Get Started

AI Email Writing Best Practices: Getting Reliable, On-Brand Results

Published

The gap between “made an email” and “made a good email”

Connecting an AI assistant to your email builder is the easy part. The MCP setup takes a few minutes. What separates a polished, on-brand email from generic filler is everything that comes after — how you ask, how you iterate, and how you review.

The good news: AI email writing responds predictably to a handful of disciplines. Get these right and your hit rate climbs sharply. This is the checklist of what consistently matters.

1. Give real content, not placeholders

The single biggest predictor of quality. “Add a button” produces a button labeled Button. “Add a button labeled ‘Shop now’ linking to temway.com/shop, brand-colored, with a large shadow” produces your button. AI can’t invent your offer, your link, or your headline better than you can — it can only assemble what you give it.

  • Supply the actual copy, links, and image URLs in the prompt.
  • If you don’t have them yet, ask the assistant to propose options first (“give me three subject lines under 50 characters”), then pick and proceed.

2. Anchor every prompt to your branding

“Use my branding” or “start from my branding” is the highest-leverage phrase in AI email writing. It tells the assistant to pull your colors, fonts, and logo into the email automatically — everywhere, consistently — instead of you setting them block by block.

Set up your workspace branding once, then make “from my branding” the default opener for every new email. Every subsequent prompt gets easier.

3. Name the audience and tone

“Make it good” steers nothing. “For busy product managers, confident and concise, no exclamation marks” steers everything. Audience and tone constraints shape the copy more than any stylistic instruction, because they tell the assistant who it’s writing for and how it should sound.

4. Describe the structure block by block

Emails are a single column that stacks top to bottom — there are no side-by-side columns. Spell out the order you want: headline, then intro, then image, then three bullets, then a button, then the footer. A prompt that lists the structure produces a clean first draft; a vague one produces a guessing game.

Need an image beside text? Stack them (image, then text below) or design that piece once as a layout and drop it in.

5. Iterate one change at a time

Stacking five edits in one message works, but it’s hard to tell which change caused what — and hard to undo one without undoing all. Make one refinement per message: “make the hero button larger”, then “shorten the intro”, then “add a divider above the footer”. Smaller steps keep you in control and make the conversation easy to follow.

6. Describe effects, not properties

You don’t need to know the preset names. “Give it depth”, “make it feel more premium”, “add some breathing room” all work — the assistant maps them to the right shadow, spacing, and color settings. Reach for effect language first; fall back to specifics (“4px more padding”) only when a tweak refuses to land.

7. Preview often, and preview twice before sending

A web preview after each major change keeps you aligned, and it’s fast — use it liberally. Before you actually send, ask for an HTML preview once to confirm how the email lands in Gmail and Outlook. The two previews serve different purposes: web for speed and iteration, HTML for inbox fidelity.

8. Always preview before you publish

It sounds obvious, but it’s the most-skipped step. End every editing session with “show me a web preview.” Once it looks right, publish. Never publish blind — a structural change that looked fine in your head can break in the render.

9. Keep layouts for artwork, emails for messages

Design your hero banner, feature section, or promo card once as a reusable layout, publish it, then drop it into every email that needs it. AI is excellent at composing and refining layouts — and once a layout exists, you can ask the assistant to “start a new email from my Gradient Hero layout” in a single line. See the reusable layouts guide.

10. Treat the assistant as a proposal engine

When you’re stuck, don’t force a structure — ask the assistant to propose one: “I want a welcome email that feels warm and short; suggest a structure, then build it from my branding and preview it.” You keep or adjust. Letting the assistant draft first, then reacting, is often faster than planning the whole thing yourself.

Common mistakes to avoid

  • Publishing the first draft. First drafts are starting points. Always preview, then refine at least once.
  • Embedding images. AI tools that emit raw HTML tend to base64-embed images, which blows past Gmail’s clipping limit. A builder that hosts images by URL keeps your email lean — see why Gmail clips emails.
  • Skipping branding. Without “use my branding,” you’ll manually fix colors and fonts on every block.
  • Asking once, judging forever. The conversation is the workflow. Treat it as an editing session, not a one-shot generator.
  • Forgetting the inbox check. Web preview is fast but not pixel-identical to an inbox. Run the HTML preview before you send.

Where to go next