Marketing Scrappers · MS-AES Execution System · Client Zero Edition
The MS Authority Execution System
From MS Resource Sheet architecture to a published, interlinked, citable page — every step, every prompt, every tool. Built for the team that must ship 2–3 assets a week without quality collapse.
21 Steps
Wired to Parts 0–12
MS Brand Voice Locked
Pre-Publish Gate Built In
GEO + Entity + DEO
Hotel Proof Layer
Table of Contents
Phase 0 — Foundation (Already Decided — Do Not Reopen)
Phase 1 — Research
- Step 4 Manual SERP & Keyword Validation (No AI)
- Step 5 AI Keyword Expansion Inside the Cluster
- Step 6 Cannibalisation Check
- Step 6B Entity Verification (Knowledge Graph)
- Step 7 Semantic & PAA Expansion
- Step 8 Write the MS Brief (No Brief, No Page)
- Step 9 SERP & Competitor Analysis
- Step 10 Competitor Entity & Authority Analysis
- Step 11 Content Gap & Information Gain Lock
Phase 2 — Content Strategy & Writing
Phase 3 — Optimization
Phase 4 — Publish & Track
Phase 5 — Off-Page & Authority
Phase 0 — Foundation
The Niche, the Voice and the Page Are Already Decided. Load Them Correctly.
Best AI Tool for This Step
CHATGPT (CHATGPT)
Long context window. The MS context block + full brief + draft will not fit comfortably in smaller models, and losing context mid-draft is exactly how off-brand copy gets written.
👉 For You Load MS context at the start of every new AI session
AI has no memory of MS between chats. So Paste this first, every time in the beginning of Chat On Any Page.
📋 PROMPT — MS CONTEXT LOADER (run once per new AI chat)
// Paste this as the FIRST message in any new AI session about MS
You are the strategic content partner for Marketing Scrappers (MS).
WHO MS IS:
An SEO-first digital marketing agency founded by Hassan Shroff, based in Pakistan,
selling globally. Evolving from a service agency into an authority platform.
MS’S REAL DIFFERENTIATOR: Diagnosis, not execution. Execution is a commodity.
MS tells a business what is broken BEFORE they waste money fixing the wrong thing.
WHAT MS CAN HONESTLY CLAIM TODAY:
– Real hotel/hospitality clients with real results
– Technical SEO and WordPress speed capability
– NOTHING ELSE. MS has NO proprietary dataset yet. Never imply it does.
THE 5 CONTENT CLUSTERS (nothing is written outside these):
1. SEO & Search Visibility 2. Growth & Performance
3. Web Development & Performance 4. Content & Brand Authority
5. Local & Hospitality SEO
BRAND VOICE – non-negotiable:
“Here is the situation. Here is the evidence. Here is the opportunity. Here is the plan.”
– No hype. No guru language. No “in today’s fast-paced digital world.”
– Educate first. Show the work. Real numbers, or an honest admission of no numbers.
– Never fabricate a result, a client, a quote, or a statistic.
Confirm by replying exactly: “MS context loaded. Ready for the page brief.”
Do not write anything else yet.
📤 Save This Output — Label it “MS Context Block”
The context loader prompt, saved as a reusable snippet
You will paste this at the start of every AI session, permanently
👉 For You Input Needed — What you fill in manually
Nothing. The voice is fixed. You paste it as-is. This is the only step in the system with nothing.
Best AI Tool for This Step
CHATGPT (chatGPT)
Best at holding a specified tone across long output without sliding into generic marketing register. Paste the voice block into every session — do not assume it remembers from yesterday.
📋 PROMPT — MS VOICE ENFORCEMENT (append to every writing prompt)
// This block gets pasted into Steps 14, 15 and 16B every single time
MS BRAND VOICE — enforce strictly. This overrides your default style.
STRUCTURE OF EVERY ARGUMENT:
Here is the situation. Here is the evidence. Here is the opportunity. Here is the plan.
MANDATORY:
– Lead with the answer. Never build up to a point — state it, then explain it.
– Every claim carries a number, a source, or an honest “we don’t have data on this yet.”
– First-person practitioner voice: “When we audited a 40-room property…”
– Disqualify readers openly. Say who this is NOT for.
– Short sentences. Plain words. Explain jargon the first time it appears.
BANNED — delete on sight:
– “In today’s fast-paced digital world” and every variant
– unlock, supercharge, game-changer, secret sauce, elevate, seamless, robust
– Any claim of scale, data, or results MS has not actually produced
– Fabricated statistics, clients, or quotes
TEST EVERY SECTION:
“Could any agency in the world have written this sentence?”
If yes, it is a commodity sentence. Rewrite it with MS’s specific evidence.
📤 Save This Output — Label it “MS Voice Block”
The voice enforcement prompt, saved as a snippet
Used in Steps 14, 15 and 16B — every time, no exceptions
👤 You Do This Pull the row — do not invent it
- Open the MS Resource Sheet and the cluster map sheet.
- Go to Coda, and find your assign task, then select it from the MS Resource Sheet. It already has: Keyword, Topic Cluster, Search Intent, Global Volume, KD Level, CPC, Parent Topic, Recommended Content Type
- Copy that row into your working doc as Step 3 (1st Output) That row is your instruction set, and baseline.
- Search the primary keyword on Google, and Then Open Live Gemini
- Run The below prompt to collect the SERP analysis, GEO behavior.
📋 PROMPT — MS AUTOMATED SERP, AIO & GEO ANALYSIS ENGINE
// Run this in Gemini to instantly automate SERP observation, intent mapping, and generative search extraction for apprentices.
Act as a Senior SEO Architect, Knowledge Graph Engineer, and Generative Engine Optimization (GEO) specialist enforcing the MS-AES framework.
Your task is to analyze raw SERP data, validate search intent, spot authority barriers, and deliver a definitive technical blueprint for our apprentice content creation team.
—
REQUIRED INPUTS:
– PRIMARY KEYWORD: [INSERT PRIMARY KEYWORD HERE]
– INTENT TYPE TARGETED BY TEAM: [e.g., Transactional/Service OR Informational/Blog]
– LIVE PAGE 1 OBSERVATIONS / SERP TEXT DUMP: [GOOGLE SEARCH RESULTS, Gemini’s answer, Web Reults (title, and descriptions) are seeing, and approachable on the attached page.]
—
EXECUTION MANDATE:
Analyze the provided data inputs and execute the following 4 automated evaluation layers:
1. SERP STRUCTURE & TOP 3 ARCHITECTURE (Task 1 Automation)
– Identify the exact format dominating the organic landscape (e.g., Deep Informational Guides, Tool/Software Landing Pages, Commercial Service Pages, Interactive Tools).
– Isolate and extract the Top 3 organic ranking domains from the data.
– Provide a 1-sentence strategic breakdown explaining why Google is rewarding these specific top 3 formats for this query.
2. INTENT MATCH & AUTHORITY BARRIER DIAGNOSIS (Task 2 Automation)
– Execute an Intent Match Validation: Does our targeted intent match what Google is serving? If we target a service page but Page 1 is entirely long-form blogs, explicitly flag it: “🛑 MISMATCHED INTENT DETECTED — FLAGGED FOR ARCHITECTURE OWNER”.
– Check for “Authority Barriers”: Are massive industry players or tools (e.g., Ahrefs, HubSpot, Semrush, TripAdvisor, Booking.com) dominating the top positions?
