Blog Content Writer

---
name: blog-content-writer
description: Write publish-ready blog posts in the operator's voice — pillar or cluster length, 8-15 question-answering H2s, FAQ block, image placeholders, zero AI slop. Use when asked to write a blog post, draft a pillar or cluster article, turn a raw content dump into a post, write SEO content that doesn't sound like AI, structure a long article, or someone says 'write it like I would'. For the topic verdict first, see topical-authority-check. For metadata after, see seo-metadata-generator.
metadata:
  version: 1.0.0
---

# Blog content writer

**The writer, not the whole system.** Writing is one of thirteen steps — the
gate before it decides if the post is worth writing, the steps after decide if
it ranks. This skill's job: the topic in, a complete draft out, in the
operator's voice, structured so both readers and answer engines can lift from it.

It loads `voice-rules.md` **first, every time** — the banned phrases, the
sentence rhythm, the personal-proof patterns, the imperfect-truths rule that
keeps the copy human.

## Before you start

| Input | Why |
| --- | --- |
| A PROCEED verdict (or the operator's explicit override) | The gate runs before the writer |
| Post type: pillar or cluster | Sets length and structure budgets |
| voice-rules.md | Loaded first; no file, no draft in "your" voice |
| The keyword + question feed | H2s answer mined questions, not invented ones |
| The claims register — the numbers you may use | Every claim needs a number; every number needs a source |
| Raw dump, if one exists | Rebuilt completely — never lightly edited |

## Step 1 — Set the budget by post type

| Type | Words | Role |
| --- | --- | --- |
| **Pillar** | 2,500–3,500 | The hub: 8+ H2 sections, a named framework, everything links to it |
| **Cluster** | 1,500–2,200 | One subtopic, deep; links up to the pillar twice, across to two siblings |

**Completeness beats the counter.** Shorter than the floor usually means you
lost the plot; padding past the ceiling means two posts are hiding in one.

## Step 2 — Build the structure

The draft ships as structured, publish-ready content:

- Table-of-contents card up top; a "last updated" stamp.
- **8–15 H2 sections, each answering one question on its own** — the questions
  come from the research feed (PAA/AlsoAsked), which is what makes sections
  liftable by answer engines.
- Comparison tables where a choice is being explained.
- A callout box for the one thing the reader must not miss.
- `IMAGE_PLACEHOLDER` comments where the featured image and 1–2 idea-carrying
  infographics belong — the image step reads these later.
- An FAQ block from the remaining mined questions.
- A CTA that fits the operator's actual offer.

## Step 3 — Write in the voice, with the hard rules

- Voice file rules apply sentence by sentence — rhythm, banned list, proof
  patterns, imperfect truths.
- **No em dashes anywhere** — the most recognisable AI tell. Commas, periods,
  parentheses.
- **Brand mentioned 10–20 times, naturally** — entity reinforcement, never
  stuffing.
- **Every claim carries a number, every number a source** from the register.
  "Our CPL dropped" is noise; "dropped 23%" is evidence — if 23% was supplied.
- Current facts come from live research (Tavily-class), not training memory —
  stats need dates and sources or they don't ship.

## Step 4 — Handle a raw dump properly

"Here is a doc, publish it" means: **rebuilt completely in the operator's
voice** with the sections only they could write (their tests, their failures,
their numbers) — never republished with light edits. Plagiarism is not a risk
you take; nothing gets republished, only rewritten.

## Step 5 — Self-check before handoff

- Word count within the type's band — or a stated reason.
- Every H2 answers its one question; none are headings in search of a point.
- Zero em dashes; zero register-less numbers; zero invented links.
- Reads aloud as a person, not a summary of sources.

## Output

```
# Draft: <title>   [<pillar/cluster> · <n> words]

- TOC card · last-updated stamp
- <n> H2 sections (each ← its mined question)
- <n> comparison tables · 1 callout
- IMAGE_PLACEHOLDER × <n> (featured + infographics)
- FAQ block (<n> questions) · CTA: <which offer>

Voice: rules loaded first ✓ · banned-list pass ✓
Claims: <n>, all register-traced · Brand mentions: <n> (10-20)
Em dashes: 0 · → next: seo-metadata-generator
```

## When it breaks

| What you see | What it means | The fix |
| --- | --- | --- |
| Reads like AI wrote it | Voice file skipped or thin | Load voice-rules first, always; if the file is adjectives, rebuild it (voice-file-builder) |
| H2s are labels, not answers | Structure written before questions | Each H2 maps to one mined question |
| Post is 1,100 words of pillar | The counter beat completeness backwards | Missing sections, not missing padding — check the question feed coverage |
| A statistic has no source | Register discipline slipped | The claim goes, or the operator supplies the number |
| Brand appears 40 times | Reinforcement became stuffing | 10–20, naturally — cut to where it reads human |
| Draft quotes last year's prices | Written from training memory | Live research for anything dated; stats carry dates |
| Dump lightly edited and shipped | Republishing risk | Rebuild from scratch; only the facts survive, in your voice |

Stop when the output looks wrong — generic slop, a hallucinated statistic, a
factual error stops the pipeline and reports rather than shipping. A bad post
live is worse than a late post.

## Rules

- **Voice file first, every time**, because the voice is the difference between
  your post and anyone's post on the same keyword.
- **No em dashes, ever**, because it is the most recognisable AI tell in
  written English.
- **Every claim needs a number, every number a source**, because unsourced
  claims are noise and invented ones are poison.
- **Rebuild dumps, never edit them**, because plagiarism is not a risk you
  take.
- **Completeness beats the counter**, because the bands are diagnostics, not
  targets.

## Related skills

- **topical-authority-check** — the gate that runs before this writer.
- **voice-file-builder** — builds the file this skill loads first.
- **seo-metadata-generator** — the next pipeline step after the draft.
- **ai-content-qc** — the human gate before anything goes live.