Flagship Skill ยท Editorial QA

The editorial QA skill.

Catch problems before they ship. Catch-problems QA, not checkbox theater.

A senior editor's playbook for pre-publish content QA. Brief adherence, voice consistency, fact accuracy, structure and clarity, AI-content audit, SEO and AEO compliance, internal linking and schema validation, QA at scale via sampling, the QA workflow, and the discipline that distinguishes catch-problems QA from process theater.

Audience: editorial leads, content directors, in-house content QA, agencies with production lines, content ops managers, anyone running a writer (human or AI) and accountable for what ships.

View the skill on GitHubBrowse the full catalog

What this skill is for

The gate-scope skill in a six-skill content workflow.

Six skills compose into the content discipline. Each has its own scope; the distinction is what keeps each skill sharp. This skill plugs in at the END of every other skill's output: the gate where quality is actually enforced.

  1. 01

    content-strategyPROGRAM scope

    Decides what to produce.

  2. 02

    pillar-content-architectureHUB scope

    Designs the topical hub structure.

  3. 03

    content-brief-authoringPER-PIECE scope

    Briefs each piece.

  4. 04

    content-and-copyEXECUTION scope

    Writes each piece.

  5. 05

    programmatic-seoSCALED scope

    Generates pages from data at scale.

  6. 06

    editorial-qaGATE scope (this skill)

    Verifies before publish. The gate where every previous decision gets tested against actual output.

The keystone distinction

Three QA shapes. Most teams ship one of the broken two.

QA is broken in one of two directions: too thin (read it once, ship it) or too thick (47-item checklist nobody completes honestly). The catch-problems discipline cuts both failure modes by sharpening the question every check has to answer: what class of failure does this catch?

Too thin

Thin QA

"I read it once, it is fine." Catches obvious mistakes; misses brief drift, voice inconsistency, hallucinated facts, structural problems, AI tells. Output: shipped content that is fine but not memorable.

Too thick

Thick QA

47-item checklist nobody completes honestly. Every conceivable check listed; no triage. Reviewers skim and check boxes (theater) or burn out under the cognitive load. Throughput collapses; reviewers leave.

The discipline

Catch-problems QA

Each check earns its keep by catching a class of failure that would reach readers. Checks that have not caught anything in 6 months get cut. Reviewers are accountable for what they caught, not for box-completion.

The litmus test. If the team can name the last 3 problems each QA check caught, the check earns its keep. If a check has not caught anything in 6 months, it is theater. Cut it; reallocate the attention to checks that catch real problems.

The AI-content audit

The QA check that did not exist as prominently 2 years ago.

AI-co-authored drafts have detectable patterns even when reviewed by a competent human editor. The audit scans for 11 AI tells (excessive em-dashes, predictable openings, forced bilateral framing, hedge stacking, and more) plus 6 hallucination patterns (specific decimals with named-source-that-does-not-exist, fabricated quotes, fake case studies). Below is a 4-paragraph excerpt with tells highlighted on the left and the audit checklist on the right.

AI-drafted excerpt

In today's fast-paced digital landscape, content marketing has become more important than ever. Whether you're a small business owner or a Fortune 500 executive, understanding the nuances of effective content strategy is crucial.

There are many factors to consider. On one hand, you need high-quality content. On the other hand, you need optimization for search engines, social media, and more.

Studies show that companies prioritizing data-driven decisions see significant improvements in their performance metrics.

By following these steps, you'll be well on your way to creating a content strategy that drives results.

AI-content audit

The editor scans for AI tells and hallucination patterns. Each match flags a revision item; each absence confirms a quality signal.

  • Throat-clearing opener
  • Forced bilateral framing
  • Sentence-end filler ('and more')
  • Hedge stacking
  • 'Studies show' with no study
  • Wrap-up filler ending
  • Specific statistics with sources
  • Named experts with attribution
  • Distinctive POV

Verdict: 6 of 9 audit items flagged. Halt the piece; return for revision with the specific tells called out.

The verdict in this example: 6 of 9 audit items flagged. Halt the piece; return for revision with the specific tells called out. Without a named gate, these patterns slip through routine review.

The framework

Twelve considerations for editorial QA.

When designing or auditing a QA process, walk these 12 considerations. Skipping any of them produces one of the failure modes documented in the skill.

