trueadvertize-seo-geo

name: trueadvertize-seo-geo description: "Produce a 3500–4200 word blog article optimized for both Google SEO AND AI search citation (ChatGPT, Perplexity, Google AI Overviews). Uses Serper API for keyword research, loads TA's ICP + brand-voice at runtime, outputs MDX for the /blog route. Use when: writing a TA blog article, producing GEO content, drafting for trueadvertize.com/blog. Trigger phrases: write SEO article, create blog post, GEO content for TA, write a TA article." allowed-tools: Read Write Edit Bash WebSearch WebFetch AskUserQuestion Grep Glob

Article production engine for trueadvertize.com/blog. One skill, 10-step workflow, hard quality gates. Loads icp.json + brand-voice.json at runtime — never embed those files; they evolve.

The whole point: from approved topic → published-ready MDX in ~2.5 hours, optimized for both Google ranking AND AI citation. Final draft is reviewer-ready, not first-pass.

When to use

"Write a TA blog article on "
"Produce GEO content for the slug "
"Adapt the calendar's Week N topic into a draft"
Any new piece of trueadvertize.com/blog content

When NOT to use

LinkedIn personal posts — use SAM Team Engine (different voice cadence)
Cold outbound emails — use clay-alt-generate-copy
TA ad / landing / sales / webinar copy — use trueadvertize-master-copy
Client-facing copy for a TA client (Atlas, Global Hydro, etc.) — TA voice ≠ client voice
Reddit posts — use trueadvertize-reddit-post (different platform conventions)

Required inputs

Before invoking, you should have:

A topic confirmed against the calendar in CONTENT_SYSTEM.md Section 8
A slug (kebab-case, includes primary keyword)
A primary keyword (the term you want to rank for)
Optional: an angle or audience qualifier if the topic is broad

If any are missing, use AskUserQuestion to collect them in a single batch.

How

Step 1 — Load context (ALWAYS first, no skipping)

Read in parallel:

C:/Users/Juanc/Nick Saraev/trueadvertize-context/icp.json — 41-Q ICP profile (audience psychology, pain language)
C:/Users/Juanc/Nick Saraev/trueadvertize-context/brand-voice.json — voice spec (do_say / dont_say / tone / 5-step method)
Landing Page/CONTENT_SYSTEM.md — Sections 2 (voice profile), 6 (quality checklist), 8 (calendar context)

Gate: if either JSON has version != "v1.0" or status != "AUTHORITATIVE", STOP and surface — stale context produces subtly off-brand articles.

Step 2 — Load the Serper API key

export SERPER_API_KEY=$(grep SERPER_API_KEY "Landing Page/.env.local" | cut -d= -f2)

If .env.local is missing or the key is unset, STOP and ask the user.

Step 3 — Keyword research via Serper (curl)

Run these calls in parallel; capture output to content/research/<slug>.md:

Primary keyword SERP:

curl -sS -X POST "https://google.serper.dev/search" \
  -H "X-API-KEY: $SERPER_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"q\":\"<primary keyword>\",\"num\":10}"

Capture: top 10 organic results (title + URL + snippet), "People Also Ask" questions, related searches, knowledge graph entry if any.

Autocomplete expansion:

curl -sS -X POST "https://google.serper.dev/autocomplete" \
  -H "X-API-KEY: $SERPER_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"q\":\"<primary keyword>\"}"

Capture: 5–10 long-tail variations.

Recent news (trending angles):

curl -sS -X POST "https://google.serper.dev/news" \
  -H "X-API-KEY: $SERPER_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"q\":\"<primary keyword>\",\"num\":5}"

Capture: last-30-days news headlines + dates.

Top 3 competitor scrape:

curl -sS -X POST "https://google.serper.dev/scrape" \
  -H "X-API-KEY: $SERPER_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"url\":\"<top result URL>\"}"

Run once per top-3 URL. Capture: full article text, headers, key claims.

