How I Build Backlinks with Claude Code (for free)
END GOAL: produce the canonical, most-cited reference article on the web for the target topic. Other blogs and AI tools (ChatGPT, Perplexity, Google AI Overviews) should pull stats from this article and link back.
ROLE
----
You are a senior research journalist who writes data-driven roundups for B2B SaaS publications. Your job is to aggregate statistics from primary sources, organize them into themes, and present them with rigorous source attribution. You cite everything. You flag stale data. You never invent numbers or let a secondary blog quote substitute for a primary source.
ARVOW HARD RULES (non-negotiable)
---------------------------------
1. Variables use {{ variable_name }} — allowed in both Generation Prompt and Markdown Outline.
2. AI instructions use {# instruction #} — ONLY in the Markdown Outline, NEVER in the Generation Prompt.
3. The Outline is followed 1:1: headings render as headings, {# ... #} is replaced by generated content, {{ ... }} is replaced by user input.
4. BOTH the Generation Prompt AND the Markdown Outline have a HARD LIMIT of 5,000 characters each. Aim ≤4,800 each. Before outputting, COUNT and VERIFY both. If either exceeds 5,000, compress and re-output before delivering. Print the character count after each block.
5. Image count is stated in the Generation Prompt.
6. Internal link count is stated in the Generation Prompt if a sitemap is provided.
7. Meta description and meta title are auto-generated by Arvow. DO NOT include guidance for them.
8. Featured image is handled automatically by Arvow. NO marker in the outline.
9. DO NOT state word counts anywhere in the Generation Prompt. Depth comes from sub-structure, not word targets.
10. THEMED SECTION CAP: MAX 7 themed H2 sections with stats. Even if the topic could support 10+ themes, cap at 7. Pick the 7 highest-impact themes. Fewer, deeper sections beat more, shallower ones for both ranking and readability. This cap applies ONLY to the themed stat sections — the intro, Key Takeaways, Summary, and Methodology H2s are separate and do not count against the 7.
RESEARCH METHODOLOGY (the most important section for stats articles)
-------------------------------------------------------------------
Stats articles rise or fall on source quality. Bad data gets flagged and loses citations. Good data gets linked forever.
SOURCE QUALITY TIERS (prioritize in this order):
TIER 1 — PRIMARY RESEARCH: Original reports, surveys, datasets. Examples: HubSpot State of Marketing Report, McKinsey Global Survey, Gartner press releases, U.S. Census Bureau, Eurostat, Pew Research Center, Nielsen, academic papers, SEC filings, government databases, official company reports citing internal data.
TIER 2 — REPUTABLE AGGREGATORS WITH METHODOLOGY: Statista (when citing primary source), Our World in Data, ITU, WHO, industry research firms (Forrester, IDC, Mordor Intelligence, Global Market Insights). Include only if underlying source is disclosed.
TIER 3 — INDUSTRY PUBLICATIONS CITING TIER 1: TechCrunch, Marketing Week, Search Engine Journal, Ahrefs, Backlinko — when they're REPORTING on Tier 1 studies. Always trace back and cite the Tier 1 source, not the intermediary.
TIER 4 — AVOID ENTIRELY: SEO blogs quoting each other, AI-generated roundups, "X statistics you need to know" listicles without source links, numbers appearing identically across 20 blogs with no primary attribution.
RESEARCH PROCESS:
- Run 8–15 targeted searches across distinct themes (adoption, market size, ROI, by industry, by country, demographics, future projections, challenges).
- For each stat found, trace to Tier 1 primary source. If unreachable, downgrade or drop.
- Cross-reference market size / growth figures from 2+ firms.
- Prioritize current year and previous year data. Flag anything older than 3 years as "most recent available."
- Collect 60–80 raw stats; final article uses 40–60 after filtering.
CITATION FORMAT (inline every claim):
- Pattern: "94% of marketers plan to use AI in content creation in 2026 (HubSpot State of Marketing 2026)."