– If authority barriers dominate, output this exact directive: “⚠️ HIGH AUTHORITY BARRIER: Abandon broad terminology. Pivot content strategy to proprietary data layers, case studies, and hyper-niche execution proof.”
3. LIVE AIO, AEO, & GEO RESEARCH ENGINE (Critical Evaluation Layer)
Look specifically for an AI Overview (AIO) or Gemini Auto-Answer within the live SERP data provided. If an AI Overview is present or indicated, execute the following audit:
– CITED DOMAINS: Map out exactly which domains the AI engine is pulling citations from. Are they identical to the top 3 organic listings, or is the AI sourcing from niche sites, forums, or secondary pages?
– RESTRUCTURING PATTERN: Define how the AI structures its auto-answer. Is it a bulleted list, a raw definition snippet, a comparative matrix/table, or a step-by-step workflow?
– GEO OPTIMIZATION VECTOR: Dictate the precise implementation framework our writer must use to intercept this generative block (e.g., “Deploy an H3 FAQ schema layer,” “Write a strict 40-word answer-first summary block at the top of the page,” or “Inject specific relational entity terms into headings”).
4. THE APPRENTICE BLUEPRINT SUMMARY
Conclude your response with a clean, scannable Markdown table organizing these findings into actionable metrics for the writer:
| Core Target Keyword | Dominant Content Format Required | AI Overview Present? (Yes/No) | Primary Action Item for the Writer |
| — | — | — | — |
📤 Save This Output — Label it “Step 3: Page Assignment”
Save The full row from the MS Resource Sheet As 1st Output. (URL, type, cluster, primary keyword, intent, parent pillar)
Save the “AIO & GEO ANALYSIS ENGINE” result as 2nd output of this Step 3.
Best AI Tool for This Step
None — this step is manual
AI cannot see the live SERP, and the page assignment comes from a MS Resource Sheet, human-approved sheet. Using AI here would reintroduce exactly the improvisation this system exists to prevent.
📋 PROMPT — MS AUTOMATED SERP INTENT & FORMAT LOCK
// Run this in ChatGPT after pasting your live search data to instantly dictate the page strategy.
Act as an SEO Strategist enforcing the MS-AES content format rules.
I am providing you with the primary keyword and the live top-10 Google search results (or a description of what dominates Page 1).
Analyze the search landscape and provide a definitive strategic playbook based on the MS matrix.
PRIMARY KEYWORD: [PASTE KEYWORD]
INTENT TYPE TARGETED BY TEAM: [e.g., Transactional/Service OR Informational/Blog]
LIVE PAGE 1 OBSERVATIONS (What kinds of pages are ranking? Are there tools like Ahrefs/HubSpot? Is there an AI Overview?):
[PASTE COPIED GOOGLE SEARCH RESULTS OR A 2-SENTENCE DESCRIPTION OF PAGE 1]
Evaluate the data and provide the following 4 tactical outputs:
- 1. INTENT MATCH: Does our targeted intent match what Google is serving? (e.g., If we want to build a service page but Page 1 is entirely long guides, state: “MISMATCHED INTENT – FLAGGED FOR ARCHITECTURE OWNER”).
- 2. AI OVERVIEW PLAN: Is an AI Overview present? (Yes/No). If Yes, remind me: “Write a strict 40-word answer-first block to be lifted.”
- 3. AUTHORITY BARRIER: Are massive players (Ahrefs, Semrush, Moz, HubSpot) dominating? If Yes, state: “HIGH BARRIER: Abandon breadth. Pivot strategy to proprietary data layers and specific hospitality proof.”
- 4. THE FORMAT VERDICT: State clearly what format this content MUST take to compete (e.g., Commercial Service Page, Deep Informational Guide, or Specific Case Study).
Phase 1 — Research
Validating the Assigned Keyword, Confirming the Cluster, and Finding What Nobody Else Has
👤 You Do This Five manual sources — work all five
- Google Autocomplete: Type the primary keyword. Record every suggestion. Then add a letter after it (“technical seo audit a”, “…b”) and collect more. This is live Google data.
- People Also Ask: Search the keyword. Expand every PAA question — more appear as you click. These become your H2s in Step 13. Copy every one.
- Related Searches: Bottom of the SERP. Copy all 8. Then search one of those and harvest its related searches too.
- Ahrefs or Semrush (ONE tool, ONE date): Pull volume, KD and CPC. Record the tool and date in the sheet header. Part 4 is explicit — one tool, one date, or the data is not comparable.
- Local volume column: Pull volume for MS’s actual target geography, not global. A 97,974-global-volume keyword MS cannot serve is a vanity metric.
You should finish with 30–50 manually sourced keywords and every PAA question for this page.
⚠ Do NOT use AI in this step. The entire point of Step 4 is a list built from live data. Mixing in AI suggestions defeats the purpose and reintroduces the exact hallucinated-keyword problem that corrupted the original file. AI gets its turn in Step 5.
💡 The AI Overview check: While you are in the SERP, note whether an AI Overview appears for this query and who it cites. Log the cited domains. This is a free data point for the AI Visibility Log (Part 9 s.1) and it costs you ten seconds.
Best AI Tool for This Step
None — fully manual
Live SERP data cannot be retrieved by a model without web access, and even then it is filtered. Your eyes on the actual results page is the standard here.
📤 Save This Output — Label it “Step 4: Manual Keyword List”
All autocomplete suggestions
All PAA questions (these become your H2s)
All related searches
Volume / KD / CPC from ONE tool on ONE date — record which and when
Local volume for MS’s target geography
AI Overview present? Which domains did it cite?
🤖 For AI Cluster-constrained expansion
AI expands what you already proved with real data. The cluster lock is the constraint that makes it safe.
📋 PROMPT — MS KEYWORD EXPANSION (cluster-locked)
// Run AFTER the MS Context Loader from Step 1. Paste your manual list.
Act as an SEO keyword specialist working inside a FROZEN content architecture.
ASSIGNED PAGE: [PAGE URL FROM STEP 3]
PAGE TYPE: [Service / Industry / Pillar / Supporting Article]
PRIMARY KEYWORD: [PRIMARY KEYWORD FROM STEP 3]
ASSIGNED CLUSTER: [ONE OF THE 5 MS CLUSTERS]
SEARCH INTENT: [Informational / Commercial Investigation / Transactional]
My manually researched keyword list (from live Google):
[PASTE YOUR STEP 4 MANUAL LIST]
HARD CONSTRAINT: This page has ONE macro-context and ONE intent. Do not suggest
keywords that would pull it toward a second topic or a second intent. If a keyword
belongs to a different MS cluster, put it in a separate “BACKLOG — DIFFERENT CLUSTER”
section rather than including it.
Expand my list with keywords I may have missed, grouped as:
1. Same-intent variations (10) — must match the page’s single intent
2. Long-tail questions (10) — 4+ words, lower competition
3. Problem/symptom phrases (5) — e.g. “traffic dropped after AI overview”
4. Entity-adjacent terms (5) — tools, standards, concepts, not synonyms
Then a separate section:
5. BACKLOG — DIFFERENT CLUSTER: any keyword that does not belong on this page,
with the MS cluster it should go to instead.
For each keyword: the phrase, the intent, and whether it is already covered by my manual list.
