Going global · Grow
The SEO water-system model
An SEO framework for any site: 14 components in the order water flows, each with its rules and the one number to watch. Effort goes first to the hardest inflows that AI cannot do for you, not to the easiest work of spinning up pages.
The SEO System: Water System Model v8 (14 Components · 5-Layer Meters · 6 Release Gates)
v8 tested · v9 pendingBuild SEO as a water system: you need inflows before pipes can carry water, and effort goes first to what is least automatable (human-led authority), not to what is easiest (spinning up pages). Every rule from v4 to v8 comes from a mistake we made: three months of practice (v5), Google Search Central's technical gates (v6), Ahrefs' demand catchment (v7) and the six gates from Gefei's tutorials (v8). The 14 components are numbered in the order water flows:
Water in
Store
Carry
Water out
Look back
↺ 14 Iteration reads the five meters and decides which inflow to open or which pipe to fix next, then back to 1.
1Catchment · Demand & Commercial Proof
v7/v8 · 🤖 lookup · 🧑 decideWhere the rain falls: record Search Intent, Traffic Potential (#1 page total traffic), and Business Value (0–3 score backed by monetization proof).
📊 Water meter: % of target clusters with intent + traffic potential + revenue proof
2Inflow · Link Authority
v5/v8 · 🧑 Human-ledTwo tiers: the weight tier is earned coverage and citable datasets (e.g. a Zenodo DOI); the discovery tier is free directories, which trigger indexing but pass no weight. No paid dofollow links since Oct 6; v8 gate: start links only once indexing grows steadily. Before pitching, `backlink-guard ai-access <domain>` shows whether a site blocks AI crawlers (a reference, not a score).
📊 Water meter: Authoritative referring domains
3Inflow · GEO & AI Citations
v5/v7 · 🤖 prep · 🧑 shipBeing mentioned > being linked (YouTube mentions r=0.737 >> backlinks 0.28). v7: llms.txt is optional; 301-redirect hallucinated AI 404 URLs monthly.
📊 Water meter: AI sessions + brand mentions + AI visits landing on 404s
4Inflow · Native Platform Search
v5 · 🤖 AIInternal search engines of GitHub, npm, PyPI, Chrome Web Store, Toolify, and Product Hunt (repo-name keywords, all 20 GitHub topics filled, structured READMEs, daily auto-updated awesome repos with domain attribution, plus CLI/MCP packages).
📊 Water meter: In-platform rank, installs & GitHub repo views
5Reservoir · Brand Entity
v5 · 🤖 schema · 🧑 namingWeight accumulates in your brand entity first.
📊 Water meter: Brand-term position + sitelinks
6Pipe Material · SSR & Status Codes
v6 · 🤖 AIVerify Google reads what you see: (1) monthly GSC Live URL Test per template to confirm body copy and links exist in initial/rendered HTML; (2) separate URLs + bidirectional hreflang per language (never JS/cookie text swapping); (3) real <a href> links; (4) true 404/410 for missing pages (no SPA soft-404s).
📊 Water meter: Live URL test pass rate + 404 / soft-404 / redirect errors
7Hierarchy · Site Architecture
v5 · 🤖 AIArchitecture before scale: structure as hub → cluster → leaf and enforce one keyword cluster per URL.
📊 Water meter: Click depth (≤3 hops)
8Flow · Internal Link Routing
v5 · 🤖 AIRoute authority from link-earning pages (home/reports) to money pages (/best/ scenarios) using keyword-rich anchors in server-rendered HTML. Orphan pages = broken pipes; never let search/tag filters spawn crawlable URLs.
📊 Water meter: Orphan page count (target 0) & SSR internal link coverage
9Sub-Pages · Programmatic & Topic Pages
v5/v8 · 🤖 GatedBuild sub-pages only for terms at pos 8–20 or KGR < 0.25 (pos 50+ is authority starvation). Every page needs its own methodology and hands-on test block. v8 batch gate: release new page types in 5–10% batches; require ≥1% click-earning or 7-day impression rate before unlocking the next batch.
📊 Water meter: Batch impression rate & click-earning page ratio (≥1%)
10Living Water · Fresh Feeds & 48h SLA
v5/v8 · 🤖 AIDated content streams (weekly reports, daily digests, immutable monthly stats snapshots) train crawlers to visit daily. New-term waves last ~2.5 weeks: ship event pages within 48h, link them back to evergreen hub pages, and run a mandatory 48-hour GSC impression check after launch.
📊 Water meter: 48-hour new-page impression rate & feed crawl frequency
11Drainage · Index Hygiene & Link Recovery
v5/v6/v7/v8 · 🤖 AIProtect site-wide quality evaluation: (1) noindex,follow thin pages; (2) track N/M/X monthly and verify every child sitemap is actually read; (3) v6 monitor 'Google-selected canonical differs from user'; (4) v7 301-redirect any 404 URL that has backlinks; (5) v8 gate: halt new page creation if Discovered/Crawled-not-indexed rises for 2 consecutive months.
📊 Water meter: N/M/X ratio + canonical mismatches + backlinked 404s (0)
12Outlet · How the Result Shows
v6 · 🤖 AIRanking is not clicking: (1) unique title matching H1 ≤60 chars + unique meta description; (2) homepage WebSite schema with name + alternateName; (3) structured data only for what the page shows; (4) if CTR drops at the same position, check whether Google rewrote the title link.
