Imported from rob1-alt/refine-seo-geo (SKILL.md). Install upstream with npx skills add rob1-alt/refine-seo-geo. Copyright stays with the author.

🚀 SEO/GEO Skill (v2 — brand-aware)

A reproducible methodology for ranking on Google and being cited by ChatGPT, Claude, Gemini, and Perplexity. Adapts to whatever brand you're working on — never assumes.

🎯 What's new in v2

v1 implicitly assumed you were working on a B2B SaaS called Refine. v2 throws that assumption out:

• Auto-detects the brand from the current project (domain, package.json, CLAUDE.md, README, git remote, website crawl)
• Asks once for what it can't infer (ICP, competitors, voice constraints) and saves it to seo-geo/brand-profile.md for the rest of the engagement
• Every output uses the user's actual brand name, domain, voice, ICP, competitors — nothing hardcoded
• Adapts the playbook to the brand's archetype (B2B SaaS, agency, e-commerce, dev tool, consumer, local business, creator) — different tactics, different sources, different prompt universes
• Default output directory is seo-geo/ in the current project (or wherever the user requests) — so the deliverables ship with their codebase, not somewhere generic

🧭 When to use

• A brand wants organic acquisition through Google + AI search
• A team needs a content strategy
• A founder asks "how do I get cited in ChatGPT?"
• A marketing manager wants to systematize GEO
• A user has a blog and wants to make it perform
• Anyone with a website who wants to be visible in AI answers

Skip this skill if: the user only wants a single article with no strategy context, a one-shot keyword check, or technical SEO debugging (use direct tools instead).

⚠️ Hard rules (read before running)

1. Never hardcode a brand name. Always pull from the brand profile.
2. Never assume the brand category. Ask if not obvious from the codebase.
3. Never copy-paste examples from this file into deliverables. Examples here are illustrative only.
4. Never recommend tactics that don't fit the brand archetype. A solo creator doesn't need a G2 listing. A regulated medical brand can't farm Reddit.
5. Always save discoveries to seo-geo/brand-profile.md so the user (and future skill runs) can reuse them.
6. Always cite the user's actual competitors, not placeholder competitors from this skill.
7. Always use the user's actual domain in the article slug, internal-link map, and tracking universe.

🔍 Phase 0 — Brand context discovery (NEW in v2)

Before any SEO/GEO work, build a brand profile. This is the single source of truth for the rest of the engagement.

Auto-detection (run silently first, before asking the user)

Read these in order. Each fills part of the profile:

1. cwd / git remote → brand name candidate, repo description
2. package.json → name, description, homepage, author
3. CLAUDE.md / README.md → brand description, voice, stack
4. .env / config / next.config → production domain, locale
5. public/ favicon, og-image → visual brand cues
6. Existing /blog or /content directories → editorial voice samples
7. sitemap.xml or /robots.txt at the detected domain → page inventory
8. WebFetch on detected domain → tagline, hero copy, CTA, footer

Then ask only for what's missing

Use AskUserQuestion for the gaps. Keep it tight — never ask more than 6 questions in one batch:

Q1: What's the one-line description of your product/service?
    (auto-fill from package.json description if present)

Q2: Who is your ICP? (role + company stage + pain)
    (auto-fill from CLAUDE.md or README if mentioned)

Q3: Who are your 3-5 direct competitors? (names + websites)

Q4: What's your primary geo focus? (global / region / country / local)

Q5: Which AI/LLM platforms do you care about most?
    (default: ChatGPT, Claude, Gemini, Perplexity)

Q6: Any constraints on voice, claims, or sources?
    (e.g. regulated industry, no superlatives, no comparisons)

Save to seo-geo/brand-profile.md

# Brand Profile — [DETECTED_BRAND_NAME]

## Identity
- **Brand name**: [from detection or user]
- **Domain**: [detected]
- **One-liner**: [from package.json or user]
- **Category**: [inferred or user]
- **Archetype**: [B2B SaaS | Agency | E-commerce | Dev tool | Consumer | Local | Creator | Other]

## ICP
- **Role**: [user input]
- **Company stage**: [user input]
- **Pain**: [user input]

## Geography
- **Primary**: [user input]
- **Locales**: [detected from i18n config or user]

## Competitors
- Direct: [user input]
- Adjacent: [user input or detected from competitor analysis]

## Voice & Constraints
- Tone: [detected from existing content or user]
- Hard constraints: [user input — regulated, accuracy-bound, etc.]

