Short answer / recommendation
Use a staged, reusable prompt set: 1) Outline generator, 2) Section-by-section drafts (JSON), 3) Citation & internal-link finder, 4) Final SEO polish. Run the outline through human review, then generate sections in parallel. Use ChatGPT for speed and Claude as an alternative when you need very long-context consistency or more cautious analysis.

Why this works
Breaking the job into deterministic steps produces predictable outputs you can QA and feed into your CMS. JSON outputs let your pipeline import headings, word targets, and link suggestions directly.

Prompt templates (copy-paste ready)
1) Outline generator (fast skeleton)
System: You are an SEO content strategist. Output only JSON with: title, meta_description (max 155 chars), primary_keyword, secondary_keywords[], search_intent, outline[] where each outline item = {h, brief, target_words}.
User: Produce a comprehensive long-form outline for the target keyword: "{KEYWORD}". Target length: {TOTAL_WORDS}. Audience: {AUDIENCE}. Competitor urls: {COMP_URLS}. Include recommended internal pillar pages (3) and content gaps to cover. Keep meta and keywords SEO-focused.

2) Section draft (repeat per H2/H3)
System: You are a professional SEO copywriter. Return JSON: {heading, word_count, draft_html, sources[] (url, anchor_text)}.
User: Write the section for heading: "{HEADING_TEXT}". Use the outline brief. Tone: {TONE}. Include 1–2 inline source citations with URLs and an anchor suggestion. Keep factual claims numbered and list sources at bottom. Avoid hallucinated sources; mark any uncertain claim with [citation_needed]. Target words: {WORDS}.

3) Citation & fact-check pass
System: You are a research assistant. Return an array of verified sources for numbered claims in the draft: {claim_number, snippet, source_url, source_type (news, study, gov, blog), confidence 0-1}.
User: For this draft text: "{PASTE_TEXT}", find the best supporting URLs (top 5 SERP + domain authority >=20). If claim is not verifiable, say so. Prefer primary sources.

4) Internal linking & FAQ builder
System: You are an SEO engineer. Return JSON: {internal_links: [{anchor, target_url, reason}], suggested_faqs: [{q,a,source_url}]}.
User: Given this outline and site map: {SITEMAP}, suggest 6 in-article internal links, anchor text, and 6 FAQ Q&A to add under an FAQ accordion (with sources).

Decision criteria (when to use what)
- Use lightweight prompts + human edit when producing 20+ articles/month and budget/time is tight.
- Use more iterations, aggressive citation checks, and Claude (or GPT with large context) when quality and fact accuracy are mission-critical.
- Team size: small teams should batch outlines; larger teams can parallelize section drafts.

Practical checklist before publishing
- Run Outline -> approve headings and word counts.
- Generate sections as JSON; import to CMS draft.
- Run citation pass; replace any [citation_needed].
- Run internal link pass and add canonical + structured data (FAQ schema).
- On-page SEO check: headings, meta, images (alt + size), schema, and page speed.
- Human edit for brand voice and factual accuracy.

Best-for / Avoid-if
- Best-for: agencies needing repeatable high-volume, consistent structure, quick QA.
- Avoid-if: you require investigative journalism or primary research—AI will need heavy human verification.

Final tip
Set model temperature to 0–0.3 for outlines and citations, 0.2–0.5 for creative parts. Always keep a short human review loop for accuracy and brand voice.

Compare ChatGPT and Gemini