# Quality gate and checks Every generated page is scored and checked before it is allowed to go live. The score can be displayed on every plan; the **enforced gate** and **AI detection shield** are Starter and above. ## How the gate works Writrex runs **38 checks** on each page: **31 weighted checks** produce a score out of 100, and **7 blocking checks** must pass regardless of score. | Score | Status | | --- | --- | | 80 and above | Good — publish-ready (if no blocking failure) | | 60–79 | Needs review | | Below 60 | Weak | When you choose *Publish* and the enforced gate is on (Starter and above), a page below 80 — or with any blocking failure — is saved as a **draft** instead, with the reasons recorded. Blocking checks and the scrubber always apply on every plan. ## Scored checks Some checks apply only when the relevant data exists (for example, keyphrase checks need a focus keyphrase). | Check | Points | What it looks for | | --- | --- | --- | | Unique title | 10 | No other page or post uses this title | | Keyphrase in H1 / title | 10 | Focus keyphrase in the title or H1 | | Keyphrase in content | 9 | Mentioned naturally in the body | | Keyphrase in meta description | 6 | Present in the SEO description | | H1 / title present | 7 | A clear H1 or title | | Thin content | 12 | At least 700 words | | Comprehensive coverage | 6 | 1,000+ words | | Section headings and subheadings | 8 + 5 | H2 sections and H3 detail | | Internal links | 8 | At least one useful link or CTA link | | Strong CTA | 8 | An obvious next step | | FAQ coverage | 7 | FAQs present | | Specific and complete FAQ answers | 6 + 6 | Practical detail; every part of a question answered | | Schema present / FAQ structured data | 7 + 5 | Structured data is output, including FAQ schema when FAQs exist | | Local relevance | 5 | Area and service-area context | | Local proof / evidence | 7 | Real local facts, research or proof points | | Trust and experience | 6 | Experience, client or proof language | | Meta description length | 6 | Roughly 80–170 characters | | Readable sentences | 6 | Average sentence length 24 words or fewer | | Location repetition control | 7 | Exact area mentions within your limit | | No fake-testimonial risk | 8 | No generated testimonials without a real review supplied | | Unsupported claim check | 7 | Exact percentages, prices, years and results are backed by supplied context | | Starting price used | 6 | A configured price appears on the page | | Banned phrase check | 6 | None of your banned or generic phrases | | Unique internal anchors / no repeated targets | 5 + 4 | Varied anchor text; no duplicate link targets | | Original headings | 5 | Headings aren't reused from other pages | | No competing page | 5 | No unresolved overlap with another generated page | | Doorway similarity guard | 8 | Not too similar to another generated page | ## Blocking checks These carry no points — they simply must pass. - **No AI output leak** — no editorial notes, placeholders or assistant commentary - **All links resolve** — no link to a URL that doesn't exist on your site - **No unreplaced template tokens** — no stray `{{TOKEN}}` - **No stray escape artifacts** - **Meta description is purpose-written** — not cut from the page body - **No leftover template copy** - **CTA links are populated** — contact URL and phone set ## Pre-publish checklist and Approve & Publish Every page, post and product shows a *Writrex — Pre-Publish Checklist* in the editor sidebar, listing each check and whether it passed. Two more panels help: *Research Brief* (when used) and *SEO Plugin Guidance*. **Approve & Publish** publishes a generated page, records who approved it and when, and requests a Search Console sync when connected. A blocking failure always prevents approval. IndexNow pings fire on publish for generated content when configured. ## Output scrubber The final safety net between raw AI output and saved content. It removes editorial notes, unfilled placeholders, prompt echoes, assistant preambles and sign-offs, and literal escape sequences. With *Hold for review* on, anything it had to clean is saved as a draft with a notice. It is a safety feature, not a paid one. ## Link integrity validator Strips invented external URLs and unsafe link output, then re-validates after internal links are injected. Stripping all external links is the safe default. ## AI detection shield (Starter+) Reduces the patterns that make scaled AI copy read as generic: - Cleans stock AI phrases and repeated patterns. - Caps exact area mentions and swaps in nearby terms. - Works alongside banned phrases, heading variation and section-label options. **What it is and isn't.** It improves how pages read. It does not guarantee any detector will or won't flag a page, and good prompts, real business facts and human review matter more than any filter. Phrase cleanup is list-based and will always lag new AI stock phrases. ## Naturalness refinement and E-E-A-T enhancer - **Naturalness refinement** is an optional extra pass that rewrites against the checks that failed. The rewrite is kept only if it doesn't make the page worse. It can add a little generation time. - **E-E-A-T enhancer** pushes the model to use supplied experience, examples, pricing and case studies. It should not invent proof. - **Doorway protection** and **campaign variation** encourage unique local angles and section variety across scaled pages.