> ## Documentation Index
> Fetch the complete documentation index at: https://docs.synscribe.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Rewrite brief

> How to strip the AI voice from training pages while keeping the substance.

You are rewriting existing Synscribe training pages so they read like a sharp, experienced
colleague wrote them for a new teammate — not like an AI content mill. **This is a voice and
readability pass, not a restructure and not a re-research.** Keep all the facts; change how they
sound.

## Read first

`../reference/AI-WRITING-PATTERN.md` (repo path `docs/reference/AI-WRITING-PATTERN.md`) — the full
list of tells to remove. Remove every one you find.

## The tells that are worst in these docs — hunt and kill

1. **Negative parallelism.** "X is not Y — it's Z", "not because A but because B", "The question
   isn't X, it's Y." We overuse this badly. Say the positive thing directly. (e.g. "Volume is not
   the criterion — the SERP cliff is" → "We pick keywords on whether the SERP is winnable, not on
   search volume.")
2. **Em-dash addiction.** We use 10–20 per page. Get to roughly 0–2. Use commas, full stops,
   parentheses, or a reworded sentence.
3. **Bold-first bullets.** Bullets that start with a `**bolded label**:`. Rewrite as normal
   sentences, or a plain bullet, or fold into prose. (Keep bold only for the rare genuinely-scanned
   term.)
4. **Unicode arrows `→` and smart quotes.** Replace `→` with a word ("to", "then", "leads to") or
   restructure. Use straight quotes `"` `'`, never curly.
5. **Rhetorical question-answers.** "The result? Devastating." Just state it.
6. **Tricolons / rule-of-three stacking**, magic adverbs (quietly, deeply, fundamentally,
   remarkably), "serves as / stands as / represents" instead of "is", "it's worth noting",
   "here's the thing / the kicker", "think of it as", "let's break this down", grandiose stakes,
   signposted conclusions ("In summary…").
7. **The formulaic `**The stance:**` opener** on every Part 1 page — keep the idea, but open each
   page differently and naturally.

## Voice target

Plain, confident, specific. Varied sentence length (mix short and long — do NOT write one-thought-
per-line staccato). A little dry humour or personality is welcome; this is a founder-run team, not
a compliance manual. Write the way you'd explain it to a smart new hire over coffee. Don't lecture,
don't hype, don't pad. Same length or shorter — never longer.

## Preserve EXACTLY (do not touch the substance)

* **YAML front-matter** — keep all keys/values. You may fix a trope in the one-line `description`,
  but keep it one line and keep every field (`title, description, part, audience, video, status`).
* **Code blocks, especially ` ```text ` prompt blocks — VERBATIM.** These are real operator prompts
  mined from production. Do not reword, tidy, or "improve" them.
* **Video lines** `> 🎥 **Video:** [text](url)` — keep the URL and the marker; you may improve link text.
* **Cross-links** `[text](../path.md)` — keep the path exactly; you may improve the link text.
* **`❓ [needs Raymond: …]` markers — verbatim.**
* **Checklists** `- [ ]` in SOPs stay as functional checklists (don't prosify them); you may fix
  trope-y wording inside an item.
* **Tables** of real tabular data stay as tables — but strip `→` arrows and bold-first cells; keep
  them scannable.
* **Audience callouts** (`> **For AMs —** …` / `> **For self-serve —** …`) stay; clean their prose.
* **Every fact, number, example, skill name, route, and ❓.** Do not invent, drop, or soften
  information. If you're unsure, keep what's there.

## MDX-safety (these render in Mintlify)

* No bare `<` right before a letter or digit in prose — write "under 5" not `<5`, or wrap in
  backticks. No stray `{ }` in prose (wrap in backticks).
* Straight quotes only.

## How to work

Rewrite each file in place (overwrite). Do a real editing pass sentence by sentence — don't just
delete em-dashes, re-flow the sentence so it reads naturally. When you finish a file, reread it
once as a human would and make sure it doesn't sound like a robot and doesn't sound like it's
trying too hard either.