How to Write Cold Emails With AI That People Actually Reply To

Why most AI cold emails fail

A client came to me last year frustrated. They’d used AI to write and send 3,000 cold emails and gotten almost nothing back. A reply rate so low it rounded to zero. They assumed AI just didn’t work for outreach.

Then I read the emails. Every single one was the same. Same opening, same pitch, same forced compliment, with only the first name swapped in. The AI hadn’t failed. It had done exactly what they asked: write one generic email three thousand times. That’s not outreach. That’s spam with extra steps.

Here’s the uncomfortable truth about cold email in 2026. Inboxes are smarter, people are more tired of being sold to, and a generic message is dead on arrival. AI makes it trivially easy to send more bad emails faster. The skill now isn’t volume. It’s using AI to do the one thing that used to be too slow to do at scale: genuine personalisation.

The research step that changes everything

A cold email lives or dies on its first two lines. If the opening proves you actually know something specific about the person, they keep reading. If it’s a generic “I came across your company and was impressed,” they’re gone.

So before AI writes a word, it needs raw material. Spend two minutes gathering a few real details about the person or company: a recent LinkedIn post, a company announcement, a specific challenge their industry is facing, something on their website. Then hand that to the AI.

That research is the difference between a 0.2% reply rate and something worth sending. AI can’t research the person for you in a browser tab, but the moment you give it real detail, it turns that detail into a natural, personal opener in seconds. That’s the leverage.

The prompt framework that works

A cold email that gets replies follows a simple shape, and you can brief AI to hit every beat. I teach it as four parts:

• The hook: one specific, researched detail about them. Proves you’re not blasting.
• The relevance: why you’re reaching out to them specifically, tied to that detail.
• The value: one concrete thing you can do for them, in plain terms, no buzzwords.
• The ask: a small, low-friction next step. Not “book a 30-minute demo.” More like “worth a quick reply?”

Notice how much I’m telling it what NOT to do. AI defaults to corporate filler, so you have to actively steer it away. The same brief-it-properly principle runs through all our marketing work, like our guide to writing emails with ChatGPT.

Personalising at scale without faking it

Here’s where AI genuinely earns its place. Real personalisation used to mean a human researching every prospect, which capped you at maybe twenty good emails a day. AI changes the maths: you do the research, and it does the writing, so you can send fifty genuinely personalised emails in the time one used to take.

The workflow that scales: build a simple sheet with a few real research notes per prospect, then feed each row to the same proven prompt. Every email shares your framework and voice, but each opener is built from that person’s actual details. It’s personalised at scale, not faked at scale. There’s a real difference, and your reply rate will tell you which one you’re doing.

If your outreach is B2B and LinkedIn-led, it’s worth reading our take on what actually works for B2B lead gen in 2026 alongside this, because the channel and the message have to match.

The follow-up most people skip

Here’s something that surprises people every time I bring it up: a large share of replies come from the follow-up, not the first email. People are busy. Your first message landed at a bad moment, got buried, and was forgotten. That’s not a no. It’s a not-right-now.

So write the follow-up at the same time you write the first email, and let AI help. The rules are simple: keep it shorter than the original, add one new angle or piece of value instead of just “bumping this up,” and never guilt-trip. Nobody replies to “I’m following up again as I haven’t heard back.”

One follow-up is plenty. Two at the absolute most. Past that you’ve crossed from persistent into annoying, and AI will happily write you ten if you let it. This is exactly the kind of judgement the machine can’t make for you, which brings us to the part that’s still entirely on you.

The 5-step cold email workflow

Here’s the whole thing as a routine you can run for any campaign.

The research step is the one people want to skip. Don’t. It’s the step that makes everything after it work.

Mistakes that send you straight to spam

Let me save you some pain. These are the errors I see most often when people let AI run cold email unsupervised.

• Identical emails at volume. The fastest way to get flagged. Vary the message, not just the name.
• Fake personalisation. “I love what you’re doing at [Company]” with no specifics is worse than no compliment at all. People can smell it.
• Walls of text. AI loves to over-explain. Cold emails should be short enough to read on a phone in ten seconds.
• The big ask. Asking for a 30-minute call in a first email from a stranger almost never works. Start smaller.
• No human check. Always read what AI wrote before it goes out. It will occasionally produce something off-tone or just wrong.

Here’s the bottom line. AI didn’t kill cold email and it didn’t save it either. It just raised the stakes. The lazy operators now send more garbage and get ignored harder. The smart ones use it to personalise at a scale that was impossible before. Decide which one you want to be, and prompt accordingly.

Hina Mian, Co-Founder of Future Factors AI

Hina is a marketing strategist with over a decade of hands-on campaign experience across B2B and consumer brands. She writes about using AI to run leaner, sharper marketing without losing the human touch. Future Factors offers AI Bootcamps, Corporate Workshops, and Speaking & Consulting for teams that want to put AI to work properly.

More about Hina →