Seo quality gates

SEO Quality Gates

The last line of defense. Other skills produce; this one blocks. An AI agent optimizing for search will, unchecked, drift toward exactly the patterns Google's spam policies punish. Run this before anything is published. Output a clear PASS / BLOCK per gate with the evidence and the required fix.

How to run

For each gate below, inspect the proposed change. Mark BLOCK if it trips, with the specific evidence and the fix. Anything BLOCK = do not ship until resolved. When the agent can run a shell, scripts/lint.py flags several of these heuristically on a draft.

The gates

1. Fabricated data — BLOCK on any invented fact

No invented ratings, review counts, reviews, prices, availability, statistics, quotes, authors, credentials, or citations. Structured data must describe real, visible, accurate content. Missing data → "data needed", never a guess.

2. Fake / ineligible review & rating markup — BLOCK

No aggregateRating/review without genuine, customer-provided reviews shown on the page. No self-serving reviews of your own business/products. No star markup to fake credibility.

3. FAQPage abuse — BLOCK

FAQPage only for genuine Q&A actually visible on the page. No FAQs invented to grab snippets; no promo/ad content as FAQs; no FAQPage on pages where it doesn't belong.

4. Keyword stuffing — BLOCK

Unnatural repetition of keywords, lists of cities/variants, or text written "for the algorithm". Read it aloud: if a human wouldn't write it that way, it's stuffing.

5. Thin / doorway / scaled-content abuse — BLOCK

• Pages with little original value created mainly to rank.
• Near-duplicate location/keyword pages (doorways) with swapped names.
• Bulk AI-generated pages with no added expertise, data, or first-hand value. Google's "scaled content abuse" and "doorway" policies target exactly this.

6. AI slop & people-first failure — BLOCK

Generic, padded, obvious content that restates the query without insight. No demonstrated experience/expertise. Written for search engines, not people. Disclose AI assistance where relevant (Who/How/Why).

7. Cannibalization — BLOCK (or flag for consolidation)

Two pages targeting the same intent that will compete. Require consolidate / differentiate / canonicalize before shipping both.

8. Cloaking & hidden text — BLOCK

Content shown to crawlers but not users, hidden text/links, sneaky redirects, schema for invisible content. Always a violation.

9. Risky technical edits — REQUIRE confirmation

noindex, robots.txt disallows, canonical changes, redirects, and hreflang edits can de-index a site if wrong. Require explicit human confirmation of intent and a rollback note before applying to live pages.

10. Over-automation / no human in the loop — WARN/BLOCK

Mass automated publishing without review, auto-generated metadata at scale without checks, or schema injected sitewide without per-type verification.

Output

| Gate | Status | Evidence | Required fix |

End with an overall verdict: SHIP (all pass), or DO NOT SHIP with the blocking list. If risky technical edits are involved, restate the rollback plan.

See references/google-policies.md and scripts/lint.py.