Data-Driven SEO Content Engine Methodology (How to Operate It, Not "What Google Says")

The knowledge base already has 36 articles gsc-* that are official references("what Google says"). This article fills in the operational layer: how to actually run a set of SEO content production using those rules. Refined from real-world practice in May 2026.

Core mindset: one closed data loop, not AI bulk article generation

Use real search data to decide what to write → produce thick, uniquely valuable content that reads like a native speaker wrote it, following official best practices → after publishing, use GSC data to identify what to optimize → keep the loop turning.

Fit human judgment + AI production capacity into one pipeline backed by official guidance and with data feedback.

One ironclad official rule that runs through the whole process (from GSC Blog: Evolution of AI Search and Google's Stance on AI Content (2015-2026)):"It's just content" — Google judges content quality, not how it is produced. AI assistance + humans injecting real experience/data/cases = OK; unattended scaled generation of keyword pages = scaled content abuse, and will be penalized.Orientation: AI boosts capacity, humans provide unique value and judgment; never run unattended bulk page generation.

9-stage workflow

#	Stage	What to do	Focus points	Official basis
0	Data acquisition and analysis	Pull GSC performance (queries/impressions/clicks/CTR/rankings) + index coverage + GA + Trends	Serves both as the keyword selection entry point and post-publishing feedback	How to use the Search Console + Analytics + Trends trio (traffic diagnosis / keyword research / Looker Studio)
1	Keyword selection	Real data + four intent categories (informational/navigational/transactional/commercial investigation)	Prioritize transactional + commercial investigation; do not create separate pages for every long-tail variant (= scaled abuse)	Google SEO Starter Guide + How Search Works + Search Essentials
2	Planning and intent mapping	Keyword-to-page mapping, one unique angle per page	Eliminate duplicate pages across the site	Complete Guide to Canonical / Redirects / Site Migration (including full HTTP status code table + 503 temporary closure playbook)
3	Drafting	Write according to the plan, with clear heading hierarchy	Structure for human readers; cover all sub-questions of the intent	Google SEO Starter Guide + How Search Works + Search Essentials
4	Remove AI flavor / machine translation flavor	Inject real experience/data/cases	Remove boilerplate/generic sentences; vary sentence length; use specific nouns and numbers; remove translationese	GSC Blog: Evolution of Quality Policies and E-E-A-T (Webmaster Guidelines → Search Essentials)
5	Multilingual localization	Write natively in each language, not translate from English	Genuinely localize the body text (translating only header/footer counts as duplicate content)	Complete Guide to Internationalization (hreflang/multilingual/multiregional) + Explicit Content (SafeSearch/Adult)
6	on-page	title/meta/canonical/hreflang/internal links/structured data	title is the strongest explicit signal; use absolute URLs for canonical; hreflang must point to reachable final URLs	Complete Guide to Googlebot Crawling Mechanisms and Index Control (URL/Links/Mobile-first/Meta/Testing/AMP)
7	Depth + quality gate + fact verification	Pass the gate before publishing	Depth ≠ word count = density of unique value; at least half of each page must be exclusive content;verify external facts online(see Lesson 1)	Google SEO Starter Guide + How Search Works + Search Essentials
8	Closed-loop iteration	Check data every month / after changes	Signals trigger actions for the next round	How to use the Search Console + Analytics + Trends trio (traffic diagnosis / keyword research / Looker Studio)

Signal-to-action mapping (the brain of the closed loop): impressions but no clicks → revise title/meta; rankings 5–15 → add content to push onto the first page; high-impression queries with no corresponding page → build the page; not indexed → diagnose root cause per official criteria.

Four hard practical lessons (learned the hard way)

Lesson 1: Competitor/external facts must be verified online + marked with a verified date

Competitor prices/free tiers/features expire extremely quickly. Empirical evidence: on one site's own 8 competitor comparison pages, online verification found that 6 pages carried wrong data(free tier off by 5x, "no mobile version" when there actually was one, pricing tiers completely wrong) — wrong enough that users would spot it immediately and drag down E-E-A-T.

For any cited external fact (especially about competitors), mandatory online verification before publishing and mark the page "verified <month>", never trust old internal copy. Tools should automatically trigger web verification for "competitor facts" type content.

