Every message SalesMind AI writes is assembled from content you control. This page shows exactly where each piece of content goes, how the AI consumes it, and which of your instructions are enforced as rules versus read as guidance. If output does not match your intent, this is the first page to check.
The Five Content Pools
| Pool | Where to set it | How the AI consumes it | What belongs here |
|---|---|---|---|
| Sender Tone of Voice | Agent settings → Sender → Tone of Voice | Read as voice only — style, rhythm, word choice. Built from your last 70 LinkedIn messages, editable anytime. Very short profiles (under ~100 words) fall back to Brand Tone of Voice. | How you sound. Not rules, not strategy. |
| Brand Tone of Voice | Agent settings → Brand | Fallback voice when a sender profile is thin or unusable | Company-level voice |
| Persona Playbook | Agent → Personas → persona description | Distilled into a compact strategy summary (about 1,000 characters) before writing. Your content survives as strategy points, not verbatim text. | Selling strategy per persona: angles, differentiators, proof points, competitors |
| Campaign Solution | Campaign setup → solution / offering | Product truth: what you sell, to whom, and the claims the writer may use | Offering description and value proposition |
| Touchpoint Instructions | Campaign sequence → each touchpoint | Parsed into enforced, structured rules (see below), plus guidance for the writer | Rules for that specific message |
💡 The most common mistake: pasting rule documents or strategy frameworks into Tone of Voice. That field shapes style only — rules placed there are read as voice, and long content is trimmed. Put strategy in the playbook, product truth in the solution, and rules in touchpoint instructions.
Touchpoint Instructions: What Is Actually Enforced
Each touchpoint instruction is parsed into a structured contract the writer must honor. These are enforced, not suggestions:
| You write | The system enforces |
|---|---|
| "No CTA" or "no call to action" | The CTA is suppressed for that touchpoint, including link CTAs |
| "No question" (also "Sem pergunta", "Sans question") | The message may end without a question, and quality checks will not fail it for that |
| "Use exactly this text: ..." | Verbatim mode — your text is used as written, with placeholder substitution such as [first name], [company] |
| "Include this link: ..." | The link is required in the generated message |
| "Mention X" or "avoid Y" | Content beats and vocabulary guidance passed to the writer |
Instructions are understood in multiple languages (English, French and Portuguese today; quality-check exemptions also recognize short forms in German, Spanish, Italian and Dutch).
💡 Keep each rule short and explicit — one rule per sentence extracts best. "No CTA. No question. Mention the Q3 report." beats a paragraph describing your philosophy of selling.
What Is Not Enforced From Prose (Yet)
Scoring weights, send-or-skip decisions, and custom gates written as prose ("only message people who...") are treated as guidance, not guarantees. Targeting belongs in your persona filters; message rules belong in touchpoint instructions. An explicit send-decision layer with structured fields is on our roadmap for advanced governance use cases.
Quality Checks and Message Exemptions
Every message passes a quality review before sending (readability, personalization, clarity, engagement). Two exemptions worth knowing:
- Connection requests are exempt from the "must end with a question" requirement.
- Any touchpoint whose instruction says "No question" (or supported short forms) is exempt as well.
For the full system — FK readability scores, soft warnings vs hard rejections, and the retry loop — see Understanding Message Quality Checks.
Size and Distillation Reference
| Content | What happens to it |
|---|---|
| Persona playbook | Distilled to a strategy summary of about 1,000 characters, plus per-touchpoint strategy rules |
| Sender Tone of Voice | Analyzed into a voice profile; very short profiles fall back to brand tone |
| Generated messages | Target of roughly 50 words, one question maximum |
| Touchpoint instruction | Parsed into structured rules; short explicit sentences extract best |
For Technical Evaluators
If you are auditing or integrating with SalesMind AI: every touchpoint is governed by a structured instruction contract (intent, CTA type including NONE, question permission, required links, vocabulary, verbatim text). Customer rules can always make sending more restrictive. Platform-level compliance, deliverability and safety rules form a floor that campaign content cannot loosen. For programmatic access, see How To Connect To SalesMind AI API and Using Webhooks.