AI is strong at volume and weak at judgment. Where it reliably earns its place in an email program, where it quietly costs you, and how agents move the job from writing emails to approving them.
Cyril Nie
Author
AI won’t replace you next Tuesday, and it isn’t just fancy autocomplete either. The truth is in the middle. Here’s where it genuinely helps — and where it doesn’t.
Subject lines. Generate 50 options in seconds, pick the best 2–3, A/B test them. Consistently lifts opens 5–15% for five minutes of work. The easiest AI win there is.
First drafts. The blank page is the silent killer. An AI draft gets you 70% of the way; you edit in your voice. 15 minutes instead of 45.
Send times. AI predicts when each individual person reads and delivers right then — typically 10–25% more opens. No human can do this for 50,000 people.
Predictions. Churn risk, next order date, lifetime value per customer. A well-timed email to a high-value customer showing early leaving signals is worth more than a thousand optimized subject lines.
Brand voice — generic AI copy has a sameness readers can feel, even if they can’t name it. Strategy — AI can polish a subject line, but it can’t decide whether this week needs a promo or a helpful guide. And creative leaps — no algorithm would have suggested Patagonia’s “Don’t Buy This Jacket.” Every AI-written email should be read, edited, and approved by a human before it sends. No exceptions.
As of early 2026, three platforms are visibly in agent territory: Klaviyo’s K:AI Marketing Agent, ActiveCampaign’s Active Intelligence (34+ AI capabilities, including a Claude MCP connector), and Intuit Assist inside Mailchimp.
The newest shift: instead of AI helping you click buttons faster, you describe the goal and an agent builds the thing — the flow, the segment, the copy — and you review and approve. The job changes from “build campaigns” to “approve recommendations.” The marketers who win will be the best editors and strategists, not the best button-clickers.
Producing volume where a human then selects: subject line variants, plain-text versions, translations, product description drafts, and summarizing behavioral data into segment ideas. It compresses the work that used to make good practice too slow to do.
On judgment. It does not know which customer pain matters, cannot verify a claim, and drifts to a generic register that reads as machine-written. Used unsupervised across a whole program, it produces email that is competent, interchangeable and ignored.
The work moves up a level: instead of writing each email, you define triggers, guardrails and success criteria, and the agent assembles and adjusts sends within them. That makes clean event data and explicit rules more valuable than copywriting speed.
Discover more insights on similar topics
The complete 2026 guide to email marketing: 16 chapters on list growth, segmentation, automation, deliverability, testing, attribution and AI. Built from 908 sources and 4,798 insights.
Email returns about $36 for every $1 spent — roughly 30 times the next best channel. Here is why it keeps winning, and the six parts every working program is built from.
Apple inflates open rates and Gmail’s scanners click your links. Here are the email metrics that survive scrutiny, the thresholds worth holding yourself to, and why your own last 90 days beat any industry average.