AI in email marketing: better emails, same rules

Where AI helps in email marketing

The rules in Germany and the EU

• Word the consent precisely: what will the address be used for? If you recommend products or measure opens and clicks, say so.
• Keep the proof: time, IP address and double opt-in confirmation. Most newsletter tools do this automatically.
• Bought lists: practically unusable in Germany, because consent for you as the sender is missing.
• AI segments: only with data you have a legal basis for. Don’t let AI guess sensitive traits.

Outside the EU the rules differ; in the US, for example, the CAN-SPAM Act doesn’t require prior consent but does require an opt-out and a physical address. If you email people in Germany, the German rules apply.

Deliverability: the foundation before any AI

The best AI subject line is useless if the email lands in spam. Gmail has applied stricter sender rules since 2024; many other providers check similarly.

Source: Google Workspace Admin Help: email sender guidelines.

• One sender name readers recognise, and a sending domain that matches your website.
• No bare URLs and no domain in the subject line; use link text instead.
• A plain-text version matching the HTML email.
• Inactive contacts: ask after a while, then remove them.

Subject lines with AI

Subject lines are the classic AI use: ten variants in seconds, sorted by pattern. The skill is in choosing.

• Honest: the subject promises what the email delivers. Clickbait costs opens over time.
• Short: on a phone, it often cuts off after a few words.
• Use the preheader: the first line after the subject is your second chance.
• No spam signals: ALL CAPS, lots of exclamation marks, piles of “free”.
• Test with enough recipients: with small lists, a difference in open rate is often chance.

Also: open rates have become less precise because some email apps preload content automatically. Clicks and replies are the more reliable signals.

Segments and send time

AI features in newsletter tools suggest segments or pick a send time per recipient. That can help when there’s enough data. With a small list, a simple plan is often better.

My rule of thumb: AI widens the cracks in the foundation. If your list is messy (old addresses, unclear consent, no sign-up source), AI segmentation only makes the problem bigger. Clean the list first, then automate.

Planning a welcome series with AI

AI helps outline the series and draft the emails. You decide each email’s goal, when to sell and which recommendation you genuinely stand behind. One email, one goal, one link destination.

• Label affiliate links: “Ad” or “affiliate link” right next to the link.
• No income promises: not in AI drafts either.
• No invented experience: “I use this daily” only if it’s true.

Tools

Prices exclude tax. For the EU: sign a data processing agreement and check where data is processed.

Checklist before sending

• Every recipient has consented (double opt-in) or clearly falls under the existing-customer exception
• Consent wording covers content, recommendations and tracking
• SPF, DKIM, DMARC set up, consistent sender name
• Honest subject line, preheader set
• One email, one goal; affiliate links labelled as ads
• Every number and claim checked
• Imprint and unsubscribe link in the email
• Test sent to several inboxes (e.g. Gmail, Outlook, a regional provider), checked on mobile
• A person has signed off the email

Metrics

AI features in newsletter tools

Almost every newsletter tool now advertises AI. The features are similar, and how useful they are depends heavily on the size of your list.

Check which data the features use and where it’s processed. And check whether the AI feature is in the free plan: at MailerLite, for example, the writing assistant is only included in the paid plan, per its pricing page.

My rule of thumb: a person signs off what goes out. That applies doubly to email, because you can’t recall a message once it’s sent.

Reusing content for emails

The biggest time saving from AI in email marketing often isn’t writing new emails but turning existing content into them: blog posts, guides, news, webinars, call notes.

1. Pick a source: an article or page with a clear core message.
2. Core message for the audience: AI summarises what’s new or useful about it for your readers.
3. Email format: an opener, three points, one link to the full piece.
4. Your own take: one sentence from you on why it matters. That’s what separates your newsletter from an RSS feed.
5. Check and sign off: facts, links, labelling.

That way a good article becomes an email in minutes that brings readers back to your site, rather than telling the whole story in the email. Make sure each email still makes sense on its own and isn’t just a list of links.

A/B tests that tell you something

Many tools offer A/B tests for subject lines or content. AI makes it easy to write variants, but not easier to learn from small numbers.

• One variable: test only the subject or only the button, not both.
• Enough recipients: with a few hundred recipients, differences of a few percentage points are often chance.
• Clicks over opens: open rates have become less precise; clicks tell you more.
• Record it: hypothesis, variants, result, decision in a simple spreadsheet.
• Big questions first: with small lists, test content and offer rather than punctuation.

How often to email? Frequency and unsubscribes

There’s no single right frequency. What counts is what you promised at sign-up and whether each email has a reason to be read.

Say at sign-up how often emails will come, and stick to it. An honest expectation lowers unsubscribes and spam complaints more than any AI optimisation.

12 prompts for email marketing

Write ten subject lines (max 45 characters) and matching preheaders for this email: [text]. Two each in the patterns curiosity, benefit, question, number (only with a sourced number), personal. No all-caps words, at most one exclamation mark.

Cut this email to 150 words max: [text]. Keep the one key message and the one link; remove filler and repetition.

Outline a five-email welcome series for [lead magnet] and [audience]: goal, subject, content in three bullets and spacing for each email. Selling no earlier than email 4, clearly labelled as advertising.

Here are three emails I wrote myself: [emails]. Describe my style in five rules, then write this new email in that style: [content].

Draft consent wording for my newsletter sign-up: content [topics], product recommendations [yes/no], open and click tracking, unsubscribe any time, link to the privacy policy. As a draft for my legal adviser.

Here’s my list data: sign-up source, clicks over the last three months, topics: [export without names and addresses]. Suggest three to five sensible segments and what each should get.

Write an email recommending [tool]: who it suits, who it doesn’t, price with an as-of date [price], one link labelled “Ad · affiliate link”. No income promises, no invented experience.

Check this email for typical spam signals: subject, wording, links, image-to-text ratio, missing plain-text version: [HTML/text]. List specific changes.

Here are the results of my last ten emails: [subject, send time, opens, clicks, unsubscribes]. What patterns do you see, and how reliable are they at this data volume?

Write an email to subscribers who haven’t opened or clicked in six months: honestly ask whether they still want emails, with one click to stay. Max 100 words.

Write an email to customers who bought [product], about [similar product of ours], with a note that they can opt out of such emails at any time. Max 120 words.

Turn this HTML email into a readable plain-text version with link text instead of bare URLs, in the same order: [HTML].

The most common mistakes

1. Sending without consent. In Germany, marketing email needs prior express consent.
2. Ignoring the setup. Without SPF, DKIM and DMARC, even good emails land in spam.
3. Clickbait subject lines. More opens short term, less trust long term.
4. Overrating open rates. They’ve become less precise; clicks and replies matter more.
5. AI without sign-off. A wrong number or promise goes to everyone at once.
6. Selling from the first email. Deliver value first, then recommend.