AI and the Future of Email

If Chapter 13 was absent from v3, this chapter couldn’t have existed at all. The AI capabilities available to email marketers in early 2026 are fundamentally different from what existed even 18 months ago. Not different in the incremental-improvement sense, but different in the ‘this changes the workflow’ sense.

I’ll be direct about where I think AI is genuinely useful, where it’s overhyped, and what’s coming next. The AI conversation in marketing is plagued by extremes: either AI is going to replace every marketer next Tuesday, or it’s just a fancy autocomplete that adds no real value. The truth, as always, sits in the middle, and the specifics matter more than the generalities.

Where AI Excels Right Now

Subject line generation is the most immediate win. AI can generate 50 variations of a subject line in seconds. Your job is to pick the two or three best ones and A/B test them. What used to take 20 minutes of brainstorming now takes 30 seconds of generation and two minutes of curation. The result is more testing, which means more data, which means better subject lines over time.

I’ve found that AI-generated subject lines perform comparably to human-written ones about 60% of the time, and outperform them about 20% of the time. The remaining 20% where humans win tends to be cases requiring cultural context, current events awareness, or brand-specific humour that the AI doesn’t quite nail. But 80% comparability at 10% of the time investment is an excellent tradeoff.

Send time optimisation has gotten remarkably good. Machine learning models now predict per-subscriber optimal send times based on historical engagement patterns. Most major ESPs have this built in. Seventh Sense takes it further with a dedicated product that analyses engagement windows for each contact individually. The improvement is typically 10 to 25% in open rates compared to batch-and-blast scheduling. It’s one of those features where the AI does something humans literally cannot do at scale: optimise timing for each individual subscriber across a list of 50,000.

Segmentation is where AI identifies patterns that humans miss. Engagement clusters, churn predictors, purchase propensity scores. Klaviyo’s predictive analytics can estimate customer lifetime value, churn risk, and expected next order date for each subscriber. HubSpot can score leads based on hundreds of behavioural signals. This data feeds into smarter segmentation, which feeds into better targeting, which feeds into better results. It’s a virtuous cycle that gets more powerful as your data grows.

Content personalisation at scale means dynamic content blocks powered by AI recommendations. Product recommendations based on browsing and purchase behaviour. Content blocks that change based on predicted interests. Subject lines that vary by segment. The goal is making each email feel individually crafted without actually writing thousands of variations. Netflix’s recommendation emails are a good example: every user gets a different email with different show recommendations, powered entirely by AI analysis of viewing patterns.

First draft generation solves the blank page problem. Staring at an empty email editor is the silent productivity killer of email marketing. AI generates a working first draft in seconds. It won’t be perfect. It shouldn’t be published as-is. But it gives you something to react to, edit, and improve, and that’s dramatically faster than starting from nothing.

Analytics and pattern recognition is quietly becoming one of AI’s most valuable applications. AI can identify anomalies in campaign performance (this email’s click rate is 40% below your average for this segment), detect trends across campaigns (subject lines with numbers have performed 15% better for you over the last 6 months), and flag potential issues before they become problems (your engagement with Yahoo recipients has dropped 20% this month).

Where AI Falls Short

Brand voice consistency is the biggest gap, and I don’t see it closing soon. Generic AI copy is detectable. Your subscribers may not consciously identify it as AI-generated, but they’ll feel the difference. There’s a sameness to AI-generated marketing copy. The phrasing is too smooth, the transitions too clean, the personality too even. The warmth, the quirks, the specific way your brand talks, that’s extraordinarily hard for AI to replicate without extensive fine-tuning. And even with fine-tuning, the output needs heavy human editing.

I tested this by sending two versions of a welcome email to a split audience. The AI-drafted version performed identically on open rate and click rate. But qualitative feedback from customer surveys showed that recipients found the human-written version ‘warmer’ and ‘more authentic’. Over a single email, the difference is marginal. Over a 12-email welcome series, the accumulated effect of generic voice erodes brand perception.

Strategic thinking remains firmly human territory. AI can optimise a subject line, but it can’t decide whether you should be sending a promotional email or a value-add piece this week. It can personalise content, but it can’t determine the right balance between education and sales for your audience at this stage of your company’s growth. Strategy requires understanding context, goals, brand positioning, competitive dynamics, and customer relationships in a way that current AI simply doesn’t.

Emotional nuance matters more than marketers sometimes admit. The re-engagement email for a subscriber who hasn’t opened in 90 days needs a different emotional register than the win-back for a customer whose subscription lapsed. Empathy in customer service replies, sensitivity in handling complaints, the right tone for a product recall, these require human judgment that AI approximates but doesn’t truly possess.