💡 The backlog section is the important one. Those keywords are not waste — they are evidence for whether a cluster needs deepening (Part 12 s.5). Log them. Do not write them.
Best AI Tool for This Step
ChatGPT (https://chatgpt.com)
Best at respecting a hard structural constraint like “stay inside this cluster” instead of maximising output. Most models will happily flood you with off-cluster keywords because more looks better.
👤 You Do This The three-part collision check
- Site search: Google
site:marketingscrappers.com [your primary keyword]. If an existing MS page already ranks for it, stop and escalate to the Architecture Owner. You are about to cannibalise.
🤖 For AI Collision analysis
Use AI as a second pair of eyes on overlap — but the site: search you ran above is the real evidence.
📋 PROMPT — MS CANNIBALISATION CHECK
// Run before writing. This is a go / no-go gate.
Act as an SEO architect enforcing a strict anti-cannibalisation policy.
I am verifying whether my assigned topic conflicts with any existing pages on our live site before I begin outlining or writing.
Current PRIMARY KEYWORD: [PRIMARY KEYWORD]
Current Primary INTENT: [e.g., Transactional, Informational, or Sector Proof]
LIVE SEARCH RESULT EVIDENCE:
[PASTE GOOGLE SITE Results from SERP: SEARCH RESULTS OR COPIED URL DETAILS IF AN MS PAGE ALREADY RANKS FOR THIS TERM. IF NOTHING FOUND, WRITE “NONE Found”]
Analyse the evidence and provide a definitive binary verdict:
1. COLLISION CHECK: Does an existing live page on marketingscrappers.com already own or rank for this exact keyword intent? (Yes/No)
2. VERDICT:
– PROCEED: (If no conflict exists, state “Proceed to Step 6B”)
– ESCALATE: (If a page already exists, state “STOP & ESCALATE TO ARCHITECTURE OWNER” and name the conflicting URL)
⚠ If the verdict is ESCALATE, you stop. You do not write “carefully” around a collision. You take it to the Architecture Owner and one of the two pages gets re-scoped.
Best AI Tool for This Step
Perplexity (perplexity.ai)
Strong at holding several page definitions in mind at once and reasoning about overlap rather than pattern-matching keywords. This is a judgement task, not a lookup.
👤 You Do This Pull real entities (free)
- Go to Google Knowledge Graph API Reference → “Try this API.”
- 🛠️ The Crucial Fields to Enter Value, as below guided:
- 1. limit (Highly Recommended) Type 5 or 10
- 2. types (Use for Intent Filtering) Leave it blank initially. If you get too much clutter, type a specific Schema.org type like Organization, Place, Person, or TechArticle.
- 3. languages: Type English: en Spanish: es French: fr German: de. Urdu: ur Persian: fa Arabic: ar
- 4. Query your primary keyword. Execute. Read the JSON.
- 5. Use this prompt with the JSON-LD Code as input with it: { Act as an expert SEO Entity & Knowledge Graph Engineer. I am going to provide you with a raw JSON payload from the Google Knowledge Graph Search API for an MS (Marketing/SEO) or hospitality optimization project. Your job is to cleanly parse the JSON and extract only the relevant semantic entity nodes, discarding any clutter. ### Extraction Rules: 1. Extract every entity from the `itemListElement` array. 2. For every entity, strictly capture: – Name (`result.name`) – Type (`result.@type` — list all schema types returned) – Machine ID (`result.@id`) – Confidence Score (`resultScore`) 3. Format the final output into a clean Markdown table sorted by `resultScore` from highest to lowest. ### Contextual Flagging: Highlight with a 🌟 star emoji if any of the extracted entities match or directly align with our core baseline stacks: – MS Stack: Google Search Console, Core Web Vitals, Schema.org, Ahrefs, Semrush, Screaming Frog, Lighthouse, PageSpeed Insights, Google Business Profile, WordPress, E-E-A-T, AI Overviews, Perplexity, ChatGPT. – Hospitality Stack: OTA, Booking.com, Expedia, Direct Booking, RevPAR, ADR. ### Content Mapping Guidance: Below the table, provide a quick 2-sentence summary of how the highest-scoring entities should dictate our heading hierarchy ($H2$/$H3$) or Schema markup. Do not hallucinate or guess fields not present in the JSON. Here is the raw JSON input: [PASTE YOUR JSON HERE] }
🤖 For AI Entity validation
Feed it the entities you confirmed. It maps them to sections and finds what you missed.
📋 PROMPT — ENTITY MAP FOR AN MS PAGE
// Paste your confirmed KG entities + the page brief
Act as an Entity SEO specialist working on Knowledge Graph optimisation.
MY PAGE: [PAGE URL + TITLE]
PRIMARY KEYWORD: [PRIMARY KEYWORD]
CLUSTER: [MS CLUSTER]
Entities I confirmed in Google’s Knowledge Graph / Wikipedia:
[PASTE CONFIRMED KG ENTITIES]
Please:
1. Confirm which of my entities genuinely belong on this page, and which are too
generic to add topical value (flag those for removal — do not pad).
2. Identify entities Google associates with this topic that I MISSED. Name specific
tools, standards, organisations, metrics and concepts — not synonyms of my keyword.
3. For a hospitality/local page, explicitly include sector metrics and platforms
(RevPAR, ADR, OTA, Google Business Profile, direct booking share) if relevant.
4. Produce the final ENTITY MAP as a table:
Entity | Type (tool/standard/org/concept/metric) | Why it belongs | Which H2 it goes in
This map becomes a MANDATORY writing checklist in Steps 14 and 16.
📥 Input Needed — What you fill in manually
[PAGE URL + TITLE / PRIMARY KEYWORD / CLUSTER] — Step 3
[CONFIRMED KG ENTITIES] — From the Knowledge Graph API or Wikipedia
Best AI Tool for This Step
Claude (claude.ai)
Strong entity recognition, and it will tell you when an entity is too generic rather than padding the list to look thorough. That refusal is the useful part.
📤 Save This Output — Label it “Step 6B: Entity Map”
Entity Map table (entity → type → why → which H2)
Entities removed as too generic
Sector-specific entities for hospitality pages
🤖 For AI Semantic and question expansion
This output IS your outline. Treat it as structural, not supplementary.
📋 PROMPT — SEMANTIC & QUESTION EXPANSION
// The output of this step becomes your H2 list
Act as a Semantic SEO and GEO specialist.
PAGE: [PAGE TITLE]
PRIMARY KEYWORD: [PRIMARY KEYWORD]
CLUSTER: [MS CLUSTER]
ENTITY MAP: [PASTE STEP 6B ENTITY MAP]
PAA QUESTIONS I COLLECTED FROM GOOGLE: [PASTE PAA FROM STEP 4]
Please produce:
1. QUESTION SET (10-14): Real questions a buyer or practitioner would ask, phrased the
way a human actually asks them — not keyword-stuffed headings. Merge my PAA questions
in. These will become the H2s of the page.
Order them in a logical reading sequence (what → why → how → what next).
2. For EACH question, draft the 40-WORD ANSWER. Direct. Self-contained. It must make
sense if lifted out of the page entirely — because an AI engine will lift it.
3. MISSING QUESTIONS: 5 questions the top-ranking pages fail to answer at all.
4. ENTITY PLACEMENT: which entities from my map belong under which question.
Rule: no question may be answerable by a different MS page. One macro-context only.