📊 Water meter: CTR at the same position (re-check 2–4 weeks after a title or description change)
13Waterwheel · Conversion, Retention & Bot Filtering
v5 · 🤖 instrument · 🧑 reviewMeasure whether water turns the wheel (install copies, audit runs, checkout clicks, lead forms with 'How did you find us?' AI attribution). v5 rule: strip datacenter headless-browser traffic (often 80%+ of raw sessions) before reading GA4, and share one analytics helper across all pre-rendered HTML shells.
📊 Water meter: Human session conversion rate, subscribers & net revenue
14Iteration · 5-Layer Meters & Weekly 1-Experiment Loop
v5/v6/v8 · 🤖 AIFive layered meters: (1) Clicks (GSC + bot-filtered GA4), (2) AI Citations (GSC Generative AI report + Bing AI Performance + Clarity + the weekly GEO citation panel, 51 questions), (3) Authority (DR, referring domains & whether targets allow AI crawlers), (4) Indexing (N/M/X + child sitemap reads + Bing index count), (5) Pipe Health (GSC Live URL test + 404/canonical match). Weekly retro picks 3 wins, 3 worries, and 1 dated experiment.
📊 Water meter: 5-layer monthly audit completion & dated experiment follow-through
🚧 v8's 6 Release Gates (Wait for the signal before the next step)
- 9 Sub-pages · Batch release gate: roll out new page types in 5–10% batches; require ≥1% click-earning or 7-day impression rate before the next batch.
- 10 Living water · 48-hour impression check: verify GSC impressions 48h after publishing; if zero, check SSR HTML, word count, title, and keyword density.
- 11 Drainage · Unindexed circuit breaker: if Discovered/Crawled-not-indexed rises for 2 straight months, freeze new page creation and clean up first.
- 2 Inflow · Signal-led link building: start backlinks only after indexing grows steadily; when pos 8–20 terms accumulate, upgrade copy against Top-5 first.
- 1 Catchment · Hard commercial proof: standardize volume on Google Ads API and require 1 hard monetization proof per commercial score.
- 14 Iteration · Monthly Bing gate: inspect Bing Webmaster scan reports and sitemap reads monthly so silent Bing drops never go unnoticed.
🧭 Jason's 6 Ranking & Social-Vote Laws (2026-10-07)
- Most keywords can reach the Top 5 or Top 3 on page quality alone when intent is focused and the page explains 'why' (e.g., our DR 13 PPT benchmark page ranking Top 3).
- Hyper-competitive head terms are won off-page through PR, brand, and social volume — on-page perfection is just the entry ticket.
- Backlinks are votes, and social/brand buzz are votes too; strong social distribution reduces dependence on DR.
- Once a site's technical baseline is solid, rankings scale roughly exponentially with brand and social momentum.
- Brand and social moats require real assets and outlast anything money can buy in link markets.
- Every page needs its own theory and testing rubric ('why these win and how we graded them'); commodity AIGC listicles get wiped out.
🧪 Pending (v9 candidates, moved here on 2026-11-21)
These come from books and talks. They make sense, but we have not tested them on our own sites. To move into the model, pages that got the change must beat similar pages that did not by more than 5 points on "cites the site" in the GEO citation panel for 4 weeks in a row (two runs on the same day differ by 1 to 4 points), or clearly beat them in Search Console.
- 12Put the conclusion, the brand name and real numbers in the first 50 words (Michal Suski's study of 50 million AI answers). Test: add it to some posts and not to similar ones, logged as an experiment.Not started
- 12The first paragraph under each H2 answers that heading on its own (JOKER, Mike King). Test: rewrite 5 posts, leave 5 similar ones.Not started
- 2Links and mentions as brand + context + relationship rather than exact-match anchors (JOKER). Test: the GIGALinks experiment, 10 treated and 6 control questions, re-run on Dec 1, Dec 15 and Jan 1.Testing
- 3Tables, datasets and video transcripts get cited more than prose (Mike King; "YouTube is cited 200 times more than other video platforms" is his figure, not ours). Test: add comparison tables to some pages, not to similar ones.Not started
- 5Fix four entity facts and keep every description in one semantic cluster; connect JSON-LD across pages with fixed @id values (JOKER, Andrea Volpini; "30% better retrieval" is his experiment). Already live site-wide with no control group, so we can only watch the trend on brand questions.Trend only
- 7New pages stay within two hops of the core entity; no off-core content (JOKER). Test: compare core and off-topic pages in Search Console.Not started
- 1SEO only captures existing demand; a category nobody searches for also needs social reach and word of mouth (Eli Schwartz, Tim Soulo). A principle for direction, not tested on its own.Reference
Dropped: "score a target site's AI impact before pitching and only pitch above 75". The score was our own; Walker Sands does not publish its method, and the command is gone.
Old numbers (renumbered in flow order on Nov 21, 2026): 1 = Ⓠ, 2 = ①a, 3 = ①b, 4 = ①c, 5 = ②, 6 = ⓪, 7 = ③, 8 = ④, 9 = ⑤, 10 = ⑥, 11 = ⑦, 12 = ⑩, 13 = ⑧, 14 = ⑨
From my post 「The SEO water model, v4」 2026-07-24