How to Automate Cold Emails With AI? (2026)

Most AI cold emails read like AI cold emails: the same flattering opener, a vague compliment, a long pitch, and a reply rate to match.

The problem is not the model. It is what you feed it and what you let through: thin data in, no checks, and no human reading the output before it sends.

This guide shows how to automate cold emails with AI so each email rests on a real fact, with a workflow we tested on 25 real sales leaders.

You can run it from an AI Agent in Claude Code, Codex or Cursor, or from an n8n form without code.

We ran both on real leads and a real Instantly account, with no email sent. Copy the workflow below, or follow the same method with your own tools.

We will walk you through the process step by step.

Everything is in one public GitHub repo: the skill, the offer and writing rules, a small helper script and the n8n workflow. Paste the prompt into your AI Agent, or pick the n8n tab to run it from a form.

Agent kit Free · Public

GitHub n8n

Let your AI agent set it up

Paste this prompt into Claude Code, Codex or Cursor. Your agent clones the public repo, follows its instructions and asks before every change.

Prompt · GitHub Copy

Clone https://github.com/evaboot-labs/ai-cold-email-writer, read AGENTS.md
and follow the skill. Write cold emails for my next 20 leads. Ask me about my
offer first, and show me every email before anything goes to Instantly.

In the repo

• The ai-cold-emails skill: leads, facts, emails, checks, approval, Instantly
• emails.py: the facts, the checks, the sheet and the Instantly push
• OFFER.md: what you sell and the rules every email follows
• The n8n workflow, with its setup guide

Open the repo on GitHub

Set it up in n8n

Paste this prompt into your AI agent and it walks you through the import, one step at a time. Or open the workflow JSON on GitHub and import it yourself.

Prompt · n8n Copy

Set up the AI cold email workflow in my n8n.
Clone https://github.com/evaboot-labs/ai-cold-email-writer and follow docs/n8n.md.
Walk me through importing n8n/ai-cold-email-writer.json, adding the Evaboot, Anthropic,
Google Sheets and Instantly credentials, and filling in Your settings, one step at a time.

In the repo

• The workflow JSON: a form in, approved emails in Instantly out
• docs/n8n.md: the four credentials and the settings, step by step
• The same checks as the script

Open the n8n JSON on GitHub Open the repo on GitHub

We used Sales Navigator, Claude and Instantly in our example, but with a few tweaks, the same method works with another model or sending tool, such as lemlist.

Can AI Write Cold Emails That Get Replies?

Yes, when it writes from facts. A model is good at turning a fact about a person into a natural first line, fast and at volume.

It is bad at knowing what is true. Ask it to "personalise" with nothing but a name and a company, and it fills the gap with flattery or invention.

So the work splits in two:

The workflow in this guide gives the AI one record of real data per lead, and puts two checks between its output and a send: a script, then you.

Ways to Automate Cold Emails With AI

Three families of tools write cold emails with AI, and they differ in where the data comes from:

1. AI Built Into Your Sending Tool

Tools such as Instantly and lemlist now write or personalise steps inside the campaign. It is the shortest path: the email is written where it is sent.

The limit is the data: the tool can only use the fields you imported, which are often a name, a title and a company.

2. Standalone AI Email Generators

Free generators and chat assistants write one email at a time from what you paste in. They are useful to draft a template or test an angle.

They do not scale to a list of 200, and nothing checks the output before you copy it into a campaign.

3. An AI Agent Working From Your Lead Data

An AI Agent connected to your data, such as Claude with the Evaboot MCP server, exports the leads with their profile and company fields and writes from those.

It is the most work to set up and the most control: you choose the data, the rules and the checks. This is the option we built and tested.

How to Automate Cold Emails With AI: Step by Step

Our workflow turns a Sales Navigator search into approved emails inside an Instantly campaign, in six steps.

The AI writes; the data, the checks and the approval are fixed and the same on every run.

1. Export Your Leads With Verified Emails

Start from a Sales Navigator search with a reason to write, such as people who changed jobs in the last 90 days. We used UK sales leaders at software companies with 51 to 500 employees.

Evaboot exports the search with verified emails and the fields a good email needs: title and time in role, headline and profile summary, company description, growth and sales team size.

Of our 20 leads, 18 came back with an email; one was marked invalid and one had none. Those two are skipped before the AI writes a word.

2. Write Your Offer and Rules Once

The agent reads one page, OFFER.md, before it writes. It holds what you sell, to whom, the proof you may use, the ask, and the rules:

The proof rule matters most: "never invent customers, numbers or results". A model asked for social proof will otherwise make some up.

3. Let the AI Write One Email per Lead

The agent writes one email per lead, or a skip with a reason. Each email names the hook it rests on, so you can judge it at a glance.

Here is one from our test, for a VP Revenue two months into the role, whose profile says he was the top seller at Stack Overflow:

Every claim in it comes from the export: the new role, his own profile, the company. The offer comes from OFFER.md.

4. Check Every Email Automatically

A script in the repo checks each email before a human reads it: subject and body length, links, the first name, template phrases, repeated openings, and the email itself.

The email check earned its place in our test. Datasite's LinkedIn page lists linktr.ee as its website, so its lead's email was built on that domain and could never have reached him.

The script now flags any email whose domain differs from the company's, and never loads one on a link site.

5. Approve the Emails in a Sheet

The emails land in a sheet, one row per lead, with the hook, the subject, the body and any email warning:

Read them, edit what you want, and write yes in the Send column for the ones to load. This is the step that keeps a bad email out of a real inbox.

6. Load the Approved Emails Into Instantly

The push adds each approved lead to your campaign with two custom fields: the body in personalization and the subject in ai_subject. The campaign's first step uses only those two:

Only verified ("safe") emails are loaded. In Instantly's preview, each lead shows the exact email the AI wrote and you approved:

Then you attach your sending accounts and launch the campaign yourself. To refill it every week, put the search on a recurring Evaboot export: that is what we call IBAO (Intent Based Automated Outreach).

How to Run It in n8n

n8n runs the same agent from a form: paste a Sales Navigator search and the number of leads. Claude writes the emails into your sheet, and a second form page asks before anything goes to Instantly:

After the yes, the workflow reads the sheet back and loads only the rows you marked, with a safe email:

We tested it on 5 new leads in IT services. It wrote 3 emails, skipped the 2 leads without a usable email, and loaded the 3 approved rows into Instantly.

Example: 20 New Sales Leaders, 18 Emails

Here is what each step returned for UK sales leaders who had just changed jobs:

The hooks were the new role in most emails, company growth where it stood out (headcount up 117%, 88% and 50% in a year), and the size of the sales team for the largest companies.

Deliverability Basics Before You Launch

A perfect email in the spam folder gets no replies. Four basics come before the first send:

Our guide to email deliverability covers each one, and why bounced emails matter.

Common AI Cold Email Mistakes

Four mistakes cause most failed AI cold email campaigns:

They are all fixed before the send, not after.

1. Feeding the AI a Name and a Company Only

With nothing to work from, the model writes "I came across your profile and was impressed". Export the profile and company fields, and require a named hook.

2. Letting It Invent Proof

A model asked for social proof will produce a customer and a percentage. Put the proof you may use in the offer file, and forbid everything else.

3. Sending Without Reading

Automation makes one bad email a hundred bad emails. Keep an approval step: reading 20 emails takes ten minutes.

4. Trusting Every Email Address

Verified does not always mean right: a company page with a link-in-bio website produced a wrong email in our test. Check the email domain against the company's.

Frequently asked questions