📥 Input Needed — What you fill in manually
[PAGE TITLE / PRIMARY KEYWORD / CLUSTER]
[STEP 6B ENTITY MAP]
[PAA QUESTIONS FROM STEP 4] — the real ones from Google, not invented
💡 The 40-word answers drafted here are the single highest-leverage output in this entire system. Part 4 calls answer-first passages “the highest-leverage GEO change available.” Part 11 says everything else is secondary to this one rule. You are drafting them at research stage, on purpose, so the writer cannot skip them later.
Best AI Tool for This Step
Claude (claude.ai) + Perplexity.ai
Perplexity for live PAA and what is currently being asked. Claude for the 40-word answer blocks — it is markedly better at writing a self-contained extractable passage than at listing questions.
📤 Save This Output — Label it “Step 7: Question Set + 40-Word Answers”
10–14 questions in reading order — these ARE your H2s
A drafted 40-word answer under each one
5 questions competitors do not answer
Entity-to-question placement map
👤 You Do This Fill every field. A blank field is a defect.
| Field | Rule |
|---|---|
| Working title | — |
| URL slug | Final. Never changed after publish. Lowercase, hyphens, 3–5 words, no stop words, no dates, no numbers. |
| Cluster | One of 5. No exceptions. |
| Parent pillar | The exact pillar URL. |
| Primary keyword | One. From the validated sheet. |
| Secondary entities | 5–10 from your Step 6B map. Entities, not keyword variants. |
| Search intent | One only. |
| Page type | Guide / How-to / Checklist / Comparison / Diagnostic / Data / Buyer Guide. |
| The one question this page answers | If you need two sentences, it is two pages. |
| 40-word answer | The most important field in the brief. Drafted now, at brief stage. From Step 7. |
| H2 question list | 6–12 questions. From Step 7. |
| MS proprietary angle | What does MS know, or have observed, that nobody else can write? If this field is empty, the article is a commodity and should be reconsidered. |
| Required links OUT | Pillar (1), siblings (2–3), service page (1+), industry page (0–1), glossary (1+). |
| Required links IN | Which EXISTING pages must be edited to link TO this page? Name them. This is how orphans are prevented. |
| Lead magnet / CTA | Exactly one. |
| Schema | BlogPosting + Breadcrumb (+ FAQPage / HowTo if applicable). |
| Author | Named human. Never “Admin.” |
| Publish date | Fixed. A Friday. |
⚠ The “MS proprietary angle” field is the one people skip. Do not skip it. If you cannot name something MS knows that nobody else can write, you are about to produce a page that competes on authority MS does not have — against Ahrefs and HubSpot, who do. You will lose. Go to Step 12B and get a real observation, or take the page back to the Content Owner.
🤖 For AI Brief assembly
Every prior step feeds this. AI assembles the brief; the Content Owner approves it.
📋 PROMPT — ASSEMBLE THE MS BRIEF
// Feed it every prior output. It assembles; you approve.
Act as MS’s Content Owner assembling a production brief.
Assemble a complete brief from these inputs. Do not invent anything not supported by them.
PAGE ASSIGNMENT (Step 3): [PASTE ROW]
KEYWORD SET (Step 5): [PASTE]
COLLISION VERDICT (Step 6): [PASTE SEPARATION LINE]
ENTITY MAP (Step 6B): [PASTE]
QUESTION SET + 40-WORD ANSWERS (Step 7): [PASTE]
Output the brief with EVERY field below filled. Where you lack information to fill a
field, write “MISSING — HUMAN INPUT REQUIRED” rather than inventing content:
Working title | URL slug | Cluster | Parent pillar | Primary keyword |
Secondary entities | Search intent | Page type | The ONE question this page answers |
40-word answer | H2 question list | MS PROPRIETARY ANGLE | Required internal links OUT |
Required internal links IN (name the EXISTING pages to edit) | Lead magnet / CTA |
Schema | Target length (estimate only) | Author | Publish date
CRITICAL: For “Required internal links IN”, name the specific existing MS pages that
must be edited to link to this new page. A page with no inbound links is an orphan,
and an orphan is a defect, not content.
CRITICAL: If “MS PROPRIETARY ANGLE” cannot be filled from my inputs, say so loudly.
That field being empty means this article is a commodity.
📥 Input Needed — What you fill in manually
All outputs from Steps 3, 5, 6, 6B and 7. This step synthesises; it does not research.
Best AI Tool for This Step
Claude (claude.ai)
Holds all five prior outputs at once and will actually flag a missing field instead of quietly filling it with plausible filler. That flagging behaviour is the point.
📤 Save This Output — Label it “Step 8: THE BRIEF” — this is the contract
The complete brief, every field filled or explicitly flagged MISSING
Signed off by the Content Owner before any writing starts
Named existing pages that will be edited to link IN to this page
👤 You Do This Manual scan of the top 3
- Google your primary keyword in a clean window.
- Open the top 3 organic results (skip ads).
- For each, record: length, structure, do they use tables, do they have FAQs, do they show real data or borrowed statistics, is there a named human author with credentials.
- The key question: is their evidence first-hand or aggregated? Almost always it is aggregated. That is the gap.
📋 PROMPT — SERP & COMPETITOR OVERVIEW
// Run in Perplexity for live data. Once per page.
Act as an SEO analyst with live web access.
Analyse what currently ranks on Google for: [PRIMARY KEYWORD]
Tell me:
1. What content type dominates page 1? (guide, service page, listicle, comparison)
2. Typical length of the top 3.
3. What H2s / subtopics appear across all three?
4. Do they use tables, FAQs, original data, or borrowed statistics?
5. Is there a named author with real credentials, or is it faceless brand content?
6. Is there an AI Overview? If so, WHICH SOURCES EARLY does it cite, and what do those
sources have in common structurally?
7. THE GAP: what does none of the top 3 have? Specifically — do any of them show
first-hand practitioner data, or are they all aggregating other people’s research?
List the 5 PAA questions currently showing for this query.
📥 Input Needed — What you fill in manually
[PRIMARY KEYWORD] — from the brief
💡 Log the AI Overview citations. Every time you note which sources an AI Overview cites and what they have in common, you are adding a row to the AI Visibility Log (Part 9 s.1) — which is the raw dataset for the Q2 research asset. MS measures its own visibility and builds its proprietary product in the same action. Do not waste the observation.
Best AI Tool for This Step
Perplexity.ai
Live web access. Claude and ChatGPT cannot see the current SERP and will confidently describe a page one from their training data. For SERP work, live access is not optional.
📤 Save This Output — Label it “Step 9: SERP Analysis”
Top 3 URLs + structure notes
AI Overview present? Which sources cited? What do they share?
THE GAP — what none of the top 3 has
Current PAA questions
📋 PROMPT — COMPETITOR ENTITY & AUTHORITY ANALYSIS
// Perplexity. Feed it the 3 URLs from Step 9.
Act as an SEO authority analyst with live web access.
Why are these 3 pages ranking for: [PRIMARY KEYWORD]
URL 1: [URL 1]
URL 2: [URL 2]
URL 3: [URL 3]
For each:
1. ENTITY DENSITY: which tools, standards, organisations, metrics do they name? Top 5.
2. AUTHORITY SIGNALS: do they cite studies or statistics? Are the numbers THEIRS or
borrowed from someone else? Be specific — this distinction matters more than any other.