  1. 01Catch-problems vs checkbox: every check earns its keep
  2. 02Brief-adherence first: cheapest, highest-value gate
  3. 03Fact-accuracy as halt-condition: AI hallucinations stop here
  4. 04Voice consistency including mid-piece sampling
  5. 05AI-content audit: 11 tells, 6 hallucination patterns, voice drift
  6. 06Structure and clarity: lede, sectioning, even-ish length
  7. 07SEO and AEO compliance: answer paragraphs, schema, entities
  8. 08Internal linking and schema validation
  9. 09Sampling at scale: 50 to 200 pages per cycle
  10. 10Threshold gating: 5% failure halts new generation
  11. 11Single QA owner per piece: accountability not diffused
  12. 12Sequencing: brief to fact to structure to AI to voice to SEO to linking to schema

What is in the skill

Fourteen sections covered in the body.

The SKILL.md spans the full editorial QA discipline from the keystone catch-problems-vs-checkbox framing through every gate (brief, fact, structure, AI, voice, SEO/AEO, linking, schema), QA at scale, workflow design, and the failure-mode catalog.

  1. 01

    What this skill is for

    Composition with five sister content skills. This skill is the gate at the end of every other skill's output: brief, hub, per-piece, execution, scaled. QA is where drafts become publishable.

  2. 02

    Catch-problems QA vs checkbox QA

    The keystone distinction. Each check earns its keep by catching a class of failure. Checks that have not caught anything in 6 months get cut. The litmus test: can the team name the last 3 problems each check caught?

  3. 03

    Brief adherence check

    Did the writer execute the brief? 8-field checklist: target keyword, intent, audience, heading structure, required entities, internal links, anti-patterns, success criteria. Cheapest gate; runs first.

  4. 04

    Voice consistency check

    Vocabulary, sentence rhythm, stance, register, mid-piece sampling. Long pieces drift; sample throughout, not just at start and end. Voice drift is the dominant failure in AI-co-authored content.

  5. 05

    Fact accuracy and citation discipline

    Halt-condition gate. Every claim verified: statistics sourced, quotes verified, case studies real, dates verified, named experts consenting, product claims true. AI hallucinations stop here.

  6. 06

    Structure and clarity check

    Lede answers the user query (or establishes thesis). H2s map to user mental model, not writer enthusiasm. Section length even-ish. Specific endings. Read-aloud audit catches what line review misses.

  7. 07

    AI-content audit

    11 AI tells, 6 hallucination patterns, voice drift detection. The QA check that did not exist as prominently 2 years ago and is now load-bearing for any program with AI in the loop.

  8. 08

    SEO and AEO compliance check

    Combined audit: target keyword, headings, schema, meta, alt text, answer paragraphs at H2 level, TL;DR for pillars, FAQPage schema, statistics with sources, named entities, distinctive POV.

  9. 09

    Internal linking and schema validation

    Outbound links present and live, anchor text varies, schema validates and matches content, optional schema fields populated where data exists. Self-cannibalization check before publish.

  10. 10

    QA at scale

    Stratified sampling 50 to 200 pages per cycle. Automated checks on every page, manual review on the sample. 5% failure threshold halts new generation. Cohort tracking surfaces drift.

  11. 11

    The QA workflow

    Single owner per piece. Sequencing brief-adherence first, SEO/AEO last. Halt vs flag vs auto-fix taxonomy. Escalation when patterns surface across pieces.

  12. 12

    Common failure modes

    Eleven-plus patterns: 47-item checklist nobody completes, zero problems caught last quarter, voice disagreements, AI hallucinations to publish, thin pSEO pages, QA longer than writing, reviewer burn-out, mid-piece voice drift, ship-and-discover-on-social.

  13. 13

    The framework: 12 considerations

    Catch-problems vs checkbox, brief-adherence first, fact-accuracy as halt, voice with mid-piece sampling, AI-content audit, structure, SEO/AEO, linking and schema, sampling at scale, threshold gating, single owner, sequencing.

  14. 14

    QA is where quality gets enforced

    Every other skill in the content suite produces drafts. QA turns drafts into publishable work. It is also where every previous decision (brief shape, voice doc, hub architecture, programmatic template) gets tested. Skipping QA ships the failure modes of every upstream decision unfiltered.

Reference files

Ten references that go alongside the SKILL.md.

