The problem this solves
Personalization has a reputation problem because most of it is fake: a first name, a company name, a line about the weather in their city. Recipients recognize the template instantly, and the reply rate reflects it. Real personalization references something that actually happened at the company, recently, that the recipient cares about. The information exists on every well-built list - the posting, the round, the technology change that put the company there - and it almost never makes it into the email, because writing 200 distinct openers by hand is a week's work.
The Personalization Agent writes from the evidence on the row. It reads the latest signal for each contact, decides which fact is the strongest hook, and writes a short first touch that names that fact and connects it to what you do, in the voice you describe. Ninety words is the ceiling because a first touch that a busy person reads on a phone has to fit on one screen, and because a short email with one specific fact outperforms a long one with three.
Nothing sends. The emails are loaded into Lemlist with each lead, as a campaign that exists but has not started, so every line can be read, edited and approved before the first recipient sees it. The evidence used for each email sits beside it as a variable, which makes the review fast: you can see the fact and the sentence it produced side by side.
How the mission runs
- Read each row and its latest signal. The Personalization Agent loads list <name> and, for every contact, reads the signals attached to the company - the most recent job posting, the latest funding round, any technology adopted or dropped - and picks the freshest and most relevant one as the hook. Where a row has several, it records which it chose and why, so the reasoning is auditable during review.
- Learn your voice. From the description in your prompt and any sample emails you attach, the agent sets the register - direct or warm, first person singular or plural, the level of formality - and the one-line description of what you do that every email will connect the signal to. It shows you two sample emails before writing the batch, so the voice is agreed before credits are spent on 200 drafts.
- Write the first touch under 90 words. Each email opens with the specific fact - the posting title and what it implies, the round and what it funds, the technology and what usually follows it - then bridges in one sentence to your product and closes with a low-friction ask. No two emails share an opening sentence, and the agent avoids the phrases that mark a template.
- Attach the evidence as variables. The signal type, the fact used and its source link are stored as custom variables on each lead, so the reviewer can check the claim and later steps in the sequence can reference the same fact without repeating it.
- Load Lemlist without starting the campaign. Leads are pushed into a Lemlist campaign named <name> with the first-touch email as the first step, the evidence variables attached and the campaign left unsent. A summary reports how many emails were written, which signal types were used, and any rows where no usable signal existed.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
A Lemlist campaign named <name>, unsent, with every contact from list <name> loaded and a first-touch email under 90 words written for each from its latest signal in your voice, plus the signal, fact and source stored as variables on each lead. The summary shows the mix of hooks used - hiring, funding, technology - and lists the contacts whose rows carried no recent signal, so they can be given a different opener or held back.
Make it yours
- Ask for three follow-up steps as well, each shorter than the last and each referencing the same fact from a different angle.
- Deliver to Instantly, Apollo or Gmail drafts instead of Lemlist.
- Restrict the hook to one signal type - only job postings, for example - when the campaign has a single theme.
- Write two variants per contact for a subject-line and opener test, loaded as A and B in the same campaign.
Frequently asked questions
What happens to contacts with no recent signal?
They are listed separately in the summary. You can ask the agent to write from a durable fact instead - the company's stack, its size, a customer it names on its site - or hold them for a later signal-driven run.
How does it avoid sounding like a template across 200 emails?
Each email is written from its own fact, and the agent varies structure and opening deliberately; it also checks the batch for repeated sentences and rewrites any that recur. Reviewing the two samples first sets the tone for the rest.
Can I edit the emails inside Lemlist?
Yes. They are ordinary campaign steps with variables, so any email can be edited in Lemlist before the campaign starts, and the evidence variables remain available for the edit.
Does the agent ever send anything?
No. Every mission in this category ends with a draft, a paused campaign or a staged automation; sending is always your action.