3. EXPERIENCE SIGNALS: is there evidence the author actually DID the thing? First-person
accounts, screenshots from real work, client specifics? Or is it all abstract?
4. FORMAT: tables, FAQs, step-by-step, original diagrams?
5. SCHEMA: what structured data do they appear to use?
Then answer the only question that matters for us:
COMMON THREAD — what do all 3 do that earns Google’s trust?
COMMON WEAKNESS — what can NONE of them do? Specifically, do any show original
first-hand practitioner data, or are they all synthesising published research?
📥 Input Needed — What you fill in manually
[PRIMARY KEYWORD]
[URL 1], [URL 2], [URL 3] — from Step 9
💡 The answer to “common weakness” is nearly always the same for MS’s keywords: none of them have run the audit themselves. That is not a small thing. It is the entire strategic opening MS is built to exploit. Write it down explicitly every time — it feeds Step 12B.
Best AI Tool for This Step
Perplexity.ai
Requires reading live pages. Only a tool with real web access can tell you whether a statistic on a competitor page is original or borrowed.
📤 Save This Output — Label it “Step 10: Competitor Authority Analysis”
Entity lists per competitor
Are their numbers original or borrowed? (usually borrowed)
COMMON THREAD — what earns them trust
COMMON WEAKNESS — what none of them can do
👤 You Do This Answer the gate question — in writing
“What does this page contain that the top 3 results do NOT?”
Your answer must be one of these. Vagueness is a fail:
- A real observation from an MS client engagement or audit (“across the hotel sites we audited, X was broken in most of them”)
- An original framework or decision table MS uses internally
- A counter-argument the competitors avoid because it is commercially inconvenient
- Fresher or more specific data than the competitors have
- An honest admission that changes the reader’s decision (“in this situation, do not hire us — do this instead”)
If you cannot answer, do not proceed to writing. Go to Step 12B and get a real observation, or return the page to the Content Owner as a commodity.
📋 PROMPT — CONTENT GAP & INFORMATION GAIN
// Uses Steps 9 and 10
Act as a content gap analyst.
TARGET KEYWORD: [PRIMARY KEYWORD]
TOP 3 COMPETITORS: [URL 1, URL 2, URL 3]
WHAT THEY COVER WELL: [COMMON THREAD FROM STEP 10]
WHAT NONE OF THEM HAVE: [COMMON WEAKNESS FROM STEP 10]
MS’s real, verifiable assets: hotel and hospitality client work, technical SEO audits,
WordPress speed work. MS has NO large proprietary dataset yet — do not suggest angles
that require one.
Identify:
1. CONTENT GAPS: 5 specific sub-questions all three competitors miss or barely cover.
2. HIDDEN QUESTIONS: 5 long-tail questions none of the top 3 answer directly.
3. INFORMATION GAIN ANGLES: 3 angles where MS could add something genuinely new,
using ONLY assets MS actually has (hotel client observations, audit findings,
before/after speed data, honest counter-arguments).
4. THE CONTRARIAN TAKE: what is the commercially inconvenient truth about this topic
that agencies avoid saying because it costs them a sale?
5. FORMAT GAP: is there a table, decision tree, checklist or diagnostic none of them use?
Do NOT suggest anything requiring data MS does not have. Be honest about what is achievable.
📥 Input Needed — What you fill in manually
[PRIMARY KEYWORD]
[URLs from Step 9]
[COMMON THREAD + COMMON WEAKNESS from Step 10]
⚠ The Information Gain statement is a gate, not a formality. Write it into the brief in one sentence. It gets checked at the Pre-Publish Gate (Step 18B). “This page covers the topic thoroughly” is not an Information Gain statement — it is an admission that there is none.
Best AI Tool for This Step
Claude (claude.ai) + Perplexity.ai
Perplexity to verify what is genuinely absent from the live pages. Claude to reason about what is conceptually missing and to construct the contrarian take without overclaiming.
📤 Save This Output — Label it “Step 11: Information Gain Lock”
5 content gaps
The contrarian take (one paragraph)
Your written Information Gain statement — one sentence
Format gap: which table/diagnostic to build
Phase 2 — Content Strategy & Writing
Building the GEO Strategy, Extracting MS’s Proprietary Angle, and Writing the Draft
👤 You Do This The internal link plan — this is the deliverable
Part 8 is blunt: the internal links ARE the deliverable. Nothing else compounds this cheaply. Your page has a link quota by type, and it is a floor, not a target:
| Page Type | Min links OUT | Must link to | Min links IN |
|---|---|---|---|
| Service Page | 8 | Parent sub-hub, 3–4 siblings, 1–3 industries, 1 pillar, 2–3 supporting articles, 1 case study | 5 |
| Industry Page | 8 | Parent hub, 3–5 services, 1–2 case studies, 1–2 supporting articles | 4 |
| Pillar | 15+ | EVERY supporting article in its cluster, 3+ service pages, 2+ glossary terms, 1 case study | 6 |
| Supporting Article | 6 | Parent pillar (1), siblings (2–3), commercial page (1+), glossary (1+) | 2 |
| Case Study | 5 | The service it proves, the industry it belongs to, parent pillar, 2 related articles | 3 |
Below the floor = fails the Pre-Publish Gate. A service page with fewer than 8 outbound internal links is incomplete, not “lean.”
⚠ The cross-silo mandate is not optional. Every service page links to 1–3 industry pages and back. Every informational article links to at least one commercial page. Strict siloing is dead folklore — it orphans pages and traps equity. Most agency sites have a Services section and an Industries section that never speak to each other; they read as two brochures stapled together. MS explicitly rejects that.
📋 PROMPT — GEO STRATEGY & INTERNAL LINK PLAN
// Synthesis step. Feed it Steps 8, 10, 11.
Act as a GEO strategist working inside MS’s frozen architecture.
THE BRIEF: [PASTE STEP 8 BRIEF]
COMPETITOR COMMON THREAD: [STEP 10]
INFORMATION GAIN STATEMENT: [STEP 11]
MY PAGE TYPE + LINK QUOTA: [e.g. Supporting Article: 6 out, 2 in]
AVAILABLE MS PAGES TO LINK TO (paste from the sitemap — only pages that EXIST):
[LIST OF LIVE MS URLs: pillar, siblings, services, industries, glossary, case studies]
Produce:
1. INTERNAL LINKS OUT — meet the quota exactly. For each:
– Target URL
– Exact anchor text (descriptive; never “click here” or “read more”)
– Which section of my page it sits in, in body copy — NOT in a related-posts widget
– Anchor variation: do not use identical anchor text for the same target twice
MANDATORY: at least one link to a commercial page (service or industry).
2. INTERNAL LINKS IN — name the EXISTING MS pages that must be EDITED to link to this
new page, with the anchor text to use. This is how orphans are prevented.
3. CITATION STRATEGY — 4-5 genuinely authoritative external sources to cite
(Google documentation, peer-reviewed research, primary industry data).
No vendor blogs presented as research.
4. EXTRACTABILITY — for each H2, confirm the 40-word answer block is self-contained
and would still make sense if an AI engine lifted it out of the page entirely.
5. ORIGINAL MEDIA PLAN — one comparison table, one diagnostic/decision table, and one
original screenshot or data visual MS can actually produce from its own work.
Rule: do not invent MS pages that do not exist. Only link to URLs I gave you.