The references hold the brief-adherence checklist, the voice-consistency patterns, the fact-accuracy discipline, the structure review, the AI-content audit, the SEO/AEO checklist, the linking and schema validation, the QA-at-scale patterns, the workflow templates, and the failure-mode catalog. Each reference closes with a methodology-vs-implementation section per the discipline established by the skill-creation-walkthrough.

  • references/brief-adherence-checklist.md

    Every brief field as a QA check. 8-field walkthrough with verify/pass/fail patterns per field. Halt-condition return note format. The when-briefs-are-too-vague-to-enforce upstream-fix observation.

  • references/voice-consistency-patterns.md

    Vocabulary, sentence rhythm, stance, register dimensions. Mid-piece sampling discipline for long pieces. Voice-drift detection in AI-co-authored content. Quarterly calibration sessions to keep editors aligned.

  • references/fact-accuracy-and-citation-discipline.md

    What gets fact-checked. Three verification paths per claim. Six hallucination detection patterns. Citation rules including authoritative sources, citation laundering avoidance, source-age guidelines. Fact-check log for pattern detection.

  • references/structure-and-clarity-review.md

    Lede patterns (SEO/AEO vs thought leadership). Sectioning principles, section-length discipline, reading flow audit, specificity-vs-abstraction rule, endings discipline. Five structural anti-patterns. Read-aloud audit methodology.

  • references/ai-content-audit-patterns.md

    The 11 AI tells, the 6 hallucination patterns, voice drift detection methodology. Worked example with a 4-paragraph AI-drafted excerpt with tells highlighted plus a revised version. 7-step audit checklist.

  • references/seo-aeo-compliance-checklist.md

    Combined SEO and AEO checks in one workflow. Title and headings, meta and URL, on-page checks, answer paragraphs at H2 level, TL;DR for pillars, FAQPage schema, named entities, distinctive POV. Auto-fix vs flag vs halt taxonomy.

  • references/internal-linking-and-schema-validation.md

    Outbound link discipline, anchor text variation rule, link-target liveness, internal link count bands. Schema type matching content, schema validation, schema-content consistency, optional-field populate-where-data-exists discipline. 6-step link audit, 5-step schema audit.

  • references/qa-at-scale-patterns.md

    Stratified sampling across data shape and cohort. Automated checks at scale (heading, schema, word count, duplicates, broken links, image presence). Manual checks on sampled pages. 5% failure threshold gating. Cohort tracking and the QA log.

  • references/qa-workflow-templates.md

    Single-owner principle. Sequencing template (brief-adherence to fact-accuracy to structure to AI-audit to voice to SEO/AEO to linking to schema). Halt/flag/auto-fix taxonomy. Escalation patterns. Review templates. QA cycle time targets.

  • references/common-qa-failures.md

    12 failure patterns with diagnoses and fixes. Cross-references to other reference files. The pattern across failures: QA either drifted into theater (cuts needed) or skipped gates that should have been there (gates needed).

Browse all reference files on GitHub

Pairs with these platforms

Five platforms across the QA workflow.

The skill is the QA discipline; the integration pages are the platform-specific tactics. Frase ships per-page optimization with QA features built in. AirOps wraps QA gates in a managed workflow surface. Notion archives QA reports alongside published pieces. Profound measures post-publish AEO performance as a feedback loop. Mixpanel surfaces post-publish engagement signals that feed back into the QA process.

Bridges to sister content skills

Five sister skills feed into the QA gate.

QA is the gate at the end of every other content skill's output. Briefs feed brief-adherence; hubs feed internal-linking checks; program standards feed voice and structure checks; execution gets QA'd directly; programmatic sets get sampled at scale.

  • Briefs feed brief-adherence check

    content-brief-authoring

    Authors the per-piece briefs that the brief-adherence QA gate enforces. The cleaner the brief, the easier the QA gate; vague briefs make brief-adherence checks impossible.

  • Hub adherence

    pillar-content-architecture

    Designs the hub structure that pillar and cluster pieces are QA'd against. The internal-linking-and-schema-validation gate references the hub architecture; pieces that violate hub linking discipline get flagged.

  • Program QA standards

    content-strategy

    Decides program-level quality standards: voice doc, anti-patterns, success criteria, brand stance. The QA process enforces what content-strategy specifies; both skills compose into the editorial discipline.

  • Execution gets QA'd

    content-and-copy

    Writes the pieces this skill audits. The QA gates run against content-and-copy's output; failures route back to the writer for revision. The two skills are the production line.