Creative breakthroughs don’t come from AI. AI optimises within existing patterns. It’s exceptional at taking what works and generating variations. But Duolingo’s heartbroken owl, Casper’s ‘Come back to bed’, Patagonia’s ‘Don’t Buy This Jacket’, these creative leaps came from humans who understood their brand deeply enough to take risks that no optimisation algorithm would recommend. AI would never suggest telling customers not to buy your product. A human who deeply understands Patagonia’s brand would.

The Human-AI Workflow

The best results come from collaboration, not full automation. Here’s the workflow I’d recommend, based on what I’ve seen working across dozens of email programmes:

Start by briefing the AI with context. Brand voice guidelines, audience information, campaign goals, product details, examples of past winning emails. The quality of AI-generated email copy is directly proportional to the quality and specificity of the input. A prompt that says ‘Write an email promoting our sale’ will produce generic output. A prompt that includes your brand voice document, three examples of emails that performed well, the specific products on sale, the discount structure, and the audience segment will produce something much closer to usable.

Generate the first draft using AI. Let it handle the structure, the initial copy, the subject line options. Don’t judge the output too harshly at this stage. You’re not looking for a finished email. You’re looking for raw material to work with.

Edit heavily. This is where your brand voice lives. Change the phrasing to match how your brand actually talks. Add the specific details, anecdotes, or personality that make your emails yours. Remove anything that sounds generic or formulaic. A good editor can turn a mediocre AI draft into a strong email in 15 minutes. Without the AI draft, that same email might take 45 minutes to write from scratch.

Test against human-written versions. Run A/B tests with AI-assisted copy versus purely human-written copy. You’ll often find that the AI-assisted version performs comparably or better on metrics like open rate and click rate, while the human-written version scores higher on brand perception and qualitative feedback. Find the balance that works for your audience.

Iterate over time. Feed the results back into your AI workflow. The winning emails become examples for future prompts. The losing ones become guardrails. Your AI-assisted output should improve with every cycle as you refine your prompts and develop a better sense of what the AI does well and where it needs more guidance.

AI Features by Platform

Every major ESP now offers AI features, though the depth and utility vary significantly.

Platform	AI Feature	What It Does
Klaviyo	AI Subject Line Generator	Generates and scores subject line options
Klaviyo	Predictive Analytics	Predicted CLV, churn risk, next order date
Mailchimp	Creative Assistant	Generates designs from brand guidelines
HubSpot	AI Content Writer	Email copy generation from prompts
Braze	Sage AI	Copy generation, channel optimisation
ActiveCampaign	AI Content Generator	Subject lines and body copy
Bento	MCP Integration	API-driven automation via MCP server
Seventh Sense	AI Send Time	Per-contact delivery optimisation
Phrasee	AI Copywriting	Enterprise subject line and copy optimisation

Klaviyo’s predictive analytics deserves special attention because it’s not generating content, it’s generating intelligence. Knowing which customers are likely to churn before they actually do, or which customers have the highest predicted lifetime value, changes how you segment, what you send, and when you send it. That’s more valuable than any copy generation feature. A well-timed retention email to a high-CLV customer who shows early churn signals is worth more than a thousand AI-optimised subject lines.

Phrasee operates at the enterprise level, working with brands like eBay, Domino’s, and Virgin Atlantic on AI-powered copy optimisation. Their approach is different from general-purpose AI: they train models specifically on your brand’s historical email data and your audience’s engagement patterns. The result is AI-generated copy that’s calibrated to your specific audience rather than a general model’s understanding of ‘good marketing copy’.

MCP (Model Context Protocol) and Email

This is new territory, and I think it’s the most important development in email marketing tooling since marketing automation itself.

Anthropic’s Model Context Protocol (MCP) enables AI models to directly interact with external tools and data sources through a standardised interface. For email marketing, this means AI can read your campaign data, analyse performance, and take actions within your email platform, all through natural language conversation. Instead of clicking through dashboards, you ask questions. Instead of building segments through a UI, you describe what you want.

Mailjet has an open-source MCP server for email marketing that provides read-only access for AI models like Claude and GPT. This lets you ask questions about your email performance in plain English and get answers drawn from your actual data. ‘What was my open rate trend for the last 12 weeks?’ gets you a direct answer with the data, not a report you need to interpret.

Bento offers an MCP server integration that lets AI models interact with your email data programmatically — querying campaign performance, managing contacts, and triggering sends through a standardised API interface. For developer-heavy teams already working with AI coding tools, this kind of integration removes the need to context-switch between a conversation and a dashboard.

These are early implementations, and the space is moving fast. The interface for creating and managing email campaigns is gradually shifting from purely graphical flow builders toward more programmatic and conversational approaches. We’re not there yet — but the direction is clear.

The implications are significant. A solo founder who couldn’t justify hiring an email marketing specialist can now describe their goals to an AI agent and get a professionally structured email programme. An experienced marketer can move faster by describing complex flows in natural language rather than clicking through builder interfaces. An agency can serve more clients by using AI agents to handle the routine build work while humans focus on strategy and creative direction.

