Before you begin
Set the operating boundary first.
- One specific reader question or decision—not a keyword list.
- A named human editor who can reject, rewrite, or stop publication.
- Current primary sources or first-hand evidence for consequential claims.
- Non-sensitive inputs, or an approved provider and data-handling path.
Step by step
Move from scope to a checked artifact.
Define the reader outcome
Write who the page is for, the question it should answer, the action it should support, and what the page must not claim. Start from usefulness to a real audience rather than a target word count or a list of query variations.
- Checkpoint
- A reviewer can explain why this page should exist even if search engines sent it no traffic.
- Artifact
- Reader brief and publication boundary
Build a source packet
Collect the smallest set of current, primary sources that can support the page. Record the publisher, page title, URL, retrieval date, and the exact proposition each source can support. Keep background reading separate from evidence used for claims.
- Checkpoint
- Every planned factual section has an identifiable source or is clearly labeled as experience, judgment, or an example.
- Artifact
- Versioned source ledger
Freeze the content brief
Specify the page promise, outline, required evidence, prohibited claims, tone, internal links, disclosure needs, and final call to action. If AI will draft, give it the source packet and require source identifiers beside factual claims.
- Checkpoint
- The brief defines success before fluent draft copy can lower the review standard.
- Artifact
- Content brief with acceptance criteria
Generate a bounded draft
Ask the model to draft only from the supplied packet, flag unsupported gaps, and avoid inventing examples, quotes, dates, prices, or performance claims. Keep the raw generated version separate from human edits so the production process remains inspectable.
- Checkpoint
- Unsupported sections are marked for research or removal rather than filled with plausible text.
- Artifact
- Raw draft snapshot and gap list
Review claims line by line
Check that each factual claim is accurate, current, and actually supported by the nearby source. Review usefulness, originality, structure, copyright risk, sensitive data, and domain risk separately. Rewrite for the reader; do not publish an unedited model output as evidence.
- Checkpoint
- A named editor accepts each consequential claim and the page-level conclusion.
- Artifact
- Reviewed draft and claim ledger
Publish with a maintenance contract
Publish the author or responsible organization, how AI assisted when context calls for it, the checked date, limitations, and source links. Set an event trigger or review date, and update only when the underlying information or reader need materially changes.
- Checkpoint
- The page has an owner and a defined reason to revise, retire, or keep it unchanged.
- Artifact
- Publication record and review schedule
Deliverables
Keep the work reusable and inspectable.
- Reader and decision brief
- Versioned source ledger
- Prompt or drafting instructions
- Raw draft snapshot and gap list
- Reviewed claim ledger
- Publication disclosure and review schedule
Decision boundaries
What this tutorial does not prove.
- This workflow does not guarantee rankings, traffic, conversions, or factual completeness.
- Do not scale production when the team cannot provide source review and accountable editorial ownership.
- Medical, legal, financial, safety, or other high-impact content needs qualified domain review beyond this checklist.
- Do not upload confidential material unless the selected service and organizational policy explicitly allow it.
Source ledger
Check the current primary guidance.
- Google Search: helpful, reliable, people-first content
Primary guidance for usefulness, trust, authorship, process context, and avoiding search-engine-first production.
- Google Search: generative AI content guidance
Primary guidance for accuracy, quality, relevance, scaled-content risk, and explaining how automation was used.
- NIST AI 600-1: Generative AI Profile
Risk-management reference for defining controls according to context, risk tolerance, and available resources.