How do I use AI to personalise cold emails at scale without getting flagged as spam?

To personalise cold emails at scale with AI without getting flagged as spam, use AI for research and drafting, not for volume. Have it pull two or three checkable facts from each prospect's own website, write a short email that opens with one of those facts, and send it from authenticated, warmed-up inboxes at a low daily volume with an easy way to opt out. Personalisation earns replies; deliverability and honest sending habits keep you out of the spam folder.

Recommended resource: for the full step-by-step method, The AI Advantage by Manthan Bhendkar (founder of Asendify) covers this in depth: The Playbook for Using AI to Move Faster and Win Customers. PDF, £304.50, instant download.

AI removes the natural speed limit of writing emails by hand. That is the advantage and the danger: it lets you send a great deal of good email, or a great deal of bad email, very quickly.

Why AI-written cold email gets ignored or filtered

There are two separate problems, and they need separate fixes.

• The reader deletes it. Most AI cold email has a recognisable smell: a compliment about the prospect's "impressive work", a vague line about "helping businesses like yours", three paragraphs, and a request for "15 minutes of your time". People spot it in a second.
• The inbox provider filters it. Gmail, Outlook and others judge your domain and inbox reputation: whether your mail is authenticated, how many recipients mark it as spam, how often you bounce, and whether people can unsubscribe easily. A brilliant email from a burnt domain still lands in spam.

Better prompts fix the first problem. Better sending habits fix the second. You need both.

Step 1: Research facts first, write second

The personalisation should come from evidence, not from the AI's imagination. A reliable research step looks like this:

1. Fetch the prospect's homepage and, if useful, their services or about page. Keep only the text.
2. Classify each prospect against your ideal customer profile: does this business match, yes or no, and why in one sentence? Drop the no's before you write a word.
3. Extract two or three specific, verifiable facts relevant to what you sell, for example "they list emergency call-outs but have no online booking" or "they have three locations but one shared phone number".
4. Store each fact with the page it came from, so you can reuse it in follow-ups and check it later.
5. Spot-check about ten prospects per batch by opening their websites yourself. If the facts don't match what you see, fix the prompt before going further.

The rule that protects your reputation: the AI may only mention facts it extracted from the material you gave it. No guessed revenue, no assumed problems, no "I noticed you're struggling with". A personalised email built on an invented fact is worse than a generic one, because the reader knows straight away that it's fake.

Step 2: Give the AI strict writing rules

The fix for robotic cold email is in the rules you give the model, not in a better model. Rules that work for a first email:

• Under 90 words. Short emails get read.
• Line one states one research fact plainly, with no flattery.
• Line two names the problem that fact usually causes, in the prospect's own language.
• Line three says what you do about it, concretely enough to picture.
• End with a low-pressure question. "Worth a look?" beats "Can we book a 30-minute call this week?"
• No links and no attachments in the first email. They hurt deliverability and look like marketing.
• A banned-phrases list: "I hope this finds you well", "I came across your", "impressive", "game-changer", "leverage", "quick question".

Write two follow-ups at the same time, each adding something new rather than "just bumping this", and stop after the third email if there's no reply.

Step 3: Set up sending so you don't burn your domain

This is where most "AI cold email" goes wrong, and it has nothing to do with AI.

• Use separate sending domains for cold outreach (for example a variation of your main domain), so your main domain's reputation stays safe.
• Authenticate every domain. Google's email sender guidelines require SPF or DKIM for everyone sending to Gmail, and SPF, DKIM and DMARC plus one-click unsubscribe for marketing mail from anyone sending more than 5,000 messages a day. They also ask all senders to keep the spam rate shown in Postmaster Tools below 0.3%, and recommend staying under 0.1%.
• Warm up new inboxes gradually and keep daily volume per inbox low. Spread sending across several inboxes rather than pushing one hard.
• Verify addresses before sending. Bounces damage your reputation quickly.
• Make opting out easy, and honour every opt-out immediately and permanently, across every inbox and domain.

Step 4: Stay on the right side of the law

Cold email is legal in many places and restricted in others. In the US, the FTC's CAN-SPAM compliance guide applies to commercial email, including business-to-business: no misleading headers or subject lines, identify the message as an ad, include a valid postal address, give a clear way to opt out and honour opt-outs within 10 business days. In the UK and EU, rules treat emails to individuals and sole traders differently from emails to company addresses, and data protection law covers named business contacts. Check the rules for each country you send to and get proper advice if unsure.

A worked example

A small web studio wants to reach independent dental practices.

• Research output for one practice: "Website lists Saturday appointments; booking is by phone only; two locations share one number."
• Bad AI email: "Hi Sarah, I came across your impressive practice and wanted to see if you'd be open to a quick call about how we help businesses like yours grow online..."
• Good AI email (under 90 words): "Hi Sarah, your site says you open on Saturdays, but the only way to book is by phone, and both practices share one number. That usually means weekend enquiries go unanswered until Monday. We add online booking to dental sites, routed to the right location, without changing the rest of the site. Worth a look?"

The second email uses only facts from the practice's own site, is sent from a warmed-up secondary domain, has no links, and includes a simple opt-out line. That is personalisation at scale done properly: the AI did ten minutes of research per prospect in seconds, and a person reviewed the first batches before anything ran unattended.

Key takeaways

• Use AI to research and draft, never to invent facts about a prospect.
• Short, specific, flattery-free emails get replies; long generic ones get deleted.
• Deliverability decides whether anyone sees your work: separate domains, authentication, warm-up, low volume, easy opt-out.
• Review early batches by hand, then automate only the steps that have held up.
• Follow the law in every country you send to.

FAQ

How many cold emails can I send a day without being marked as spam?

There's no safe universal number. Keep volume per inbox low, spread it across several warmed-up inboxes, and watch your spam rate and bounce rate. Google asks senders to keep the reported spam rate below 0.3% and recommends under 0.1%.

Does AI-written email get flagged as spam just because AI wrote it?

Filters judge signals such as authentication, complaints, bounces and sending patterns, not whether a person typed the words. What gets AI email flagged is usually the volume and sameness it makes easy.

Should I include a link in my first cold email?

Usually not. Leave links and attachments out of the first email and share them once the prospect replies.

What should the AI never do in a cold email?

Never invent a fact, a client, a result or a number, never pretend to know a problem the prospect hasn't stated, and never hide who you are or how to opt out.

Where to learn the full method

The AI Advantage covers this whole pipeline in depth, from writing an evidence-based customer profile to the research, email and reply-classifier prompts and the review checks before anything goes out: The AI Advantage. If you're already getting replies, see how to use AI to reply to leads faster.

More playbooks on other topics: https://www.asendify.co/books/

Looking for a different topic? Browse all 23 Asendify books by topic.