📥 Input Needed — What you fill in manually
[STEP 8 BRIEF]
[STEP 10 COMMON THREAD], [STEP 11 INFORMATION GAIN]
[LIST OF LIVE MS URLs] — only pages that actually exist. Linking to an unbuilt page creates a broken link, and Part 5 forbids publishing with broken links.
Best AI Tool for This Step
Claude (claude.ai)
Holds the brief, the quota, the competitor analysis and the live URL list simultaneously, and will respect “only link to URLs I gave you” instead of inventing plausible-looking internal links. Invented internal links are a real and common failure.
📤 Save This Output — Label it “Step 12: GEO + Link Plan”
Internal links OUT with exact anchors and placement — meeting the quota
Internal links IN — named existing pages to edit
4–5 authoritative external citations
Original media plan (table, diagnostic, screenshot)
👤 You Do This Pull one real observation from the six capture sheets
Before writing, you must contribute at least one of these to the page. Pull it from the sheets — do not invent it.
| Source Sheet (Part 9 s.1) | What you can pull | What it sounds like on the page |
|---|---|---|
| Audit Log | The 10 most common defects across audited sites; CWV scores before; indexation ratio; schema coverage | “Across the hotel sites we audited, the booking path was blocked from crawling on most of them.” |
| Client Baseline / Delta | Before/after rankings, traffic, speed, indexed pages | “A 40-room property went from LCP 6.1s to 1.9s. Here is exactly what we changed, in order.” |
| Experiment Log | Hypothesis → change → result → confidence | “We tested X on two properties. It worked on one. Here is why we think it failed on the other.” |
| AI Visibility Log | Which engines cited whom for a given query | “We ran this query across four AI engines monthly. Here is who got cited and what they had in common.” |
| Case Study Intake | Full before/after, timeframe, what worked, what did not | A named result with a real number and an honest caveat. |
⚠ If the sheets are empty, the sheets are the emergency — not this page. If you cannot pull a single real observation because nobody has been logging client work, stop and go build the six sheets today. Every engagement that closes unlogged is proprietary data destroyed permanently. That is a bigger loss than any single blog post. Build them, then come back.
💡 Anonymous with numbers beats named with vagueness. “A 40-room boutique property in Lahore” is more credible than “a leading hospitality brand.” You do not need client permission to publish an anonymised, accurate observation — you need permission to name them. Publish the number, anonymise the client.
📋 PROMPT — MS PROPRIETARY ANGLE EXTRACTION
// AI helps you PLACE the observation. It does not invent it.
Act as a content differentiation strategist.
PAGE TOPIC: [PAGE TITLE / PRIMARY KEYWORD]
WHAT COMPETITORS ALL LACK: [COMMON WEAKNESS FROM STEP 10]
MS’s REAL, VERIFIED OBSERVATIONS available for this page (pulled from our client sheets):
[PASTE YOUR ACTUAL OBSERVATIONS — real numbers, real findings, anonymised clients]
Help me:
1. Which of my real observations is strongest for THIS page, and why?
2. Exactly which section should it anchor? It should carry the article’s authority,
not sit in a box at the bottom.
3. How do I present it honestly — including its limitations (small sample, single sector)?
Stating limitations increases credibility. It does not reduce it.
4. What is the contrarian, commercially inconvenient truth this observation supports?
5. What ONE table or visual would make this observation maximally citable by an AI engine?
HARD RULE: Do not invent observations, numbers, clients or quotes. Work only with what
I gave you. If what I gave you is too thin to carry the page, say so directly and tell me
what I would need to capture to fix it.
📥 Input Needed — What you fill in manually
[PAGE TITLE / PRIMARY KEYWORD]
[COMMON WEAKNESS FROM STEP 10]
[YOUR REAL OBSERVATIONS] — from the six capture sheets. If this bracket is empty, you cannot proceed.
Best AI Tool for This Step
Claude (claude.ai)
Will tell you when your evidence is too thin to carry a page rather than dressing it up. That refusal is exactly what you need here — a model that enthusiastically fabricates a plausible client anecdote is actively dangerous to MS.
📤 Save This Output — Label it “Step 12B: MS Proprietary Angle”
The real observation, with its real number
Its honest limitations, stated
Which section it anchors
The table or visual that makes it citable
Confirmation it came from a capture sheet, not from imagination
👤 You Do This Answer all three yourself
- The contrarian take. The commercially inconvenient truth agencies avoid because it costs them a sale. MS’s voice permits this — in fact it requires it. Example: “Most ‘SEO audit’ offers are lead-generation forms with a PDF attached. If the audit does not tell you something that changes what you spend money on next month, it is not an audit.”
- The original contribution. Your real observation from Step 12B, written as one clear paragraph, with its number and its limitation.
- The real data asset. A screenshot from actual client work, a before/after table you built from real numbers, a decision table MS genuinely uses internally. Something that exists.
💡 Why the contrarian take matters commercially: MS’s audience is people who have been burned by agencies. The single fastest way to earn their trust is to say the thing the last agency would not say — including “in your situation, do not buy this.” Part 3 requires a “who this is NOT for” section on every service page for exactly this reason. Honest disqualification raises lead quality and it builds trust.
⚠ Do not let AI write the contrarian take from scratch. It will produce something that sounds contrarian but is safe — “SEO takes longer than people think.” That is not contrarian, it is a cliché. A real contrarian take costs you something. If it does not make a salesperson slightly uncomfortable, it is not one.
📤 Save This Output — Label it “Step 12C: Human Injection Brief”
Contrarian take (1–3 sentences, written by a human)
Original contribution paragraph (from Step 12B, real numbers)
The real data asset — described and located
All three of these get pasted into Step 14
👤 You Do This The structural rules, non-negotiable
- One H1. Contains the primary keyword.
- Answer-first block directly under the H1: 40 words, before any other content.
- Every H2 is a real question, phrased the way a human would ask it — from your Step 7 question set.
- Under every H2: a 40-word answer. Self-contained. It must make sense lifted out of the page entirely, because it will be.
- At least one table. Tables are the most extractable format there is.
- Then elaboration, context, nuance.
Never build up to a point. Lead with it. Context comes second, always.
📋 PROMPT — ANSWER-FIRST OUTLINE
// Start with the pillar if you have one. Feed everything.
Act as a GEO content architect building an answer-first outline.
BRIEF: [PASTE STEP 8 BRIEF]
QUESTION SET + 40-WORD ANSWERS: [PASTE STEP 7]
ENTITY MAP: [PASTE STEP 6B]
CONTENT GAPS: [PASTE STEP 11]
HUMAN INJECTION BRIEF: [PASTE STEP 12C]
LINK PLAN: [PASTE STEP 12]
Build the outline:
1. H1 — contains the primary keyword. Under 60 characters.
2. ANSWER-FIRST BLOCK — 40 words, directly under H1, before anything else.
This is what an AI engine lifts. Write it to be lifted.
3. H2 SECTIONS — every H2 is a REAL USER QUESTION, in logical reading order.
Under each H2:
– The 40-word answer block (self-contained, extractable)
– 2-3 bullets on what the elaboration covers
– Which entities from the map appear here
– [INTERNAL LINK: target + exact anchor] where the link plan places one
– [CITE: source] where an external citation belongs
– [TABLE: what it compares] — at least one across the page
4. [HUMAN INJECTION HERE] — mark exactly where the contrarian take and the real MS
observation anchor the page. This is NOT a box at the bottom. It should sit where the
article’s authority is established.
