AI assistant
Two ways to use AI inside the editor — a chat for whole-email edits, and AI sections that regenerate on every send.
EmailBuilder has two distinct AI surfaces. They look similar but serve different jobs.
| Chat | AI sections | |
|---|---|---|
| Where | Right panel, Chat tab | Inline on a section |
| What it edits | Anything you ask | One section at a time |
| When it runs | When you send a chat message | Every time the email is sent or previewed |
| Best for | Bespoke edits, exploration, one-off rewrites | Recurring sends where the content changes (digests, newsletters) |
Chat
Open the Chat tab in the right panel. Type what you want and the AI rewrites, restyles, or extends your email accordingly. A few example prompts:
- "Add a 3-card section under the hero with my latest blog posts."
- "Rewrite the body in a more casual tone."
- "Make all buttons emerald green."
- "Translate to Spanish."
The AI sees your full MJML as context and replies by modifying it. Edits land directly in the editor — there's no manual copy/paste step.
Quick prompts
Above the input, + text opens a list of your saved quick prompts (Casual, Formal, Shorter, …). Click one to drop it into the textarea. Use the gear icon to add or rename your own quick prompts.
Attaching context
The AI works better when it knows what to use. Above the input:
- + module — attach one of your saved modules as a layout template. The AI will replicate its MJML structure exactly (same tags, attributes, nesting) and only swap the content. Useful when you want a new section to match the look of an existing one.
- + file — attach an image from your media library. The AI will use it as the
srcof any image it inserts.
Attached chips persist across chat turns and survive switching between Visual / Code / Chat modes.
Source-aware context
If your email is attached to a content source — an RSS feed, a Notion database, a Google Sheet — the AI can read items from it on demand. When you ask "build a digest of my latest 3 articles," it pulls the real items from the source rather than inventing them.
This is opt-in by attaching the content source on the email's settings page. The AI will only call the source-read tool when your prompt mentions items, links, or summaries — routine styling questions don't trigger an extra round-trip.
AI sections
AI sections are different. Instead of asking the AI to do something once, you mark a section as AI-editable and write a prompt as that section's text. From then on, every time the email is sent or previewed, the AI replaces that section with freshly generated MJML.
Marking a section as AI-editable
- Right-click the section → toggle ✨ AI content.
- The section's contents are replaced with an empty column. Click it — instead of the block menu, you get a prompt chip picker with templates like Welcome intro, Product highlight, Quote, etc.
- Pick a chip (or "Custom prompt") and edit the resulting text. That text is now the section's prompt.
A small teal sparkle badge appears on the section's top-left corner so you can tell at a glance which sections are AI-driven.
Regenerating
Click the sparkle badge to regenerate just that section. The badge pulses while the AI runs (10–30 seconds typically). The rest of the email is left untouched.
What happens on send
When you send the email (or any automation creates a send from this template), the pipeline:
- Loads your attached content sources.
- Sends the prompt + source items + your full MJML to the AI as context.
- Replaces each AI section with the AI's generated MJML.
- Renders the final HTML and dispatches to your ESP.
The original template is never modified — each send is a new copy with that send's generated content. The prompt itself lives in the template, so the next send picks up where the last left off.
Preview before sending
The Preview Next Send button (eye icon in the Sends list) shows you exactly what the next AI-composed send will look like, without sending it. You can then Save as Draft to schedule it for later editing or sending.
Limits
- Each generation runs the same Claude model your account is configured for. There's a per-month token budget on each plan — usage shows on the Account page.
- The AI never returns invalid MJML — it self-validates before emitting — so failed generations are rare. When they do fail, the section is left unchanged and a warning surfaces in the preview drawer.
- The AI only edits sections you've marked AI-editable. Static sections are passed through untouched, even if your prompt references them.