Composer: drafting a pool's rows with AI
Draft rows with AI writes the messages of a pool from a brief, your positioning, the brand's brief, and the plan the pool belongs to.
The dialog
- Brief: what the messages should do. In a pool built from a campaign plan it is optional; the plan's brief and this pool's job are used.
- Count: Auto follows the plan's outline for this pool, otherwise three.
- Positioning (preselected) and Brand.
- Review: three critics score each draft before you see it, and failing rows are revised from their findings. On by default; two or three more AI calls.
What it reads
The pool's existing rows, so drafts continue after them. The positioning spine: the pillar this pool carries with its messages, the audience the pool serves with its benefit ladder and the targets behind it (need, barrier, posture), the value proposition, the voice, the short sentence, vision and mission. In a pool built from a plan the pillar and audience come from the plan; otherwise every pillar and audience is given and the drafts argue one pillar per message. The plan's Campaign Brief, this pool's purpose and message outline, the other pools' jobs (so it does not repeat them), and the account's live articles, which are the only article links it may use.
With a Target chosen in the dialog (the targets come from Identity > Targets), the rows are written to that one person: who they are from their card, what they need, what blocks them.
It also reads what HubSpot shows about the account (the record at the top of Identity > Targets): who the customers are, as counts, and the portal's posture. Rich, and the rows lean on it where it and the positioning disagree about who the reader is. Messy, and only the filled fields count. Blank, and the positioning decides. The posture is logged with every draft.
The spine is settled: the short sentence, the pillar claims, the proposition, vision and mission never change in a draft. What may flex for the audience is the posture (which claim leads), the wording and examples, the level of detail, how the risk is framed, and the ask.
Baseline comes first. The Composer, Review messages and the Advisor all start from what the business confirmed in Baseline: the Product Brief (offers, who it is not for, the stakes), the approved claims, and where each pillar stands beside the alternatives. Where Baseline and the site disagree, Baseline stands.
What moved readers before is read by lead status too. Once the account has 60 days of stage movement on record, the messages after which contacts left a lead status (more often than that status's usual) are kept as examples for that status. When the pool's rows are for one lead status, the Composer reads those first, then the ones for this audience. It learns what they did at their move, never their sentences.
Authorship, first
Before the arc and before AURA, one principle: empathy is accuracy about the reader's situation, not warmth about it. Every row is written so the reader recognises their own week described more precisely than they would put it. The subject names the thing they are dealing with, in their terms; the preview adds the one fact that makes it worth opening. The drafter never crows about the sender, never apologises for writing, and never announces its own candor or brevity ("I'll be blunt", "no fluff", "I'll keep this brief"). Those phrases are counted as tells, and the Reader critic checks that the first line and the subject would be recognised by the person they are written to.
The arc
Before any row is written, the sequence is planned as beats. Each row makes one move in the reader's mind: names the situation, teaches one thing, proves it, reframes the problem, answers an objection, or asks. Each beat says what the reader believes afterwards, one step past the row before. The move shows beside each draft's name; hover it for the belief.
The review
Three critics read the first draft. They score; they never rewrite.
- Minto editor, structure: Anchor (opens on the reader's situation), One idea (answer first), Arc (makes its move, repeats nothing).
- Toulmin auditor, argument: Grounds (every claim backed by your material), Warrant (the step from grounds to claim holds), Reason (timing only with a real trigger).
- Reader, a busy operator who deletes cold mail on sight: Ask (one, plain), Reads human (no cold-email phrasing, no machine tells), Useful anyway (worth reading if they never buy).
A row fails a line only where a critic can point at the words. When anything fails, the sequence is revised from the findings and scored again, so the score you see is the score of what you get. With a positioning chosen, a row that steps outside the spine fails too. When the pool has a named audience, the Reader reads each row as the audience it is written to, using the targets under it in your positioning. If the revision would drop a row, empty a field or cut a row below 60 percent of its words, the first draft is kept and the review screen says so.
The Reader's first rule is the reflection test: a technique is allowed only if the email still works after the reader understands exactly what it did. The critics also know nine fallacies by name (made-up urgency, a false either/or, a figure with no source, flattery, "everyone is doing it", a loaded question, a straw man, sunk cost, and "after X, Y happened" with no mechanism). Each counts against one of the nine lines, never as a tenth, and a fallacy finding must quote the email's own words; one that cannot be found in the email loses its label. Sound moves stay: an analogy that fits, a concession, a worked example. The Reader also asks whether each row would hold with your site, the sender's public profile and your HubSpot evidence open beside it, and gives Useful anyway to a row that says who it is not for, never to one that claims to fit everyone.
Each draft shows Critic score n/9, what failed and, for a fallacy, the words it was found in. The number is a critic score, not a quality score, until the quality test shows the critics agreeing with people. Beside it, Specific is counted in code: numbers, names and links per 100 words, and any vague praise words ("seamless", "innovative"). Rewrite to fix writes that one row again from its findings, with your edits in it, and scores it.
Approved claims from Evidence (under Hubs) are the grounds the drafts stand on: the pool's pillar first, strongest first, customers unnamed until you allow it. A claim's link becomes a link the row may use, and each draft shows which approved claims it used.
Add the sender's Public profile link (usually LinkedIn) under Senders. A reader looks the sender up before replying, so no row may claim a role or history that page would not show.
Flags
Each draft shows anything worth checking: a link off your sites, an email address not on your domain, a phone number the material did not contain, a never-say phrase. Nothing is removed.
Written from a plan
A pool built from an Advisor plan can be written without opening this dialog: Write pools in the background on the plan page runs the same drafting and the same critics as a job, and the rows arrive here as Drafts with their rubric recorded on the plan. Open the pool to edit, rewrite or make them Ready.
Accepting
Accepted drafts become Draft rows in the pool, filed in the folder chosen under File in (the campaign's message folder by default) and recorded on the plan. Save each row's fields to make it Ready.