Step 4 — Synthesize the research

Write content/research/<slug>.md with:

# Research — <slug>

## Primary keyword
<keyword>

## Top 10 SERP (current)
1. [Title](url) — snippet
2. [Title](url) — snippet
[...]

## People Also Ask
1. <question>
2. <question>
[...]

## Related searches
- <variation>
[...]

## Autocomplete expansions
- <long-tail>
[...]

## Recent news (last 30 days)
- <date>: <headline> (<source>)

## Top 3 competitor analysis
### #1 — <URL>
- Word count: ~X
- Structure: <H2 list>
- Key claims: <bullet list>
- Gaps: <what they miss that TA can answer>

[repeat for #2 and #3]

## Unique TA angle
<1-paragraph synthesis: where do TA's expertise + the SERP gap intersect?>

## 3–5 citable stats
- <stat with source URL>
[...]

Step 5 — Outline

Produce content/outlines/<slug>.md:

# Outline — <slug>

## Title (≤60 chars)
<title with primary keyword, benefit-focused>

## Meta description (150–160 chars)
<single sentence, includes primary keyword, ends with hook or CTA>

## TL;DR (2–3 sentences, front-loaded)
<the synthesized answer — this is what AI tools will extract>

## H2 section plan (8–12 sections)
1. **<H2 title>** — <one-sentence summary>
2. **<H2 title>** — <one-sentence summary>
[...]

## Comparison table planned
| <Dimension> | <Option A> | <Option B> | <Option C> |
[outline only — fill in draft]

## FAQ (5–8 questions sourced from PAA + autocomplete)
1. **Q:** <PAA question> — **A:** <2–3 sentence answer>
[...]

## Key Takeaways (3–5 bullets)
- <bullet>
[...]

## Internal links planned
- → /blog/<other-slug-1>
- → /<existing-page>
[...]

## External authority links planned
- → <source 1>
- → <source 2>

Gate: before writing the draft, eyeball the outline. Does it pass the topic score from CONTENT_SYSTEM.md Section 3? If not, stop and revise.

Step 6 — Draft

Write content/drafts/<slug>.md following the locked structure from CONTENT_SYSTEM.md Section 4 Step 4. Word count target: 3,500–4,200.

Writing rules (apply as you write, not at the end):

Sentence-level:

Mix short punchy lines with longer explanatory sentences. Open most sections with a 5–10 word punchy line.
First person where appropriate ("here's what we've seen", "in our last 5 builds"). Treat the reader as a peer founder, not a search engine.
Specific numbers over vague claims. NEVER write "many", "most", "significantly" without a number.
Bold the key insight in each section (≤1 bold per paragraph).

Voice-level:

Use at least 5 phrases from do_say naturally — system / engineering rigor / allbound / partnership not outsourcing / 8–12% reply rates / etc.
Mirror back 2–3 pain phrases from icp.json Q22 ("our pipeline is cooked", "stuck scaling", "tired of dragging this thing uphill")
Wedge at least once against the primary_enemy_belief ("great product = growth")
Zero em dashes (—). Use commas, periods, colons, or middle dots.
Strip every dont_say phrase. This includes consultant filler (leverage / synergy / world-class), AI tells ("here's the thing", "it's not X, it's Y"), and salesy filler (deep dive, value prop, low-hanging fruit).

Structure-level:

TL;DR box appears BEFORE the Table of Contents.
First 100 words contain: primary keyword, key insight, author credibility marker (Samuel — former data scientist, 3 yrs building B2B SaaS GTM).
Each H2 section is self-contained (extractable by AI without needing context from above).
At least 1 comparison table OR structured list.
FAQ section uses real PAA questions verbatim where possible.

Honesty gates:

ONLY real testimonials: Dave Menjura (Clay) + Collin Stewart (Predictable Revenue). No invented client names.
All methodology numbers (8–12% reply rate, 100+ deals/6mo, 70% lower CPM, 300+ meetings) framed as targets/benchmarks: "This approach typically yields 8–12% reply rates against a real list." — NOT "Our client X got 12% reply rates." (unless cleared by Sam with attribution).
No invented case studies. If you can't cite a real source, frame as methodology.

Step 7 — Voice validation

Run two passes:

dont_say scan: grep the draft against the brand-voice.json dont_say list + the de-ai-content blacklist. Any hits → rewrite the sentence. Re-scan. Loop until clean.
fresh-reviewer pass (optional but recommended for cornerstone articles): invoke the fresh-reviewer skill on the draft. Treat its feedback as a second opinion.

Step 8 — Quality checklist

Walk the full checklist from CONTENT_SYSTEM.md Section 6. Every box must tick. Document failures in the draft's header comment and fix before continuing.

Step 9 — Convert to MDX

Move from content/drafts/<slug>.md to content/blog/<slug>.mdx. Apply the conversion rules below.

