The Content Engine

A voice card is the document that teaches your drafting skill how a specific author sounds. It has four components.

10 exemplar paragraphs — the author writing at their absolute best, across a range of topics. These are the few-shot examples the model learns from.

Banned phrases — words and constructions the author genuinely never uses. Knowing what to exclude is as important as knowing what to include.

Sentence-rhythm notes — observations about how long sentences tend to be, how punctuation is used, where the author speeds up or slows down.

A stance list — a set of 'we believe X' and 'we're skeptical of Y' positions that give the model a worldview to write from.

This voice card loads into every drafting task. It's the mechanism that separates recognizable content from slop. The model can only match a voice it can actually see.

A hook library is not a folder of inspiration — it's a scored database of your own hooks, with fields for the hook text, the content format, and the engagement it earned.

A learn-loop skill pulls your LinkedIn analytics weekly and updates the library with new performance data. An angle-development skill retrieves the top performers as few-shot examples when you're developing a new angle.

This turns hook quality from a one-off effort into a compounding asset. Every post you publish makes the next hook better, as long as the loop is closed.

A digital twin is a voice-matched drafting system built around a specific person. Here's how it works in practice.

Voice capture starts with a 45-minute interview and 20 writing samples, which are processed into a voice card and a stance map. The stance map captures the author's positions on the topics they write about.

Idea supply is continuous: weekly triggers arrive from the research system and from the author's own transcripts, calls, and notes.

The draft-and-approve loop is the heart of it. The twin drafts in the author's voice; the author approves or edits in a lightweight UI. Edits feed back into the voice card, improving future drafts over time.

Measurement is per-persona: engagement dashboards and hook-library learning are scoped to each individual voice.

After the setup sprint, marginal cost is near zero — which makes this structurally better retainer revenue than project work.

One pillar asset — a blog post, a webinar, a recorded conversation — should produce the following minimum set.

5–8 LinkedIn posts, each built around a single idea with its own hook and native formatting. A carousel spec: 8–10 slides, one claim per slide, designed for the Canva MCP template. One newsletter section that treats the reader as already warm. 3–5 short-video scripts at 30–60 seconds each, with a hook, the core point, and a CTA. One FAQ block structured for GEO (Generative Engine Optimization — extractable Q&A that AI engines can cite directly). And quote graphics pulled from the 2–3 most contrarian lines in the piece.

That's 90% of the value that most teams leave on the table by publishing once and moving on.