AI Cold Email Personalization at Scale: Write a Unique Email for Every Lead (Without Sounding Like a Robot)

Why "Hi {{First Name}}" isn't personalization any more

Merge tags were a big deal in 2015. Today, a first name and company name inside an otherwise identical email reads as automation, because it is. Real personalization means the reason you're writing is specific to them: something on their website, a job post, a review, a product launch, their local market.

It also matters for deliverability. Thousands of near-identical emails form a pattern filters recognise. Thousands of genuinely different emails don't.

The workflow: one email per lead

1. Build a focused list with the fields AI needs: name, company, website, role, and anything you know (city, industry, a recent event).
2. Research each lead. An AI assistant with web access can read their website and note one or two specific, true observations.
3. Generate the email from those observations with a tight prompt and your offer.
4. Save everything in a spreadsheet: one row per lead, with columns for each email's subject and body.
5. Review before sending: spot-check the writing, and run automatic checks for empty fields, placeholders and near-duplicates.
6. Import and send so every lead gets exactly the email written for them, with follow-ups.

A prompt that produces believable emails

The difference between a robotic AI email and a good one is almost entirely the prompt. Constrain length, ban the clichés, require a specific observation, and give it your real offer and proof.

You write short B2B cold emails for {{Your company}}, which {{one-line offer}}.
Proof you may use: {{one real result}}.

For the lead below, write ONE email:
- Subject: 2-5 words, lowercase, specific to them. No "quick question".
- Under 90 words. Plain text. No links. No emojis.
- First line: one specific, true observation from their website or the
  details given. If you can't find one, say so instead of inventing.
- Then: the problem it suggests, our offer in one sentence, the proof.
- End with one low-effort question.
- Sign off with exactly: [Your Name]
- Never use: "I hope this finds you well", "I came across",
  "I wanted to reach out", "game-changer", "revolutionize", "synergy".

Lead: {{name}}, {{role}} at {{company}} ({{website}}), {{city}}.

The spreadsheet format

Keep one row per lead and put each email in its own columns: Subject 1 and Email 1 for the first email, Email 2 for the first follow-up, and so on. This is the format of Kaysid's import template, so no column mapping is needed:

Quality checks before you send

AI makes mistakes at scale, and a mistake in a cold email is public. Check for:

Using AI assistants that can browse

Assistants with web access, like ChatGPT, Claude or Grok, can visit a lead's website and pull out real details, which is what makes the first line credible. A practical setup:

• Give the assistant your offer, proof, rules and the column layout of your spreadsheet.
• Ask it to work through the list in batches of 20 to 50 and return rows in that layout.
• Ask it to mark rows where it couldn't find anything specific, instead of guessing.
• Paste the results into your spreadsheet and import it.

Kaysid's import screen includes ready-made instructions you can copy straight into an AI assistant, plus a template spreadsheet in the right format.

Personalized follow-ups

If the first email is personal, the follow-ups can build on it. A good pattern: email 1 is fully personal; email 2 adds a second, personal angle; the breakup email can be shared across everyone. Leads without a personal email for a step simply get the shared one. See How to write a follow-up sequence.

Does AI personalization actually improve replies?

When the observation is real and relevant, yes, often substantially, because the email answers the reader's first question: "why are you writing to me?". When it's shallow ("I saw you're the CEO of Acme"), it performs no better than a template, and sometimes worse, because people recognise the pattern.

Keep it honest

• Never claim you did something you didn't ("I loved your podcast episode" when you didn't listen).
• Don't fake familiarity or invent mutual connections.
• Make sure the observation is about the business, not intrusive personal details.

Send every lead its own email with Kaysid

• Import a spreadsheet (Excel or CSV) where every lead has its own emails. Columns are detected automatically.
• Review every lead before launch: empty fields, unknown tags, leftover [Your Name] placeholders, emails that read almost the same, and leads with nothing written.
• Leads with no email for a step are held, never sent a blank or half-filled email, until you write it or skip them.
• Your name is filled in automatically from the mailbox that sends each email, even with several mailboxes in one campaign.
• Follow-ups thread under the first email and stop when the lead replies.