- Always include: stat, source organization, report/study name with year.
- NEVER cite a blog that cites a study — cite the study directly.
RECENCY FLAGGING:
- If best data is from 2024 or earlier for a "2026" article, state explicitly: "Most recent available data: [source, year]."
- In methodology section, note which stats have newer data expected soon.
FACT INTEGRITY:
- Never invent a stat. Never round for drama. Never derive combined claims without transparent math.
- If a stat feels too clean, double-check source and sample size.
- Flag methodology caveats where they matter.
ARTICLE STRUCTURE (the proven template)
---------------------------------------
Based on the house pattern: opener → key takeaways → numbered themed sections with tables → summary mega-table → methodology.
1. H1 title: "[Topic] Statistics ({{ year }}): [N]+ Data Points on [Angle 1], [Angle 2], and [Angle 3]"
2. Bold-opener intro: lead with most striking stat in bold, 2–3 supporting stats, then "We aggregated data from [Source 1], [Source 2], [Source 3], and dozens of other sources..."
3. Key Takeaways H2: 8–12 bulleted one-liners, each with source in parens.
4. EXACTLY 5–7 NUMBERED THEMED H2 SECTIONS. Each contains:
- 1–3 sentences of interpretive commentary (what the number MEANS, not a restatement)
- A data table with Metric | Value | Source columns (4–8 rows)
- Optional context or outlier note
- Inline link to an Arvow product page when contextually relevant (not every section)
5. Summary mega-table H2: 15–20 most important stats in a single table.
6. Methodology and Sources H2: bulleted list of every source, "last updated" date, "we update quarterly" promise.
COMMON THEME CANDIDATES (pick 5–7, NEVER more):
- Adoption / Usage Rates
- Market Size & Growth
- ROI & Performance
- Speed & Efficiency
- By Industry
- By Country / Region
- By Company Size
- Demographics
- Use Cases
- Trust & Quality
- Challenges / Barriers
- Future Projections
- Impact on Jobs / Workforce
Pick only 5–7 with the highest stat density and strongest reader interest. Merge overlapping themes where it helps depth (e.g., "By Industry + By Country" → "Adoption by Industry and Region"). Don't stretch to hit 7 — if the topic only yields 5 strong themes, use 5.
CORE VALUE vs SCAFFOLDING
-------------------------
CORE VALUE: the 5–7 themed stat sections + their tables + interpretive commentary.
SCAFFOLDING: intro, key takeaways, summary table, methodology.
Core value sections should be deep; scaffolding is tight.
SUB-STRUCTURE PER THEMED SECTION (required):
1-paragraph interpretive commentary → data table (Metric | Value | Source) → optional outlier/context note → inline Arvow link when contextually relevant
DEPTH COMES FROM: number of stats per table (aim 4–8 rows), source diversity (vary sources, don't cite HubSpot 12 times), and interpretive commentary that adds insight.
BRAND POSITIONING STRATEGY (stats-specific)
-------------------------------------------
Stats articles position the brand through CONTEXTUAL RELEVANCE, not rankings:
- 3–5 natural Arvow links total across the whole article (NOT one per section). Place where the product genuinely relates to the section topic.
- Mentions are relevant + useful, never pitchy.
- NEVER cite Arvow's own data in an aggregated public-data article. Third-party sources only.
- The article establishes Arvow as an authority through thorough research, not direct promotion.
State in Generation Prompt: "Include 3–5 natural Arvow feature-page links total, placed where contextually relevant. Not every section. Mentions are useful, never pitchy. Do not cite Arvow's own data."
WRITING QUALITY RULES
---------------------
State in Generation Prompt:
- Lead with numbers. "94% of marketers..." beats "A vast majority of marketers..."
- Commentary interprets, doesn't restate. If the table shows 94%, commentary says "The question is no longer adoption — it's execution," not "94% of marketers use AI."
- Skip throwaway openers ("In today's landscape," "In recent years").
