Personalizing at Scale
What it does: makes a message that goes to two thousand people read like it was written for one, without anyone writing two thousand messages.
Run it when: the copy already works and you want it to hold up at volume.
Time: an hour to set up properly, once.
Before you start
A message that already reads well with no personalization at all
The brand profile filled in, including target buyer titles
A list where you know which fields are actually populated
The steps
1. Start with the free layer: static merge tags.
First name, last name, company, position, headline, location. They swap in the lead's own values at send time and cost nothing. First name is the only one present on effectively every lead, so reach for the others only when you know the list has them filled. Company always means the prospect's company, never yours.
2. Write copy that survives a missing value.
If a message reads strangely when the position field is empty, it will read strangely for a few hundred people. Test each tag by mentally deleting it.
3. Add one AI opener, not five.
An AI tag runs a per-lead generation at send time and writes a line from that lead's profile. The standard one writes just the opening line, which is spliced into "Hi Sarah, ___." The built-in options cover a first line, a restrained compliment, a "saw you" opener, and a whole-message writer. One per message is plenty; more makes the message a collage.
4. Make the message stand on its own without the AI line.
When a generation fails, the tag is removed and the seam repaired, and the rest of the message sends as written. So your trust line and CTA must carry the message alone, and nothing after the opener may refer back to it with a "given that" or a "that's why".
5. If you write a custom prompt, carry everything in it.
The prompt runs against a fast model with no house context, so it needs: the role and task in one line, the lead placeholders, explicit handling of missing data, the voice rules restated, a hard output contract ("return only the line, no quotes, no preamble"), a word limit under 15 words, and the grammar seam (start lowercase, no ending period, because the template supplies it). Output is capped around one sentence, so asking for more gets it clipped mid-word.
6. Use the verified opener when you want a fact, not a compliment.
One tag behaves differently from all the others: instead of writing from the profile, it writes from a verified fact on the lead's company website, an open role or a news item. It needs target buyer titles in the brand profile, and research only runs while the campaign is active, so a draft shows an example rather than real facts. Put it first in the message, do not stack a greeting in front of it, and never refer back to the fact from the body.
7. Put the fact in one step, never in follow-ups.
By the second message the same fact is stale and mentioning it again reads like a script running. One step per campaign.
8. Never put a researched fact in a public comment reply.
A fact about someone's employer, posted publicly under their comment, reads as surveillance rather than homework. Keep it in private messages.
9. Know what personalization cannot do.
It cannot make a weak offer interesting or a long message short. If reply rates are poor with a clean message, the problem is the offer or the audience, not the opener.
Do this / Not this
Do this | Not this |
|---|---|
One AI opener per message | Three AI tags in four lines |
Copy that reads complete without the AI line | "Given that, we thought..." |
First name as the default merge tag | Position and headline on a list where they are empty |
Verified fact in the first message only | The same fact repeated in follow-up two |
Target buyer titles set on the brand profile | Wonder why no facts are being found |
Facts in private messages | Facts in a public comment reply |
Fix the offer when replies are low | Add a fourth personalization layer |
Done when
Every static tag has been tested against a missing value
At most one AI-generated line per message
The message reads complete with the AI line removed
Custom prompts specify output format, length and missing-data handling
If the verified opener is used, buyer titles are set and research is started on the live campaign
No researched fact appears in a follow-up or a public reply