5. FAQ H2 at the bottom — 6-10 real PAA questions, FAQPage schema.
6. ONE CTA — appropriate to the intent. Not three.
RULE: if any H2 is not phrased as a question, rewrite it until it is.
RULE: an informational page does not sell. A service page does not teach.
📥 Input Needed — What you fill in manually
[STEP 8 BRIEF], [STEP 7 QUESTIONS], [STEP 6B ENTITIES]
[STEP 11 GAPS], [STEP 12C INJECTION BRIEF], [STEP 12 LINK PLAN]
⚠ Check the injection point placement before you write. If [HUMAN INJECTION HERE] has been parked in the last section, move it. The real observation is the article’s authority — it belongs where the reader is deciding whether to trust you, which is usually the second or third section, not the conclusion.
Best AI Tool for This Step
Claude (claude.ai)
Produces the most logically ordered outlines and reliably enforces “every H2 must be a question” instead of drifting back to noun-phrase headings after the first few.
📤 Save This Output — Label it “Step 13: Outline”
Full outline: H1, answer-first block, question H2s with 40-word answers
Entity placement per section
Internal link and citation markers
[HUMAN INJECTION HERE] placed correctly
FAQ questions + the single CTA
📖 Why This Exists Brief the model completely, or it writes a commodity
If you tell AI “write about technical SEO audits,” it writes what every other agency published. If you tell it who MS is, what the competitors lack, what MS observed in a real hotel audit, and hand it a validated outline — you get something that could only have come from MS.
This step is the difference between good and unusable. Send it as its own message. Do not combine it with the writing instruction.
📋 PROMPT — PRE-WRITING CONTEXT BRIEF
// Send this FIRST. Then the writing instruction in the SAME chat.
You are about to write a page for Marketing Scrappers. I am briefing you completely
first. Do not start writing until I give the writing instruction.
=== MS CONTEXT ===
[PASTE THE MS CONTEXT BLOCK FROM STEP 1]
=== MS BRAND VOICE ===
[PASTE THE MS VOICE BLOCK FROM STEP 2]
=== THE PAGE ===
URL: [PAGE URL]
Page type: [TYPE] — its ONE job: [SELL / PROVE SECTOR FLUENCY / TEACH / SOLVE ONE NARROW PROBLEM]
Primary keyword: [KEYWORD]
Single intent: [INTENT]
Cluster: [CLUSTER]
=== WHAT IT MUST NEVER DO (cannibalisation map) ===
[e.g. “This is the SERVICE page. It must NEVER teach hotel SEO — that is
/guides/local-seo/. It sells. It shows deliverables, process, proof, price signal.”]
=== ENTITIES TO COVER (mandatory) ===
[PASTE STEP 6B ENTITY MAP]
=== WHAT COMPETITORS ALREADY DO (do not just repeat it) ===
[PASTE STEP 10 COMMON THREAD]
=== WHAT NONE OF THEM CAN DO ===
[PASTE STEP 10 COMMON WEAKNESS]
=== INFORMATION GAIN — why this page deserves to exist ===
[PASTE STEP 11 STATEMENT]
=== HUMAN INJECTION BRIEF — the core anchor of this page’s authority ===
Contrarian take: [STEP 12C]
MS’s real observation: [STEP 12B — real numbers, real limitations]
Real data asset: [STEP 12C]
=== EXTERNAL SOURCES TO CITE ===
[PASTE STEP 12 CITATION LIST]
=== THE OUTLINE — follow it exactly ===
[PASTE STEP 13 OUTLINE]
=== END OF BRIEF ===
Confirm by replying exactly:
“Brief received. MS voice loaded. Ready for the writing instruction.”
Do NOT start writing.
📥 Input Needed — What you fill in manually
Every output from Steps 1, 2, 6B, 10, 11, 12, 12B, 12C and 13. This is the synthesis point. If a bracket is empty, go back and fill it — do not proceed with a partial brief.
💡 Keep the same chat window open into Step 15. The model needs this brief in context when it writes. A new tab means a new model with no memory of any of it, and you will get generic copy back.
Best AI Tool for This Step
Claude (claude.ai)
Largest usable context window. This brief is long — context block, voice block, entity map, competitor analysis, injection brief, full outline — and truncation here silently degrades everything downstream.
📤 Save This Output — Label it “Step 14: Context Brief Sent”
Confirmation the model acknowledged the brief
The chat window stays open into Step 15
📖 Why This Exists The draft — and what it is not
AI writes the draft. AI does not write the final page. Part 7 assigns the Writer the research, outline and draft — and specifically says the Writer “owns the proprietary angle.” That part is human. The draft is a starting point that a human then makes true.
📋 PROMPT — WRITE THE DRAFT
// Send in the SAME chat as Step 14, after the confirmation.
Now write the full page, following the brief and outline exactly.
MS WRITING RULES — enforce all of them:
1. ANSWER FIRST, ALWAYS. Every section opens with the answer. Context comes second.
Never build up to a point — lead with it.
2. 40-WORD ANSWER BLOCKS. Directly under each H2 question. Self-contained. It must make
complete sense if lifted out of the page entirely — because an AI engine will lift it.
3. EVERY H2 IS A QUESTION. If a heading is not a question, rewrite it.
4. BRAND VOICE: “Here is the situation. Here is the evidence. Here is the opportunity.
Here is the plan.” No hype. No guru language. Short sentences. Plain words.
5. HUMAN INJECTION: weave the contrarian take and MS’s real observation into the section
marked [HUMAN INJECTION HERE]. Do NOT generalise it. Do NOT soften the numbers.
Include its stated limitations — that increases credibility, it does not reduce it.
This is the core anchor of the page’s authority.
6. FIRST-PERSON PRACTITIONER VOICE: “When we audited a 40-room property…” This is the
Experience in E-E-A-T. Google’s raters look for it. So do LLMs choosing what to cite.
7. ENTITIES, NOT SYNONYMS. Name the tools, standards, metrics, organisations from the
entity map. Entity coverage is what topical authority is actually made of.
8. AT LEAST ONE TABLE. Comparison, summary, diagnostic, or checklist.
9. SHOW THE WORK. Real numbers where we have them. Where we do not, say so honestly —
never fabricate a statistic, a client, or a quote.
10. NO FLUFF INTROS. “In today’s fast-paced digital world” — never. Delete on sight.
11. ONE INTENT. [If informational: this page does not sell. If commercial: this page does not teach.]
12. LINKS: where the outline marks [INTERNAL LINK: target + anchor], write natural
body copy around that exact anchor. Never “click here” or “read more.”
Where it marks [CITE: source], write the sentence and mark [CITATION NEEDED: X].
13. LENGTH: [Tier 1 service 1,500-2,500 / Tier 2 1,000-1,500 / Pillar 3,000-5,000 /
Supporting article per its type]. Length is an OUTPUT of covering the topic properly,
not a target. Do not pad.
14. ONE CTA. Not three.
Begin. Follow the outline exactly.
📥 Input Needed — What you fill in manually
[TARGET LENGTH] — Tier 1 service: 1,500–2,500. Tier 2: 1,000–1,500. Tier 3: 800–1,200. Pillar: 3,000–5,000.
