Key takeaways
AI made it easy to write a thousand “personalised” cold emails. That’s exactly why they stopped working.
Gmail and Microsoft now run their own AI filters, trained on billions of messages. They spot templated outreach very accurately. The more an email looks mass-produced, the less likely anyone sees it.
What works in 2026 has three layers, in strict order.
First, the plumbing
Most cold campaigns die in the spam folder before anyone reads the copy. Deliverability benchmarks put average inbox placement around 84%. So roughly one legitimate email in six never arrives, and a badly set-up domain does far worse.
Before you write a word:
Keep outbound on separate domains from your company’s. A burned domain stays burned. You don’t want the chief executive’s email in spam because of an SDR campaign.
From the outside, a deliverability problem looks just like a copy problem. So check this layer first.
Then, precision
Every bounce from a stale address hurts your sender reputation. Every irrelevant send trains the filters against you. That makes a scraped list of 10,000 contacts a liability.
What does better is small, verified lists tied to real buying signals. Think a funding round, a new leader, hiring in the team you sell to, a new technology, or a surge of visits to your site.
Vendor data puts personalisation built on signals well ahead of personalisation built on company size and industry. That second kind only shows you looked someone up.
On benchmarks, average reply rates sit around 3 to 4%. Disciplined teams clear 10%. The 15 to 18% figures going around come mostly from companies selling personalisation tools. Below about 3%? Suspect targeting or deliverability before copy.
Last, AI for research and timing
AI earns its place in three jobs:
It fails as fully automated, unedited AI sequences at volume. Those end up with the phrasing and structure the filters are trained to catch. Spend the time the research saves on relevance, not more sends.
Email-only sequences also do noticeably worse than a mix of email, social touches and calls over about two weeks. So treat cold email as one part of the wider sales motion. Teams that work this way win twice. Their emails land, and their domains stay healthy enough to keep landing them next quarter.
Frequently asked questions
Does AI cold email still work in 2026?
Yes, but how it works has changed. Use AI to write and send more templated email and the inbox providers' own AI filters catch it. They're trained on billions of emails to spot mass outreach. What works is using AI for research: finding buying signals, timing outreach, and personalising from real context. Keep volume low and your setup clean.
How many cold emails can I safely send per day?
Common guidance in 2026 is roughly 35-40 emails per day per email address to protect deliverability. Use more warmed sending domains if you need more total volume. Blasting hundreds from one address is the fastest way to burn a domain's reputation. Once it's burned, every email from it lands in spam.
What is signal-based personalization?
It means mentioning a specific, current buying trigger in your outreach: a funding round, a leadership change, a hiring spree, a new technology. Generic facts like company size or industry don't count. Vendor-reported benchmarks keep showing signal-based emails beating generic outreach by several times. A signal shows you understand the prospect's situation right now. Firmographics only show you looked them up.
What's a good cold email reply rate in 2026?
Benchmark data puts average reply rates around 3-4%. Disciplined teams report 10%+, and vendor figures for tightly signal-personalised campaigns go higher. Below roughly 3%, treat it as a targeting or deliverability problem, not a copywriting one. Be sceptical of the highest claimed figures. They usually come from companies selling the tools.
Why are my cold emails going to spam?
Usually it's your setup, not your content. The common causes: missing or wrong SPF/DKIM/DMARC authentication, an unwarmed or burned sending domain, too much volume per address, and templated copy that looks like mass outreach. Industry benchmarks suggest roughly one in six legitimate emails never reaches the inbox. Fix authentication and warmup before you touch your copy.
Should I use AI to write my cold emails?
Use it to help with research and drafts, not to blast. AI is good at turning a prospect's context (their signals, role and likely priorities) into a relevant first draft. But fully automated, unedited AI sequences end up with the same patterns inbox filters are trained to catch. The teams getting results use AI to write fewer, sharper emails, not more, faster ones.
Three-minute lesson on this Where AI helps a seller Sales course, basic. Free, no sign-up.