

Cold email platforms do this using something called spintax.

Tactical Strategy #3: Increase Semantic Novelty Through Information Density

The AI looks for new information versus repeated information.

Use Structured Content With Specific Data Points

The key to remember here is that AI loves specificity. They like specific “claims” they can reference for context.

Instead of: Long paragraphs of general advice

Write: "Gmail now requires DMARC authentication. 46.2% of senders still don't have it enabled. Here's the three-line DNS record you need to add today."

Instead of: Abstract concepts

Write: "Apple Intelligence generates summaries from your first paragraph. The preview text you spent 20 minutes crafting? Your subscribers never see it. The AI rewrites it based on your opening sentences."

Instead of: Repeated information the subscriber already knows

Write: New research, recent data, or fresh examples they haven't heard before.

Emails with specific data points, concrete examples, and bulleted lists (when appropriate) get weighted higher for "digest" purposes than conversational filler.

Create Pivot Points That Signal New Decisions

In email sequences, the AI down-ranks messages that just repeat what's already been said. It prioritizes content that represents a decision point or change in direction.

Instead of: Email #3 that reiterates the same benefits from Email #1 and #2

Write: Email #3 that introduces a new angle, addresses a different objection, or reveals information you deliberately held back.

Instead of: "As I mentioned in my last email..."

Write: "You asked about implementation timelines. Here's what that actually looks like..."

The AI detects pivot points and semantic novelty. Content that advances the conversation gets higher priority than content that echoes previous messages.

Leverage the Unresolved Loop Dynamic

If you send an email with a question, the AI marks that thread as "Open." It's waiting for a reply to determine if the loop is "Closed."

Structure your emails to create open loops that benefit from staying top-of-mind:

"I sent you the deliverability audit last week. Did you get a chance to review the authentication section? If you have questions about implementing DMARC, just reply to this email."

"You downloaded the copy framework guide on Tuesday. I'm curious… which framework did you try first?"

These questions keep the conversation thread marked as unresolved, which maintains visibility in AI-prioritized inboxes.

When AI Fact-Checks Your Claims in Real-Time

Gmail and Outlook now use AI-controlled filtering to evaluate content accuracy.

If the AI detects misleading claims, false statistics, or misquoted sources, it may restrict your email's visibility or route it to spam. This is sometimes called shadowbanning.

You won't get a bounce message. Your deliverability dashboard might still show 95% inbox placement. But your subscribers never see the email because the AI filtered it before it reached human eyes.

The system learns from spam reports, which are increasingly triggered when subject lines don't match email content.

Traditional "clickbait" tactics - exaggerated claims, false urgency, misleading statistics—now actively harm your sender reputation with AI systems that can verify facts in real-time.

Clean, accurate, substantiated claims protect your deliverability. Verifiable data and honest language signal trustworthiness to AI filters.

What's Actually Happening in Your Email Program Right Now

Most brands are writing for humans while AI systems determine if humans ever see their messages.

Your authentication might be set up. Your copy might convert when it's read. But if the AI is down-ranking your emails based on signals you don't understand, your open rates will keep declining no matter how many subject line tests you run.

Now is a good time for an email audit. Set 2026 up for email success. We look at your general email settings, segmentation strategy, automations, customer lifecycle flows, and technical infrastructure to show you exactly where AI systems are filtering your messages.

During our audit, we identify which signals could be hurting your deliverability and which opportunities you're missing.

You’'ll get a comprehensive report that maps every point where your email program could be improved and what’s crushing it for you already.

Reply to this email with "Audit" and we'll send you the intake form. Normally we charge $1,500 for a full audit like this, but I’m working on using AI to help customize and spot trends bester and better. So for the first 3 brands that request an audit, we’ll do it for no charge.

Reply “Audit,” and we’ll send you the next steps.

Tyler