[INTENT REMINDER] — one line, stating what this page must not do
⚠ Never publish the draft as-is. Check these six things before Step 16: (1) Is the MS observation intact, with its real number and its limitation — or did the model soften it into a generic claim? (2) Did it invent any statistic, client or quote? Search for every number and verify it. (3) Is every H2 still a question? (4) Does every H2 have a 40-word answer block? (5) Is there at least one table? (6) Did it stay in one intent — or did the informational page start selling?
💡 The most common failure at this step is the model quietly generalising MS’s specific observation. “A 40-room property went from LCP 6.1s to 1.9s” becomes “clients often see significant speed improvements.” That is a downgrade from evidence to marketing. If it happens, paste the observation back in and tell it to restore the specifics verbatim.
Best AI Tool for This Step
Claude (claude.ai) — same window as Step 14
It already holds the full brief. A new chat, or a different model, loses the context you spent nine steps building — and you will not notice until the draft reads like every other agency blog.
📤 Save This Output — Label it “Step 15: Draft”
The full draft
Every [CITATION NEEDED] flagged for a real URL
Every [INTERNAL LINK TO:] marker
Every [TABLE:] and [VISUAL:] marker — this is your media production list
Verified: no fabricated numbers, clients or quotes
Phase 3 — Optimization
On-Page, Entity Coverage, Voice Repair, Schema — and the Gate That Stops Bad Pages
👤 You Do This Manual placement checklist
- Primary keyword in the H1
- Primary keyword in the URL (already locked at brief stage — never change it now)
- Primary keyword in the title tag
- Primary keyword in the first 100 words — naturally
- Every entity from the Step 6B map appears at least once
- Every H2 is a question, and every H2 has a 40-word answer under it
- The MS observation is present, with its real number and its limitation intact
- At least one table exists
- No keyword stuffing — it reads like a practitioner wrote it, not an optimiser
📋 PROMPT — ON-PAGE, ENTITY & ANSWER-BLOCK CHECK
// Paste the full draft. Targeted fixes only — no rewrite.
Act as a Semantic SEO editor. Review this draft. Do NOT rewrite it — give targeted fixes.
PRIMARY KEYWORD: [KEYWORD]
ENTITY MAP (must all appear): [STEP 6B]
INFORMATION GAIN STATEMENT: [STEP 11]
DRAFT: [PASTE FULL DRAFT]
Check and report:
1. KEYWORD PLACEMENT: is it in the H1, first 100 words, at least one H2? If not, show
exactly where to add it naturally.
2. MISSING ENTITIES: which mapped entities are absent? For each, give one natural
sentence to add and say which section it belongs in.
3. ANSWER BLOCKS: for EVERY H2, does a self-contained 40-word answer sit directly under
it? Flag every H2 that fails. Rewrite the failures so they would survive being lifted
out of the page by an AI engine with no surrounding context.
4. THE MS OBSERVATION: is it still specific — real number, real limitation? Or has it
been generalised into a vague claim? If it has been softened, flag it. This is the
most important check in this step.
5. INTENT PURITY: does an informational page try to sell anywhere? Does a commercial
page drift into teaching? Quote any section that breaks intent.
6. FABRICATION AUDIT: list EVERY number, statistic, client and quote in the draft.
For each, state whether it is traceable to my inputs. Flag anything invented.
7. Score on-page coverage 1-10 and name the single highest-impact fix.
📥 Input Needed — What you fill in manually
[KEYWORD], [STEP 6B ENTITY MAP], [STEP 11 STATEMENT]
[FULL DRAFT FROM STEP 15]
⚠ Check 6 — the fabrication audit — is the one that protects MS. A single invented statistic on marketingscrappers.com is worse than a thin page. MS sells diagnosis and honesty; publishing a made-up number destroys the exact thing being sold. Verify every figure against your source or delete it.
Best AI Tool for This Step
Claude (claude.ai)
Gives targeted, surgical edits without rewriting the whole draft — and will actually flag its own earlier fabrications when asked to audit them.
📤 Save This Output — Label it “Step 16: Optimized Draft”
Entities added and where
Answer blocks repaired
Fabrication audit — every number verified or removed
MS observation confirmed intact and specific
📖 Why This Exists The goal is not to hide AI — it is to sound like MS
AI-assisted drafts leave patterns: uniform paragraph lengths, “Furthermore,” and “Moreover,” every section with exactly three bullets, headings all in the same grammatical shape, and a relentless neutral positivity that never commits to anything.
MS’s voice is the opposite of that. It is specific, blunt, and willing to say the inconvenient thing. This pass is not cosmetic humanisation — it is restoring the brand voice the model sanded off.
📋 PROMPT — AI FINGERPRINT CLEANUP & MS VOICE RESTORATION
// Run on the optimized draft from Step 16
Act as MS’s senior editor.
MS BRAND VOICE: [PASTE THE MS VOICE BLOCK FROM STEP 2]
DRAFT: [PASTE STEP 16 OPTIMIZED DRAFT]
Find and fix:
1. Robotic transitions: “Furthermore,” “Moreover,” “In conclusion,” “It’s important to note”
2. Uniform rhythm: every paragraph the same length, every section the same shape
3. Hedging that commits to nothing: “can often help”, “may potentially improve”,
“there are many factors to consider”
4. Generic sentences that ANY agency could have written — quote each one
5. Headings all following the same grammatical template
6. Neutral positivity where MS’s voice requires a position
7. BANNED WORDS: unlock, supercharge, game-changer, seamless, robust, elevate,
“in today’s fast-paced digital world”
For each issue:
– Quote the exact text
– Say why it reads as machine-written or off-brand
– Rewrite it in MS’s voice: situation → evidence → opportunity → plan
– Vary sentence rhythm: mix short punchy sentences with longer ones
PRESERVE ABSOLUTELY: all entity mentions, all keyword placements, all 40-word answer
blocks, all internal link anchors, and — above all — the MS observation with its exact
numbers and its stated limitation. Do not soften it. Do not round the numbers.
Then answer: does this page now sound like it was written by someone who has actually
run this audit? Or does it still sound like a well-optimised article about running one?
📥 Input Needed — What you fill in manually
[MS VOICE BLOCK FROM STEP 2]
[STEP 16 OPTIMIZED DRAFT]
💡 Read the rewrites out loud. If a “humanised” sentence sounds more forced than the original, keep the original. The goal is natural variation and a real position — not performed personality. You are the final judge.
Best AI Tool for This Step
Claude (claude.ai)
Best at recognising machine-writing patterns in existing text and rewriting with genuine stylistic variation, rather than swapping synonyms and calling it humanised.
📤 Save This Output — Label it “Step 16B: Final Draft”
The voice-restored draft — this is now the writing you publish
Confirmation the MS observation still carries its exact numbers
📖 Why This Exists The title tag is the ad. The meta is the offer.
Title tag: under 60 characters, contains the primary keyword. Meta description: under 155 characters — and per Part 5, it contains the outcome and a differentiator, not a summary. Both written deliberately. Auto-generated metas fail the gate.
📋 PROMPT — META, ALT-TEXT & VISUAL BLUEPRINT
// After Step 16B
Act as an SEO copywriter for MS. Brand voice: no hype, outcome-led, honest.
H1: [FINAL H1]
PRIMARY KEYWORD: [KEYWORD]
PAGE TYPE: [TYPE]