The QA gate: a practical editorial checklist for AI-drafted articles

Every publishing team says they review AI drafts. Very few can say what "review" means precisely enough that two editors would reach the same verdict on the same draft. This post shares the checklist behind the QA gate in AI Blog Factory — the checks a draft must pass before it can ship. Use it manually even if you never use our product.

Structure checks

• Every section from the brief is present, in order, and none are near-duplicates of each other.
• Heading hierarchy is legal: no H3 without a parent H2, no headings used for styling.
• The intro makes a claim the article actually delivers on — the most common AI failure we see.

Substance checks

• Each section says at least one thing that is specific: a number, a named example, a concrete step. Sections that only restate the heading fail.
• Any factual claim that would embarrass you if wrong is flagged for human verification before publishing.
• Hedge-phrase density is below threshold — strings of "may", "can", "it depends" are how models avoid saying anything.

SEO and linking checks

• Primary keyword appears in the title, the first 100 words and at least one H2 — naturally, or the section loops back.
• Internal links resolve and point to live pages from the link map.
• Title tag and meta description exist and fit within pixel-safe lengths.

Voice checks

• No phrase from the banned list ("in today's fast-paced world", "delve", "game-changer" — build your own).
• Average sentence length within the profile's range, and readability at the target grade level.

The rule that makes it work

A failed check doesn't mean the editor fixes it — it means the draft goes back to the station that caused it. Editors who fix everything by hand become the bottleneck the gate was built to remove. The gate's job is to make failure cheap and automatic; the editor's job is judgment on the drafts that pass.