Scale should increase the number of useful, reviewed artifacts—not the speed of publishing low-value pages. A safe pipeline separates brief, sources, draft, factual/editorial checks, approval, publishing, live verification, and maintenance.

// concept

Define Artifact States

// prompt — copy me3 lines

IDEA → BRIEFED → SOURCED → DRAFTED → FACT_CHECKED
→ EDITED → APPROVED → SCHEDULED → PUBLISHED → MONITORED
NEEDS_EVIDENCE | REJECTED | OUTDATED | CORRECTED | REMOVED

Every artifact carries owner, audience, canonical target, brief/source versions, AI provider/model/prompt version if used, claims register, reviewer, approval hash, publish ID/URL, and update trigger.

// concept

Keep AI in the Draft Boundary

AI may propose structure or draft from approved notes. It cannot invent first-hand experience, client results, testimonials, product tests, prices, or authority. Do not give the publisher credential to the drafting model.

Use deterministic checks for title length, one H1, broken links, schema validity, canonical, banned claims, and required sections. Human editors verify meaning, sources, originality, tone, copyright, risk, and whether the page deserves to exist.

// worked_example

Worked Example

A Pakistani software agency plans eight API integration guides. Each has a real implementation/test artifact and official API source. The content board prevents two writers from creating the same intent page.

AI drafts from the brief, but code snippets are tested, screenshots use synthetic data, and pricing/authentication facts are dated. An editor approves exact final hash. Publishing creates a CMS draft first; after release, the workflow fetches the canonical URL, checks status/title/H1/canonical/links, and records it. Failed verification rolls back or alerts.

// failure_cases

Failure Cases to Diagnose

6 cases to diagnose

Calendar demands a post regardless of evidence
reject weak idea.
Draft publishes automatically
separate identities and approval.
One prompt generates 100 city pages
scaled low-value abuse risk.
Source added after drafting
build from evidence first.
Final edit bypasses approval
invalidate hash.
No update/removal path
assign triggers and owner.

// pakistan_angle

Pakistan Angle

Pakistan context must come from verified business operations, primary sources, or authorized expert review—not an AI’s stereotypes. Use synthetic personal data and secure client examples.

For multilingual production, maintain separate editorial owners and parity checks for price, terms, dates, and claims. Do not automatically translate regulated or contractual advice.

Track rejected drafts as well as published ones. Rejection reasons—missing evidence, duplicated intent, unsafe advice, or weak user need—show where the pipeline is producing waste. Improve the brief or source requirements before increasing volume.

// hands_on

Hands-On Exercise

5 steps

Build the content state board.
create one evidence-first brief and source pack.
produce an AI draft with version record.
run deterministic and human QA.
publish to staging/CMS draft and live-verify.

// completion_rubric

Completion Rubric

6 checks — tick as you verify

0/6

// sources

Sources

3 official sources — check every claim yourself

// check_yourself

Check yourself

4 questions · answers and options are taken word-for-word from this course

0/4

1 / 4 · diagnose
Your work shows this failure mode: “Draft publishes automatically.” What does the lesson tell you to do about it?