Draft Quality and Review

Every draft the platform produces is reviewed automatically before it reaches your screen. This page explains what that review does, what the occasional advisory note means, and how to raise draft quality over time.

---

The review, in brief

Each reply is checked against three things before you ever see it:

1. Your creator’s rules — banned topics, forbidden phrases, tone, reply length, pet names.
2. The facts of the relationship — what the fan has actually said, actually bought and actually been sent.
3. Message integrity — that the reply is complete, coherent, in the right language, and addresses what the fan actually asked.

Anything that fails is rewritten before delivery. In the overwhelming majority of cases you simply receive a good draft and never think about this at all.

---

What the review prevents

These are the failure modes that damage revenue and trust, and they are handled for you:

Prevented	Why it matters
Inventing facts about the fan	Referring to a girlfriend he never mentioned instantly breaks the illusion
Inventing facts about the creator	Two chatters giving two different answers about where she lives is how a persona dies
Promising things that were not agreed	A promised delivery date nobody can meet becomes a refund
Claiming content is attached when it is not	“Photos below 😘” with nothing below is the fastest possible way to lose a sale and gain a complaint
Inventing purchases or prices	Referring to a purchase that never happened, or quoting a price you do not charge
Answering the wrong person	Context from another conversation never bleeds across
Breaking your content rules	Banned topics and forbidden phrases are enforced on every reply
Replying in the wrong language	Especially when bilingual mode is active
Revealing that it is a drafting tool	Instructions never leak into the message
Incomplete or malformed messages	Nothing truncated or broken is ever delivered

---

The advisory note

Occasionally a draft arrives with a note suggesting a closer look before sending.

This is the system doing its job, not failing. It means the reply is worth a human judgement — usually because the conversation has moved somewhere that benefits from a person’s read.

When you see one:

1. Read the draft with the note in mind.
2. Edit it, or add a hint and regenerate.
3. Send when you are happy.

An advisory note is advice to a professional, not a lock: Send stays available and the judgement is yours.

---

The one thing that is a lock

A small number of drafts are held rather than advised on. A reply that came back incomplete, truncated, malformed, or with a placeholder left unfilled fails the integrity check outright and is marked Needs review. Copy, Insert and Send wait until it is dealt with, and the way out is the same in either direction: edit it and press Re-check text, or add a hint and generate again.

This is not a judgement call and there is no override — it applies to your chatters and to automatic sending alike. It is also rare: it catches a reply that came back broken, not one whose content someone might disagree with.

Note that editing any draft — held or not — puts it through the same re-check before it can be sent, and a re-check costs one draft from your allowance. See Writing Replies.

---

Held-back automatic messages

When Autopilot is engaged, some replies are deliberately not sent and are left for a person. The panel always states the reason, and the reason also lands in the Autopilot log so it can be reviewed later:

Reason shown	What to do
The fan appears to be in distress	Handle personally. Always
The fan is complaining	Handle personally — complaints escalate when automated
The fan is pushing on a banned topic	Read the conversation and decide
The relationship is too new	Introduce yourself properly
Quiet hours where the creator lives	Send it yourself, or leave it for the morning
The conversation changed while the draft was being written	Regenerate against the current conversation
The draft needs a closer look	Review, adjust, send

Nothing is lost. The draft is waiting in the panel.

With Autopilot off, nothing is drafted or sent automatically at all, so there is nothing to hold back — your team generates and sends every reply themselves. See Autopilot.

---

Raising quality over time

Draft quality is a function of what you have told the platform. Four levers, in order of impact:

1. Persona specificity

A creator whose tone of voice reads “friendly and flirty” produces generic replies. One with a backstory, a vulnerability, three catchphrases and two example exchanges produces a person.

Go to Creators and Personas and fill in the identity fields properly. This is the highest-return hour available to you.

2. Example exchanges

Up to three sample fan-message / reply pairs per creator. These outperform any amount of written description — one good example teaches more than a paragraph.

3. Per-fan instructions for the AI

One line on a fan’s profile prevents an entire category of recurring mistake:

• “He is here for conversation, not content. Do not pitch.”
• “Always confirm the price before offering.”
• “Responds to teasing, not to compliments.”
• “Never mention his ex-wife.”

4. Ratings

Rate drafts 👍 / 👎 in the dashboard, on the fan’s page — the controls sit under the draft on the Conversation tab, and Insights → Past drafts keeps the record. The results appear in your Analytics dashboard and tell your managers which creators need persona work and which are ready for more automation.

The natural moment for it is the persona check below: you are already generating drafts on real conversations and reading them, so rate them while you are there.

---

Reading the quality metrics

Your Analytics dashboard reports:

Metric	What it tells you
Thumbs up / down	Chatter satisfaction with draft quality
Conversion	The share of generated drafts your team actually used
Editing behaviour	How much your team is rewriting before sending

Conversion is the number that matters. A creator whose drafts are used 90% of the time is ready for more automation. One sitting at 40% needs persona work before anything is automated.

See Analytics.

---

Checking a change before your team feels it

After any change to a persona, generate two or three drafts yourself against real conversations — low-risk ones, not VIPs — and read them before the shift starts. Nothing is sent until somebody presses send, so the only cost is the drafts themselves.

Then let the numbers confirm it a week later: conversion per creator is the measure of whether the change worked. See Creators and Personas for the full routine.