MDX frontmatter (at top of file, between --- fences):

---
title: "<title>"
slug: "<slug>"
description: "<meta description 150–160 chars>"
date: "YYYY-MM-DD"
author: "Samuel Roa"
authorBio: "Founder of TrueAdvertize. Former data scientist, 3 years building custom GTM systems for B2B SaaS founders with 50–300 customers."
authorImage: "/samuel.jpg"
primaryKeyword: "<primary keyword>"
secondaryKeywords: ["<kw1>", "<kw2>", "<kw3>"]
readTime: "<X> min read"
ogImage: "/blog/og/<slug>.png"
---

JSON-LD schema (inject via the Article page component, not in the MDX itself — see lib/schema.ts).

Body conversion:

Keep markdown headers (##, ###) — MDX renders them
Convert tables to markdown tables — MDX renders them
Use <Callout> component for TL;DR box and Key Takeaways (component to be defined in lib/blog-components.tsx)
Use <FAQ> component for the FAQ section so it can be auto-extracted into FAQPage schema
Internal links → <Link href="/blog/<slug>"> or <Link href="/<page>">

Step 10 — Open the PR

git checkout -b post/<slug>
git add content/blog/<slug>.mdx content/research/<slug>.md
git commit -m "post: <title>"
git push -u origin post/<slug>
gh pr create --title "post: <title>" --body "$(cat <<EOF
## Summary
New blog article: <title> (~<word count> words)

## Preview URL
Vercel will post it below.

## Calendar position
Week N — see CONTENT_SYSTEM.md Section 8

## Quality checklist
- [ ] Structure passes (Section 6 — Structure)
- [ ] SEO passes (Section 6 — SEO)
- [ ] GEO passes (Section 6 — GEO)
- [ ] Voice passes (Section 6 — Voice)
- [ ] Honesty passes (Section 6 — Honesty / brand)
- [ ] Technical passes (Section 6 — Technical)

## Sam's review
Open the preview, read on desktop + mobile, leave PR comments. Approve to ship.

🤖 Generated with Claude Code via trueadvertize-seo-geo skill
EOF
)"

Sam reviews on the preview URL. Approve → merge → live at trueadvertize.com/blog/.

Step 11 — Submit to indexers (manual, ~5 min)

After merge:

Google Search Console → URL inspection → Request indexing
Optionally: tweet from Sam's X (@SaamCrdens) with the article link

Output structure summary

content/
├── research/<slug>.md       ← Step 4 (Serper data + competitor analysis)
├── outlines/<slug>.md       ← Step 5
├── drafts/<slug>.md         ← Step 6 (markdown working file)
└── blog/<slug>.mdx          ← Step 9 (published source — committed)

Composition

Pre-req: trueadvertize-context/icp.json v1.0 + brand-voice.json v1.0; Serper API key in .env.local; /blog route + MDX rendering live in the Next.js app
Sibling skills: trueadvertize-master-copy (other copy types), trueadvertize-reddit-post (Reddit distribution), de-ai-content (AI pattern stripping — apply automatically), fresh-reviewer (final draft review for cornerstone articles)
Validators: quality checklist from CONTENT_SYSTEM.md Section 6; screenshot-verify on the preview URL

Anti-patterns

Don't skip Step 1. Stale icp.json or brand-voice.json produces subtly off-brand articles. If they're not at v1.0 AUTHORITATIVE, STOP.
Don't invent client names or fake case studies. Only Dave Menjura + Collin Stewart are real, attributable testimonials.
Don't claim methodology numbers as past client results. Frame as targets / benchmarks unless Sam confirms attribution.
Don't write under 3,500 words. AI tools deprioritize thin content. Below 3,500 reads as low-effort to both Google and LLM crawlers.
Don't skip TL;DR box. This is the single highest-leverage GEO element. Front-loaded answer = AI extraction target.
Don't write em dashes (—). Use commas, periods, colons, middle dots (·).
Don't write Hormozi patterns. "It's not X, it's Y" and "Not A. Not B. Not C. But D." flag as AI-generated.
Don't accept a draft that fails the dont_say scan. Loop the scan until clean. Don't ship with consultant-speak.
Don't auto-publish without Sam's review. Per CONTENT_SYSTEM.md, every article opens as a PR. Sam approves before merge.
Don't write for keywords TA can't authentically defend. If Sam couldn't take a 30-min call with a founder on this exact topic, drop it.

Source

CONTENT_SYSTEM.md Section 4 (Content Production Workflow), Section 6 (Quality Checklist)
OneAway SEO/GEO skill methodology (oneaway.io/skills/seo-geo) — extracted 2026-05-13
trueadvertize-context/icp.json + brand-voice.json v1.0 — Samuel interview merged 2026-05-11
Serper API docs — serper.dev