Lesson 2: Don't compete on "free", compete on "giving you better stuff for free"

Niche competitors' free tiers are often quite generous too; the pitch "we're free, they lock you out" often doesn't hold up(users check and see through it). The real differentiator is to find what their free tier actually lacks: poor free model quality / requires account sign-up / has ads or queues / narrow topic coverage.

Upgrade the selling point from "free" to "better X for free" (better model / lower barrier / broader topic coverage). "Free" by itself is no longer a differentiator.

Lesson 3: Deliver complete final drafts, not diff-style correction suggestions

When revising competitor/landing page content, most pages are mostly good with only a few errors — but the final deliverable to downstream (engineering) should be a complete final draft(whole sections ready to paste in), not a list of "❌ old → ✅ new" correction suggestions.

Correction suggestions only serve as audit records of "what changed and why"; always deliver complete final drafts. For pages where changes are so large that the core argument shifts (angle rewrite), rewrite the whole thing directly instead of providing patches. (See also memory wiki-communication-style rule 20)

Lesson 4: After on-page content reaches a certain volume, returns diminish — don't pile content on "new sites that lost index / have no authority"

After writing a batch of high-quality on-page pages,the marginal value of writing more pages drops rapidly. Especially for a new site that has lost index / just launched with no authority: newly written pages will get stuck at "crawled - currently not indexed" just like existing pages,yielding zero short-term return(Google isn't properly indexing your site yet).

Judgment: Content is not the only lever, and often is not the current bottleneck. For new sites / sites that lost index, the real order of levers is ① fix indexing mechanics (so the site can be indexed) ② off-page authority(see below) ③ wait for Google to reindex (2–8 week lag).When you recognize "writing more pages won't get them indexed", stop writing content and shift to mechanics + authority. Don't use "write more pages" to cover up the real problem of "the site isn't being indexed at all / has no credibility".

off-page: Backlink / authority building (the #1 lever for new sites, don't only stare at on-page)

The first 9 stages of the whole engine are all on-page (on your own site). But for a new site with weight ≈ 0, the #1 lever for rankings/indexing is off-page — other sites mentioning/linking to you (backlinks = votes of trust). Just writing pages well won't get rankings if Google doesn't trust a new site.

Playbooks (reusable):

Directory submissions: list the site in AI tool directories (general + niche vertical), 30–50 high-quality directories over 4–6 weeks, quality > quantity; write original descriptions, don't mass-send.
Review list outreach(highest value): email authors of "best X 2026" review articles that already rank for your category terms, asking to be included — you get both backlinks and targeted referral traffic. Personalize each one, don't mass-send.
Natural community mentions: be a real contributor in relevant subreddits / Discords (don't spam ads, or you'll get banned).
🔑 Exclusive lever — the "hosted version of an open-source tool" brand angle: if the product is a hosted/online version of a well-known open-source project,use this angle for all backlink copy(borrow the parent project's existing authority instead of cold-starting an unknown site). This is the strongest off-page differentiator for same-origin products.

Anti-pattern: only burying your head in writing pages (on-page) with zero investment in off-page. New sites that do this won't rank no matter how good the pages are — because no one is "vouching" for you.

Tool form: main skill orchestrating dedicated subagents

Full funnel + multilingual + closed data loop,a single agent can't hold it all. Split into a main skill orchestrating 7 dedicated subagents (data analysis / keyword planning / writing / de-flavoring / localization / on-page / quality gate); the main skill strings the workflow together and pauses at keyword selection / quality gate / publishing for human sign-off. Make niche/audience/competitors/tone into swappable configurations, so migrating to another site only requires swapping configs.

When to use / not use

✅ Use: when you need to do systematic, reusable, migratable SEO content production + ongoing data optimization for a site.
❌ Don't use: writing a one-off article (just write it directly, no need to spin up the engine); pure technical SEO fixes (that's the code layer, see Complete Guide to Googlebot Crawling Mechanisms and Index Control (URL/Links/Mobile-first/Meta/Testing/AMP) / Complete Guide to Canonical / Redirects / Site Migration (including full HTTP status code table + 503 temporary closure playbook)).