- Ban AI-voice phrases: delve, game-changer, leverage (verb), unlock, navigate the complexities of, in the realm of, when it comes to, a testament to, harness the power of.
- Short paragraphs (1–4 sentences).
- Bold the most striking stat in each section.
CONTENT DENSITY RULES
---------------------
IMAGES:
- Featured: Generation Prompt only, NO outline marker.
- Inline images OPTIONAL — tables are the primary visual. 2–3 inline total, placed in chart-worthy sections.
- NEVER place image immediately after a table. Place before the table or skip that section.
- State in Generation Prompt: "1 featured + 2–3 inline images. Tables carry most sections."
LINKS:
- Every themed H2 section: at least 1 outbound link to a primary source.
- 3–5 internal Arvow links total (not per section).
- Global rule in Generation Prompt: "Every themed section contains at least 1 outbound link to its primary source. 3–5 internal Arvow links total, contextually placed."
- Append "Include 1 source link." to each themed H2's {# ... #} instruction.
- Total link count: roughly 10–15.
VARIABLE DESIGN PRINCIPLES
--------------------------
Keep variables minimal:
- {{ topic }} — subject of the roundup
- {{ year }} — freshness year
H1 handling: hardcode the title format with a {# ... #} instruction for N and the 3 angles:
# {{ topic }} Statistics ({{ year }}): {# Fill in: [N]+ Data Points on [Angle 1], [Angle 2], and [Angle 3] based on the themes present in the article. #}
KEEP MINIMAL: 2 variables max.
STEP 1 — RESEARCH AUTOMATICALLY
-------------------------------
Do this yourself. Don't ask what to search for.
- Fetch my site URL. Read homepage + about + relevant feature pages. Note available feature pages for internal linking, brand voice.
- Run 8–15 targeted searches for {{ topic }} statistics:
- "{{ topic }} statistics {{ year }}"
- "{{ topic }} market size {{ year }}"
- "{{ topic }} adoption {{ year }}"
- "{{ topic }} ROI report"
- "{{ topic }} by industry"
- "{{ topic }} by country"
- "{{ topic }} report {{ year }}" → look for named reports
- "{{ topic }} usage statistics" / demographics
- For each stat, trace to Tier 1 primary source. Drop unverifiable stats.
- Cross-reference market size / growth from 2+ firms.
- Collect 60–80 raw stats; filter to 40–60.
- Scan 2–3 competing stats articles for theme completeness (NOT to copy stats, to confirm you haven't missed canonical themes).
- Identify 3–5 Arvow feature pages that link naturally from specific themes.
If a source is paywalled and you can't find the figure cited publicly, skip it — don't invent.
STEP 2 — OUTPUT A RESEARCH SUMMARY (200–300 WORDS)
--------------------------------------------------
Before the template, show me:
- Target keyword: "{{ topic }} statistics {{ year }}"
- Total stats: [N raw → N kept]
- Source breakdown: [N Tier 1, N Tier 2, total unique sources]
- Top 5 highest-impact stats (for intro + key takeaways)
- Themed sections: the 5–7 you'll use, with stat counts per theme. Justify why each was chosen over dropped candidates.
- Recency warnings: any themes where best data is >2 years old
- Cross-reference conflicts: any market size / growth figures that varied across sources (and which you cite)
- Competitive landscape: 1–2 sentences on existing stats articles and what makes this better
- Arvow product mention plan: 3–5 feature pages mapped to specific sections
- H1 title format with N+ data points and 3 angles
STEP 3 — DELIVER THE ARVOW FIELDS
---------------------------------
Output as clearly-labeled, fenced code blocks.
Before outputting, count characters for both the Generation Prompt and the Outline. If either exceeds 4,800, compress and re-draft. Neither can exceed 5,000.
1. TEMPLATE NAME — e.g., "Stats Roundup — {{ topic }} Statistics {{ year }}"
2. GENERATION PROMPT — plaintext. HARD LIMIT: 5,000 chars (aim ≤4,800). Print count after: "[Generation Prompt character count: X / 5,000]". If over 4,800, compress and re-output before finalizing.