## Tech context
- Stack: [from package.json]
- Existing blog path: [detected]
- Sitemap: [detected]
- GSC available: [yes/no — ask if unsure]

## AI focus
- Platforms: [user choice — default: all 4]

---
_Last updated: [DATE]. Update this file as you learn more about the brand._

This file is now the source of truth for every other phase. Every prompt, every article, every CTA must read from here.

🔑 Core principles (always apply, regardless of brand)

1. One intent per article — never two articles cannibalizing on the same long-tail
2. Long-tail > head terms — 4-7 word queries convert better and are less competitive
3. Citability before rankability — content must be easy for LLMs to extract
4. Multi-model by default — ChatGPT, Claude, Gemini, Perplexity are not interchangeable
5. Measurement first — no article ships without being added to the prompt tracker
6. Source diversity — own content + 3rd-party citations adapted to the brand archetype
7. Cluster authority — pillar + 4-6 satellites, never isolated articles

🎭 Archetype-specific tactics

The skill must adapt by archetype. Pick from the brand profile.

If the archetype isn't obvious, ask. If it's mixed (e.g. dev tool + B2B SaaS), pick the dominant one and note the secondary in the profile.

📋 The 9-phase workflow

Each phase reads from seo-geo/brand-profile.md and writes to seo-geo/.

Phase 1 — Brand brief (uses Phase 0 output)

Already done. Use as input.

Phase 2 — Long-tail keyword research

Sources to mine, adapted to the brand's archetype + geo:

• Google Search Console (if user has access — ask once, save the answer)
• Google Keyword Planner / SEMrush / Ahrefs
• Conversations: sales calls, support tickets, customer interviews
• Reddit / Quora / Hacker News (search the brand's actual niche, not generic)
• Competitor visibility gaps (run the actual competitors from the profile)
• AlsoAsked / AnswerThePublic (use the brand's actual category terms)

Selection criteria — same for every brand:

Output: 30-50 long-tails customized to the brand. Save to seo-geo/02-keywords-backlog.csv.

Phase 3 — Content cluster architecture

Group keywords into 2-4 clusters. Pick pillar topics that match the brand's category, not a generic SaaS template.

Save to seo-geo/03-content-clusters.md.

Phase 4 — Article production

Use the article skeleton from assets/article-template.md. Substitute placeholders with the actual brand profile values before writing:

• [BRAND_NAME] → from profile
• [DOMAIN] → from profile
• [ICP_ROLE] → from profile
• [VOICE] → adapt sentence rhythm/vocab to the detected tone
• [CTA_URL] → from profile (e.g. /contact, /signup, /book-call)

Apply the SEO + GEO checklist (see article template) to every article.

Save articles to seo-geo/articles/[slug].md (or directly into the brand's existing blog system if detected).

Phase 5 — Citation acquisition plan

Tier list adapted to the archetype. Don't recommend G2 to a creator brand. Don't recommend Yelp to a B2B dev tool.

Save to seo-geo/04-citation-tracker.csv, pre-filled with sources actually relevant to the archetype.

Phase 6 — Multi-model tracking setup

Build a 50-300 prompt universe using the brand's actual category terms, competitor names, and geo qualifiers. Save to seo-geo/05-prompt-universe.csv.

Phase 7 — Performance dashboard

Per-article metrics. Save to seo-geo/06-performance-dashboard.csv.

Phase 8 — Weekly measurement loop

Use assets/weekly-review-template.md.

Phase 9 — Notion workspace delivery

Generate the full package in seo-geo/notion-package/ — every CSV pre-filled with the brand's data.

📝 Article writing rules (brand-agnostic)

Every article must:

1. Open with the user's actual brand context. First paragraph mentions the brand once, naturally, with a link to a relevant page on their actual domain.
2. Use the user's competitors by name in comparison content — never placeholder names.
3. Match the user's voice detected in Phase 0 (read 2-3 of their existing pieces if they exist).
4. CTA points to the user's actual conversion URL, not /contact by default.
5. Internal links target the user's actual existing pages — don't fabricate URLs.
6. External links to authoritative sources relevant to the user's category.

If the user has no existing content (new brand), use a neutral CTA and flag it in the article output: // TODO: replace with [BRAND]'s actual conversion URL once defined.

