AI Agent Email Marketing: The Complete Guide
AI agent email marketing explained: what an agent can run on its own, how to connect it in five steps, prompts that work, and where you keep the final say.
AI agent email marketing: the complete guide
AI agent email marketing means handing the operational work of email — building lists, importing contacts, writing templates, creating campaigns, running sequences, reading statistics — to an AI agent that calls an email platform's API on your behalf. You state the goal in plain language; the agent makes the calls, and you approve every real send. The short answer: an agent can already do almost everything you do in an email marketing tool, as long as the tool exposes every action through an API and refuses unsafe sends on its own. This guide covers what that looks like, how to set it up, which prompts work, and where the human stays in charge.What is AI agent email marketing?
Most "AI email marketing" features are writing assistants: they suggest a subject line inside a screen you still drive by hand. An agent is different. It acts. It reads your lists, creates the draft, renders it, checks why a send would be refused, and reports back.| Approach | Who clicks | What the AI does | Typical limit |
|---|---|---|---|
| AI writing assistant | You | Suggests copy | Stops at the text |
| Classic automation (drip, flows) | Nobody, after setup | Nothing, rules run | Rules you configured by hand |
| AI agent + email API | The agent, you approve sends | Plans, builds, checks, measures | Only what the API exposes |
Why email marketing automation with AI fits agents so well
Email marketing is a long chain of small, verifiable operations. Each step has a clear input and a clear output, which is exactly what an agent handles well:- Lists and contacts are structured data. "Import this CSV into Newsletter and tag everyone
webinar-oct" is one bulk call. - Templates are HTML. An agent writes HTML fluently, and a render endpoint lets it check the result without sending anything.
- Campaigns are state machines:
draft,scheduled,sending,sent. An agent can read the state and know exactly what is possible next. - Statistics are numbers. Opens, clicks, bounces and unsubscribes come back as JSON the agent can compare with your last ten campaigns.
What an AI agent can run in AgentsMail
AgentsMail is an email marketing tool designed to be driven by an agent. Through its API v1 and the AgentsMail skill, your agent can:- Manage the audience: create lists, add or bulk-import contacts (up to 500 per call), tag them, build segments, manage topics and hosted sign-up pages.
- Write the email: create HTML templates with merge tags such as
{{firstName}}and{{unsubscribeUrl}}, mark editable zones so a human can rewrite the copy later, and handle dark mode. - Render safely:
POST /api/v1/renderreturns the final HTML, plain text and warnings, with no side effect at all. - Send: create a draft campaign, send a test, schedule it, or launch it once you approve.
- Automate: build sequences (a journey triggered when a contact joins a list or receives a tag) and series (dated steps sent to a whole audience).
- Measure: read delivery, opens, clicks, unsubscribes and clicks per link for each campaign.
How to set up AI agent email marketing in five steps
01
Create an account and an API key
Sign up, then create a key under Account, API keys. A key opens exactly one organization, so the agent can never wander into someone else's data.
02
Install the skill in your agent
One command installs it. The skill teaches the agent the endpoints, the shape of every answer and the safety rules.
bash
npx skills add https://github.com/mikecodeur/skills --skill agentsmail03
Ask the agent whether you can send
The first call is always the diagnostic,
GET /api/v1/me. It answers with ready and a list of blockers such as domain_not_verified, so nothing is built on a setup that cannot send.04
Verify your sending domain
The agent declares the domain, reads back the DNS records to publish (DKIM and an ownership TXT record), and polls verification until it passes.
05
Import the audience and draft the first campaign
Hand over a CSV, name the list, and ask for a draft. The agent renders it and sends a test to your own inbox before anything else.
Prompts that work for email marketing automation with AI
Good prompts name the audience, the goal and the stopping point. The stopping point matters most: tell the agent where to stop and wait for you.text
Check whether my account can send. If it can't, tell me exactly what is
blocking and fix what you can. Don't create any campaign yet.text
Import customers-q3.csv into the list "Customers". Tag every row
"q3-import". Report how many were created, updated and rejected, with the
reason for each rejection.text
Draft the October newsletter for the list "Newsletter": three product
updates, one call to action to the changelog. Render it, send a test to
me@example.com, then stop and wait for my approval.text
Compare the last five campaigns: click rate, unsubscribe rate, best link.
Tell me which subject lines worked and propose two for next week.Where the human stays in the loop
Never trigger a real send without explicit human approval for that specific send.That is the first rule in the AgentsMail skill, and the product reinforces it. The agent drafts, imports and renders on its own. A real send needs your go for that campaign, and approving one never approves the next. Sending a campaign through the API is not open on every account yet. When it is not, the agent prepares and schedules everything, and you press send in the interface. Either way, the decision stays with you.
The guardrails that keep an agent safe
Giving an agent access to your list is only reasonable if the platform protects your sender reputation when the agent gets something wrong. AgentsMail does it at several levels:- Named refusals. A send that cannot go out is refused with a reason: domain not verified, ownership unproven, not enough sending units, daily quota exceeded. A refused scheduled campaign goes back to draft and owners are notified. A refusal never sends half a campaign.
- Mandatory unsubscribe link. Rendering refuses a template without
{{unsubscribeUrl}}. - Suppression list. Hard bounces and spam complaints are excluded from every future send, automatically, and cannot be removed by a careless import.
- Silent imports. Importing contacts never sends anything, so a bulk import cannot email 5,000 people by accident.
- Spam risk report. Before sending, you can get an editorial read of the subject, preview text and body: clickbait, artificial urgency, subject/content mismatch. It advises; it never blocks.
Measuring results with your agent
Once a campaign is sent, the agent reads the same figures the Stats screen shows:json
{
"success": true,
"data": {
"sent": 1840,
"delivered": 1826,
"bounced": 14,
"uniqueOpens": 712,
"openRate": 0.39,
"uniqueClicks": 151,
"clickRate": 0.083,
"unsubscribed": 6,
"unsubscribeRate": 0.0033
}
}How much AI agent email marketing costs
API calls are never billed separately on AgentsMail. You pay for the plan:| Plan | Price | Emails / month | Contacts | Sequences |
|---|---|---|---|---|
| Free | €0 | 3,000 | 10,000 | 2 |
| Pro | €19/month | 50,000 | 50,000 | 10 |
| Enterprise | €200/month | 500,000 | Unlimited | Unlimited |
FAQ
Can an AI agent send emails on its own?
Technically yes, if the platform lets it. In practice you should not let it: on AgentsMail the agent drafts and schedules autonomously, but every real send requires your explicit approval for that campaign. Test sends count as real sends too.Which AI agents work with AgentsMail?
The skill installs into Claude Code, Cursor, Codex, Copilot, Gemini and others with one command. Any agent able to make HTTP requests can also call the API v1 directly with an API key.Is AI email marketing automation different from drip campaigns?
A drip campaign is a fixed rule that runs after you configure it. An agent can build that rule for you, as a sequence triggered by a list or a tag, and then adjust it based on the statistics it reads.Will an agent hurt my deliverability?
Not if the platform refuses unsafe sends. Verified domains, a mandatory unsubscribe link, an automatic suppression list and named refusals mean an agent's mistake ends in a refusal, not in a reputation incident.Key takeaways
01An AI agent runs the operations of email marketing through an API:
audience, templates, campaigns, sequences, statistics.
02Start every session with the readiness check, so nothing is built on a
setup that cannot send.
03Render freely, test once, and approve each real send yourself.
04The platform's guardrails matter as much as the agent: named refusals,
suppressions, silent imports.