  • Sampling QA at scale

    programmatic-seo

    Generates the programmatic page sets that sampling QA enforces against. The QA-at-scale patterns in this skill are the operational discipline programmatic-seo's quality-control section points to.

Where this skill fits in the content suite

The fourth of five in the content suite.

Editorial QA is the fourth skill in a planned five-skill content suite. Skills 1 through 3 covered the production discipline: content-brief-authoring for per-piece briefs, pillar-content-architecture for hub-level architecture, and programmatic-seo for scaled-content programs. This skill is the gate at the end of all three production tracks.

Skill 5 in the suite (forthcoming) is ai-content-collaboration: how to run an editorial team where some writers are AI agents. The AI-content audit dimension of this skill anchors the human-in-the-loop side of ai-content-collaboration; the two skills compose for programs running mixed human and AI production.

The suite ships sequentially. Skill 5 does not exist yet; cross-linking lands when that skill page ships. For now, the QA discipline anchors the gate at the end of the production line.

Open source under MIT

Read the SKILL.md on GitHub.

The skill source lives in the rampstackco/claude-skills repository alongside dozens of other skills covering the full lifecycle of brand and product work. MIT licensed.

View SKILL.mdBrowse the full catalog

Frequently asked questions.

What is the difference between catch-problems QA and checkbox QA?
Catch-problems QA earns its keep by catching specific classes of failure that would reach readers if missed. Each check is justified by a problem it catches; checks that have not caught anything in 6 months get cut. Checkbox QA is the 47-item process-theater checklist nobody completes honestly because it grew over time without pruning. Catch-problems QA produces shipped content that is actually quality-controlled; checkbox QA produces shipped content that looks reviewed but is not.
How is this different from the QA section in programmatic-seo?
The programmatic-seo skill treats QA as one component of a larger pSEO program design (alongside data sources, templates, schema, internal linking, refresh, and crawl budget). This skill treats QA as the discipline itself: how to design a QA process from scratch, how to sequence gates, how to halt vs flag vs auto-fix, how to escalate patterns, how to keep QA tractable at production speed for editorial pieces, AI-co-authored drafts, and programmatic sets all at once. The pSEO QA section is one application of this skill's framework.
Why is the AI-content audit a separate gate?
Because AI-co-authored drafts have detectable patterns even when reviewed by a competent human editor. The 11 AI tells (excessive em-dashes, predictable openings, bullet-list overuse, forced bilateral framing, hedge stacking) plus the 6 hallucination patterns (specific decimals with named-source-that-does-not-exist, fabricated quotes, fake case studies, dead URLs, studies-show claims with no study, made-up product features) plus voice drift (mid-piece regression to model-default voice) all require pattern recognition the editor needs to run deliberately. Without a named gate, these patterns slip through routine review.
What sequencing should QA gates run in?
Brief-adherence first (cheapest gate, catches the largest class of failures). Fact-accuracy second (halt-condition; AI hallucinations do not progress past this gate). Structure and clarity third. AI-content audit fourth. Voice consistency fifth (with mid-piece sampling for long pieces). SEO and AEO compliance sixth. Internal linking and schema validation seventh. Final read eighth. The discipline saves editor time: pieces that fail at gate 1 do not consume voice/SEO review time, which is the most common 'QA takes longer than the writing' failure pattern.
How does QA work for programmatic SEO at 100,000 pages?
Sampling discipline replaces full-audit. Stratified random sample 50 to 200 pages per cycle, balanced across data shape (sparse vs dense, recent vs old, popular vs niche). Automated checks run on every page (heading structure, schema validity, word count thresholds, duplicate content, broken links). Manual review on the sampled subset covers what automation misses (template adherence, top-200-word answer quality, distinctive vs templated, AI hallucination spot-check). 5% sample failure threshold halts new generation until template or data fixes ship.
Who owns QA when the team has multiple editors?
Single owner per piece. Committees diffuse accountability. Common ownership models include editor-owns-piece (one person reviews and ships), editor + dedicated QA (split labor with editor as ship-decision-maker), and rotating editor pool with calibration sessions. The pattern that fails is committee review: 4 reviewers comment, nobody decides, the writer revises against contradictory feedback. Pick a single owner, treat other reviewer input as advisory, run quarterly calibration sessions to keep individual editor standards aligned.