🚨 Anti-patterns (NEVER do these)

• ❌ Generate articles that mention "Refine", "Otterly", "Profound" or any other brand the user didn't list as a competitor
• ❌ Use "/contact" as the default CTA without checking the brand's actual signup/demo path
• ❌ Recommend "publish on G2" to a brand that doesn't sell software
• ❌ Recommend "post on Reddit" to a brand in a regulated industry without flagging compliance considerations
• ❌ Default to English content for a French-only brand (check locale in Phase 0)
• ❌ Write 2500-word B2B articles for an e-commerce product page
• ❌ Suggest "Wikipedia entity" as a near-term goal for a 6-month-old brand
• ❌ Reuse the article skeleton verbatim — substitute every placeholder

📊 Success criteria

A successful engagement produces:

• ✅ A seo-geo/brand-profile.md that any teammate can read and understand the brand from
• ✅ A keyword backlog where every keyword is plausibly something the brand's actual ICP would search
• ✅ Article drafts that use the brand's voice, link to their actual pages, mention their actual competitors
• ✅ A citation plan that targets sources LLMs actually cite for the brand's category
• ✅ A prompt universe with no generic placeholders
• ✅ A Notion package the user can import and immediately recognize as theirs

If a teammate could read all the deliverables and not be able to identify which brand they belong to, the skill failed.

🗂️ Output directory layout

[user's project]/seo-geo/
├── brand-profile.md            # Phase 0 — source of truth
├── 02-keywords-backlog.csv
├── 03-content-clusters.md
├── 04-citation-tracker.csv
├── 05-prompt-universe.csv
├── 06-performance-dashboard.csv
├── 07-competitor-tracking.csv
├── articles/
│   ├── [slug-1].md
│   ├── [slug-2].md
│   └── ...
└── notion-package/
    ├── README.md               # 15-min Notion setup guide
    ├── 00-PLAYBOOK.md
    ├── 01-keywords-backlog.csv
    ├── 02-editorial-calendar.csv
    ├── 03-citation-tracker.csv
    ├── 04-performance-dashboard.csv
    ├── 05-prompt-universe.csv
    ├── 06-content-clusters.csv
    ├── 07-competitor-tracking.csv
    ├── 08-article-template.md
    ├── 09-weekly-review-template.md
    └── 10-properties-config.md

If the user has an existing content/ or _posts/ or src/blog/ directory, also offer to drop articles directly into the brand's actual blog system, in the file format they already use (Markdown, MDX, JSON, TypeScript object, etc.). Detect the format from a sample read.

🔧 Execution playbook (when this skill runs)

1. Phase 0: Discover — auto-detect, then AskUserQuestion for gaps. Save brand-profile.md.
2. Confirm — show the user the inferred profile in 5 lines. Ask "is this right?". Wait for confirmation.
3. Phase 2-3 — keyword research + cluster architecture using the profile.
4. Present plan — clusters + first 5 articles to write. Get approval.
5. Phase 4 — produce articles one by one, applying the brand voice.
6. Phase 5-7 — distribution + tracking setup, all customized.
7. Phase 8 — weekly review template (filled with the brand's actual baseline).
8. Phase 9 — Notion package delivery.

After Phase 0 confirmation, never re-ask profile questions in the same session. Read from brand-profile.md instead.

🧪 Example invocation flow (illustrative — DO NOT copy verbatim)

User: "Help me build an SEO/GEO strategy."

Skill response:

1. Reads package.json, finds "name": "their-app", "description": "...", "homepage": "..."
2. Reads CLAUDE.md, extracts a few facts about the product
3. Fetches the homepage, extracts hero + CTA + 1st nav link
4. Asks 4 questions (not 6 — already inferred 2): ICP, competitors, geo, voice constraints
5. Saves seo-geo/brand-profile.md
6. Echoes back: "Got it — you're [BRAND_NAME], a [CATEGORY] for [ICP], targeting [GEO], main competitors [LIST]. Confirm?"
7. After "yes" → runs the keyword research using the brand's actual terms

Skill response on a fresh project with no info:

1. Detects empty repo
2. Asks all 6 discovery questions
3. Cannot infer category — asks an additional clarifying question
4. Saves the profile
5. Confirms before proceeding

📜 Reference assets

🆕 Migration from v1

If you previously installed v1 of this skill: the methodology is the same, but v2 stops assuming any specific brand. Delete any old seo-geo/ outputs that mention brands you don't actually work for, and re-run Phase 0 to generate a fresh brand profile.

Author: Robin Pautigny Version: 2.0.0 Last updated: 2026-04-28 License: MIT