{{ai.opening_line|fallback}}.
Lines are written ahead of time and stored on the lead, not at the moment of sending. That is what makes bulk review possible — and it means a slow model can never delay a send.
1. Create the variable
Settings → AI & data → AI variables (Managers).
A good prompt is narrow:
One friendly sentence that refers to their current role and company. No flattery, no questions, no exclamation marks.The writer may only use facts that are on the profile. When there is nothing usable, the line is left blank and the fallback is used. It does not invent anything.
2. Generate lines
From AI review → Generate, or from the enrol flow. Choose the variable, then where the leads come from: a list, a tag, or the leads you selected on the Leads page. Try on one lead writes a single line and shows it to you without saving anything. Leads without an enriched profile are queued for enrichment first, so their lines arrive a little later. Existing approved lines are kept unless you tick write new lines for leads that already have one — which loses their approval.3. Review
The AI review tab holds the table: the lead, the source facts used, and the generated line.
Filter by status: pending (queued), generated (waiting for you), approved, skipped, blank (the profile had nothing usable) and failed.
The badge on the AI review tab counts the lines waiting for a person.
4. Send
Use{{ai.opening_line|Fallback sentence.}} anywhere you write copy.
In a sequence’s settings, Hold leads until AI-written lines are approved makes a lead wait rather than send the fallback. Without it, an unapproved lead simply gets the fallback and carries on.
The step preview never shows an unapproved line. If a preview shows the fallback, that is exactly what would be sent right now.