You don’t have to ban AI from your email workflow, but you need to decide which tasks are worth handing off and which tasks need human review, then build those checkpoints into the process.

1. Start every draft with a detailed brief

Generic prompts produce generic emails. Give the model your audience segment, the offer’s exact terms, the call to action, and a couple of past emails that performed well so it can match your voice.

Treat the output as a first draft for an editor. We recommend rewriting the opening line and the call to action yourself, since those are where readers decide whether a message sounds like a person.

2. Fact-check every claim and deadline

Build a pre-send checklist that compares prices, discount percentages, expiry dates, and product names against your source of truth. Pay particular attention to subject lines and preview text, because those are what inbox summaries and legal complaints tend to latch onto.

Anything with urgency language deserves a second human reviewer. If there’s any chance a promotion gets extended, rewrite the subject line so it stays true either way.

3. Put approval gates on automated sequences

Let the platform draft a welcome series or a win-back flow, but review every email and trigger before it goes live. Treat any AI-built sequence like a new hire’s first campaign, where nothing ships until someone senior has read it. After launch, spot-check live sends each week rather than assuming the automation behaves.

Set a cap on how often any single contact can be emailed across all your automations. Overlapping AI-built flows are an easy way to creep toward Google’s spam-rate ceiling without noticing.

4. Question what the analytics tell you

Before acting on an AI recommendation, ask what data it’s based on. A send-time suggestion built on a few weeks of activity from a small list is a guess wearing a lab coat.

Use AI to spot patterns, then confirm them with a controlled A/B test before rebuilding your program around them. The model can find the correlation, but a human still has to decide whether it means anything.

5. Protect customer data and keep personal emails personal

Check how a vendor handles your data before pasting customer details into any AI tool. Look for a written commitment that third-party AI providers can’t train their models on your prompts or subscriber data, plus a clear answer on where that data is processed.

For relationship emails, such as a thank-you to a key account, AI shouldn’t do anything more than light editing. You can use AI to write the email for you, but making the reader believe that you meant it is still your job.