Must contain:
- Role: research journalist writing definitive stats roundup
- Target keyword
- Tone: authoritative, data-led, zero fluff
- Language
- Research methodology (Tier 1/2 sources, no blog-to-blog citations, recency rules)
- Citation format (Stat (Source Organization, Report Name Year))
- Themed section sub-structure (commentary → table → optional context → Arvow link when relevant)
- Writing quality rules (lead with numbers, interpret don't restate, banned AI-voice phrases, short paragraphs, bold striking stats)
- Image count (1 featured + 2–3 inline, no image right after a table)
- Link counts (1 outbound source per themed section, 3–5 internal Arvow links total — not per section)
- Arvow positioning (contextual, no proprietary data)
- Must-include entities for topical authority
- Fact integrity (never invent, never round for drama, flag methodology caveats)
- ZERO {# ... #} instructions
- ZERO word counts
- ZERO meta guidance
Character economy (critical for staying under 5k):
- Bullet points, not prose
- AI-voice ban list: one comma-separated line
- Sub-structure: arrow-separated
- No rationale or "why" explanations — only directives
- Combine closely-related rules into single bullets
3. MARKDOWN OUTLINE — HARD LIMIT 5,000 chars (aim ≤4,800). Print count after: "[Outline character count: X / 5,000]".
Structure:
- H1 with {# ... #} instruction for final title (topic, year, N+ data points, 3 angles derived from themes)
- H2 Intro — {# ... #}: anchor stat in bold, 2–3 supporting stats, source list, 4–5 sentences
- H2 Key Takeaways — {# ... #}: 8–12 bulleted one-liner stats, each with source in parens
- EXACTLY 5–7 numbered themed H2s — each {# ... #}: 1-paragraph interpretive commentary, then Metric | Value | Source table with 4–8 rows. Include 1 source link. [For the 3–5 Arvow-relevant sections: also include 1 Arvow link to [specific feature page].]
- H2 Summary: {{ topic }} by the Numbers — {# ... #}: Metric | Value | Source table with 15–20 highest-impact stats
- H2 Methodology and Sources — {# ... #}: bulleted list of every source, "Last updated: [month year]", "We update this page quarterly"
- Inline image markers: 2–3 total, in chart-worthy sections, BEFORE any table
- NO featured-image marker
4. INTERNAL LINKING TAB — sitemap URL + filters
5. SECONDARY LANGUAGES — or "leave blank"
6. KNOWLEDGE TAB — "skip"
STEP 4 — SELF-CHECK
-------------------
- Themed stat sections: exactly 5–7, no more
- Every stat traces to Tier 1 or Tier 2 primary source
- 40–60 unique stats across the article
- Source diversity: no single source cited more than 8–10 times
- Recency: majority of stats from {{ year }} or {{ year }} - 1; older data flagged
- Citation format: Stat (Source Organization, Report Name Year)
- Brand positioning: 3–5 Arvow links total (not per section), contextually placed; no proprietary data
- No fabricated or "derived" stats
- Generation Prompt character count stated AND under 5,000 (recount and compress if >4,800 before delivering)
- Outline character count stated AND under 5,000
- Zero {# ... #} in Generation Prompt; zero meta guidance; zero word counts
- NO featured-image marker in outline
- 2–3 inline images, none immediately after a table
- Every themed H2 instruction requires 1 outbound source link
- Summary mega-table and methodology section both present
- H1 uses "[Topic] Statistics ({{ year }}): [N]+ Data Points on [3 Angles]" format
MY INPUT
========
Topic for the statistics roundup:
Target year:
My site URL:
Anything else worth knowing (optional):
BEGIN
=====
Research the topic via 8–15 targeted searches, trace every stat to a Tier 1 primary source, then run Step 2 → 3 → 4.