The AI-Native ESP Vision

The traditional ESP workflow looks like this: a human creates a campaign, selects a segment, writes the copy, designs the template, schedules the send, and analyses the results. Every step requires human initiation and execution.

The AI-native ESP workflow inverts this. AI analyses customer data and identifies opportunities (‘You have 2,400 customers who purchased once 45 days ago but haven’t returned. Here’s a suggested win-back sequence.’). It drafts the content. It optimises timing and targeting. The human reviews, adjusts, and approves.

The shift is from ‘build campaigns’ to ‘approve recommendations.’

Early examples of this shift are already visible. Klaviyo’s predictive analytics identifies at-risk customers before humans would notice the pattern. Braze’s Sage AI generates copy and optimises channel selection. MCP integrations from platforms like Bento and Mailjet let AI models query email data directly.

The key distinction is this: AI handles optimisation (what content, when to send, who to target), while humans handle strategy (why we’re sending, brand voice guardrails, ethical boundaries, overall programme direction). This division of labour plays to each side’s strengths. AI is better at processing data and finding patterns. Humans are better at judgment, creativity, and understanding context.

Practical AI Integration Today

Here’s what I’d actually recommend implementing right now, ordered by impact and ease of adoption:

1. Use AI for subject line generation. Generate 20 to 50 options, pick the best two or three, and A/B test them. This takes five minutes and consistently improves open rates by 5 to 15%. It’s the lowest-effort, highest-impact AI application in email marketing today.

2. Use AI for first drafts of email sequences. Especially for standard flows like welcome series, cart abandonment, and post-purchase. Edit heavily for brand voice, but let AI handle the structural heavy lifting. A good prompt with brand voice examples will get you 70% of the way there.

3. Use predictive analytics for churn risk and customer lifetime value. If your ESP offers it (Klaviyo, HubSpot), turn it on. Segment by predicted churn risk and send targeted retention campaigns to high-risk customers before they leave. This is pure upside with minimal effort.

4. Use AI-powered send time optimisation. Most major ESPs include this. Enable it. The per-subscriber timing adjustment is something humans cannot replicate manually, and the improvement is measurable and consistent.

5. Use AI for customer segmentation. Let AI identify engagement clusters and behaviour patterns you wouldn’t have thought to look for. Then build campaigns targeted to those AI-identified segments.

And here’s what not to do:

6. Don’t use AI as a replacement for understanding your customers. AI analyses data. Understanding comes from reading support tickets, talking to customers, watching user sessions, and building empathy for the people on your list. Data tells you what people do. Understanding tells you why.

7. Don’t use AI-generated copy without human review and editing. Every AI-generated email should be read, edited, and approved by a human before sending. No exceptions. Not even for automated flows. Set it up, review it, then let it run.

8. Don’t rely on AI for strategic decisions about your email programme direction. Should you send more or fewer emails? Should you shift from promotional to educational content? Should you launch a newsletter? These are strategic questions that require human judgment about your brand, your market, and your goals.

What’s Coming (2026-2028)

I’m going to make predictions, which means some of these will be wrong. But the direction is clear, even if the timeline is uncertain.

AI agents that build and manage email automations from natural language instructions. There are early experiments in this space, but nobody has truly cracked it yet. By 2028, I’d expect every major ESP to offer conversational automation building. The flow builder interface won’t disappear entirely, but it’ll become the ‘power user’ tool rather than the primary interface. Just as most people use a visual website builder rather than hand-coding HTML, most email marketers will describe their automations in natural language rather than building them visually.

Real-time content personalisation powered by large language models. Each recipient gets genuinely unique copy, not just different product recommendations inserted into the same template. The entire email, from subject line to body to CTA, is generated for that specific person based on their behaviour, preferences, and stage in the customer journey. This is computationally expensive today but will become practical as inference costs continue to drop.

Predictive deliverability monitoring. AI that flags potential deliverability issues before they impact inbox placement. ‘Your engagement rate with Gmail recipients has dropped 12% over the last week. Here’s the likely cause and recommended action.’ This moves deliverability management from reactive (fixing problems after they occur) to proactive (preventing problems before they happen).

Cross-channel AI orchestration. Email, SMS, push notifications, and in-app messaging coordinated by AI that determines the optimal channel, timing, and content for each customer interaction. The marketer sets the goal and the guardrails. The AI handles the execution across channels.

AI-powered compliance checking. Automatic verification that every email meets GDPR, CAN-SPAM, CASL, and other regulatory requirements before send. Checking consent records, validating unsubscribe mechanisms, scanning content for compliance issues. This removes one of the most anxiety-inducing aspects of email marketing, especially for companies operating across multiple jurisdictions.

---