How to Write a Cold Email With AI (Templates Inside)

Writing a cold email with AI sounds like a dream: paste a prospect, get a perfect message, hit send. I tried that lazy version first, and my reply rate was embarrassing. Then I tested a smarter split, where AI does the research and I keep control of the writing, and things actually started working. So let me tell you where AI genuinely helps with cold email, where it quietly sabotages you, and give you the templates I use.

Quick honesty before we start: a 12,000-email test that made the rounds showed AI writing alone hitting around a 4.1% reply rate while human-written emails hit 10.4%. That gap is the whole reason this guide exists. AI is a fantastic assistant and a mediocre author, and once you accept that, it becomes genuinely useful.

Use AI for research, not the whole email

The single best use of AI in cold outreach is finding your hook. Before you write anything, ask it to scan what a prospect or company is doing and surface a reason to reach out that is not generic. Real personalization means referencing something specific, a new hire, a recent product launch, a pivot they announced on a call, not "I loved your website."

A prompt I paste into ChatGPT or Claude:

• "Here is a company's recent LinkedIn posts and news. Give me three specific, recent things I could reference in a cold email that signal they might need help with [your offer]. No fluff, just the hook and why it matters."

Now you have a real reason to email, and the human part (turning that hook into something that sounds like a person) stays with you. This split is the whole trick, because AI is great at scanning and bad at sounding human, so give it the scanning job.

The cold email with AI template I actually send

Cold emails work when they are short, specific, and about the reader. Here is the bones of mine, and you fill the brackets using your AI-found hook. This is the exact cold email with AI structure I reuse:

• Subject: quick question about [specific thing they did]
• Line 1: Saw [specific recent event]. [One genuine sentence reacting to it.]
• Line 2: We help [similar companies] with [the one problem that connects to that event].
• Line 3: Worth a quick chat, or should I send a two-line summary instead?

That last line gives an easy out, which weirdly gets more replies than a hard ask for a meeting. Keep the whole thing under 90 words. If your draft is longer than what fits on a phone screen without scrolling, cut it.

How to make AI drafts sound human

If you do want AI to draft the body, the prompt matters enormously. Generic prompts give you generic, salesy mush that screams template. Here is the one I use to get something closer to usable:

• "Write a cold email under 80 words to [role] at [company]. Reference this specific hook: [paste hook]. Tone: casual, like a real person who did their homework. No buzzwords, no 'I hope this email finds you well,' no exclamation marks. End with a soft, low-pressure ask."

Then I rewrite at least one line in my own words, because AI has tells. It overuses phrases like "I wanted to reach out" and loves a tidy three-part structure that pattern-matches as a bot. Breaking that rhythm yourself is what keeps you out of the mental spam folder. A good test: read it aloud, and if you would never say it to a person at a coffee shop, rewrite it.

One more thing that helps a lot: ask AI to write three subject line options and three opening lines, then pick the best of each yourself. It is great at generating quantity, and you are great at judging which one sounds least like marketing. I keep a running note of the openers that earned replies and feed those back in as examples, which teaches it my voice over time. Avoid anything that looks like a template merge field gone wrong, because "Hi {FirstName}" landing as literal text is the fastest way to get deleted.

Deliverability is the part nobody talks about

Here is the thing that matters more than your words: getting into the inbox at all. Deliverability in 2026 is brutal. A spam complaint rate as low as 0.3% can get your domain blacklisted, so the best email in the world fails if it lands in spam.

Before you send any volume, handle the boring technical setup:

• Set up SPF, DKIM, and DMARC on your sending domain. These are basically mandatory now, and skipping them is the fastest way into the junk folder.
• Use a separate domain for cold outreach so a problem does not poison your main company email.
• Warm up the mailbox for a few weeks before sending real campaigns, gradually increasing volume.

Tools like Instantly and Saleshandy bundle warm-up and deliverability features, and Snov.io is a solid all-in-one if you also need to find verified addresses. Salesforge can write full sequences in many languages if you run truly large campaigns. I am not saying buy all of them, just do not skip the technical groundwork, because no clever wording rescues a blacklisted domain.

Where a cold email with AI goes wrong

Let me save you my mistakes. The biggest one is mass-personalization that is obviously fake, where every email has the same hollow "I noticed you're passionate about innovation" line. Prospects can smell it instantly, and it reads worse than no personalization at all.

The second mistake is volume without quality. AI makes it easy to send 2,000 emails, and if they are bad, you have just burned 2,000 prospects and dinged your domain reputation. Fewer, sharper emails beat a flood of mediocre ones every time. I would rather send 50 genuinely researched emails a day than 500 templated ones.

The third mistake is forgetting the follow-up. Most replies come from the second or third message, not the first, so I ask AI to draft a short follow-up that adds a new angle instead of just "bumping this to the top of your inbox." A follow-up that says "saw you also just opened a second location, this gets more relevant" performs far better than a guilt-trippy nudge. Keep follow-ups even shorter than the original, two sentences if you can.

Who should skip AI for cold email

If you only send a handful of high-value outreach emails a month, honestly just write them yourself. The personalization bar for a dream client is too high for a tool to clear, and the time you save is tiny. AI earns its keep when you are sending at real volume and need help researching hooks fast. And if your deliverability foundation is not set up, no tool and no template will rescue you, so fix that first.

The Bottom Line

The winning move with a cold email with AI is to let the machine do the research and keep the writing human. Use AI to find a specific, recent hook, drop it into a short and reader-focused template, then rewrite a line or two so it does not sound like a bot. Nail your SPF, DKIM, and DMARC setup before you send anything at scale, send fewer and better emails, and you will land far more replies than the people blasting generic AI drafts into the spam folder.