Quality review and the publish gate
What every article is checked for, how the QC score and status work, what sets needs review, and what blocks auto-publishing.
Every article is checked right after it is written. The check gives a QC score out of 100, a QC status and, when something is wrong, a needs review flag with its reasons. The Library shows the score in the QC column and the details when you unfold a row. Automatic publishing only happens for articles that pass the publish gate below. When you press Publish yourself, you are the reviewer and the gate does not stop you.
What is checked
| Check | What it looks at |
|---|---|
| Truth layer | Runs twice. Removes or rewords invented people and quotes, first-person test claims, placeholder links (example.com, "#") and statistics without a source page. |
| Language | The article is in the project's language, with no section in another language. |
| Structure | Headings, lists and tables; item sections for listicles; a comparison table for comparisons; no duplicate headings, cut-off sentences or leftover placeholders. |
| Competitors | Links to competitor sites are removed (the official sites of listed products stay); forbidden terms and promotional competitor mentions are flagged. |
| Brand claims | What the article says your product does is checked against your Brand Book. |
| Sensitive topics | Regulated topics such as health claims or CBD follow stricter rules; breaking sentences are removed, and some topics need a person to publish. |
| E-E-A-T | Experience, expertise, authority and trust signals, scored by a separate AI model. |
| Fact check | Claims are checked when web research is on. |
| Originality | How far the article goes beyond the key points of the ranking pages. |
| Keywords and readability | Keyword use and spread, the keyword in headings, sentence length, transitions and accessibility. |
| Brand, links, length | 2 to 4 brand mentions (none in Educational style), internal links, cited sources, and a length within 15 % of the target. |
| Cannibalisation | Compared with your existing articles on the same keyword. Recorded only; it never blocks. |
The QC score
| Part | Points |
|---|---|
| Keywords | 20 |
| Structure | 15 |
| Readability | 15 |
| Content | 12 |
| Brand | 13 |
| E-E-A-T | 10 |
| Factual accuracy | 15 |
The Library's "QC Score Breakdown" shows the first five parts. Factual accuracy loses points for each error found and each unsupported claim removed. When the score is below 90, Rankfender tries automatic fixes (sentence-level fixes, then up to 6 paragraph rewrites) and keeps the better version.
QC status
| Status | When |
|---|---|
passed | Score 90 or more and no blocking reason. |
passed_with_warnings | Score 75 to 89 and no blocking reason. |
needs_review | Score below 75, or any blocking reason. |
What sets "needs review"
| Reason | Code |
|---|---|
| Wrong language, or a section in another language | language_mismatch, mixed_language |
| Content policy broken (for example fake-review wording) | policy_violation |
| A competitor link left after clean-up, or a forbidden term | competitor_link_remaining, forbidden_term |
| A list item that is not a real product | invalid_list_item |
| A placeholder, a duplicate heading or a cut-off sentence | placeholder_artifact, duplicate_h2_heading, truncated_fragment |
| Listicle or comparison structure missing or unbalanced | listicle_structure, comparison_table_missing, listicle_balance |
| A promotional mention of a competitor | promotional_competitor_mention |
| A sensitive topic that a person must review and publish | ymyl_human_review |
| A claim about your product that your Brand Book does not back | unbacked_capability_claim |
| The truth layer removed 10 or more invented items | truth_guard_heavy |
| Score below 75 | score_below_threshold |
Two findings are noted but never block: a section that opens with a pronoun with nothing to refer to (orphan_sentence) and internal links lost when a sentence was removed (lost_internal_links). The reasons are stored with the article; the app shows the score, the status, the breakdown and recommendations.
The publish gate
The autopilot and auto-publish send an article only when all of these are true:
- It is not published yet (draft, scheduled, ready or idea) and has at least 500 characters.
- It is not flagged needs review, and its QC score exists and is 75 or more.
- It has no critical policy violation, and the language check passed.
- It has no blocking review reason.
- Its outline was built from live search results, not from AI knowledge alone.
- It is not a review of your own brand published on your own site.
If one fails, the article stays a draft, the attempt is logged with the reasons ("Blocked by auto-publish gate"), and the autopilot does not retry it on its own.
Note
The gate only applies to automatic publishing. Publish in the Writer always works for a person with the right role. See Publishing and autopilot.
An article needs review: what to do
Look at the numbers
In the Library, unfold the row: QC Score, QC Status, QC Score Breakdown and Recommendations show where it lost points.Read the article
Open it in the Writer and read it, with Proofread if your plan has it.Fix and save
Edit the text, or select a passage and regenerate it. Press Save.Publish yourself
When you are happy, press Publish. Or press Regenerate to write it again.
Proofread
Professional and up Proofread opens a "Proofread Report" with a summary and five checks marked pass, warning or fail: Competitor safety, Instruction adherence, Hallucination risk, AI artifacts and Brand alignment. Each issue shows its severity, the passage and a recommendation.
- Apply safe fixes: applies the fixes that need no judgement.
- Rescan: checks the article again.
- Focus in article, Regenerate flagged part and Dismiss, issue by issue.
A red badge on the Proofread button counts the open issues.