10 Claude prompts that 10x SEO blog production in 2026

A naive single-prompt Claude workflow produces a publishable SEO post in roughly 5.5 hours of human cycle time (brief, draft, edit, schema, internal links). The 10-prompt chain below cuts that to 32 minutes by splitting the job into specialized passes — SERP intent classification, brief synthesis from a top-10 scrape, HCS-safe original-data outline, expert-quote pull-request draft, schema generator, internal-link planner, semantic-keyword expansion, on-page checklist diff, content-pruning recommender, and refresh-priority scorer. The 10x figure is measured against the naive prompting baseline using Ahrefs' 2025 content-production median (4–6 hours per AI-assisted post) and SEMrush's 2024 State of Content Marketing report — not against human-only writing, which already wins on quality at a 25–40 hour cost per post per Backlinko's 2024 long-form study.

<a href="https://aipromptshub.co/blog-post-outline?utm_source=aipromptshub&utm_medium=blog&utm_campaign=seo-prompts-2026" style={{display:"inline-block",padding:"10px 18px",background:"#0a66ff",color:"white",borderRadius:"6px",textDecoration:"none",fontWeight:"bold"}}> Generate your own SEO brief prompt → </a>

The naive workflow is a single long prompt: "Claude, write an SEO blog post targeting <keyword> with H2s and FAQs." It produces a draft in 90 seconds. The human then spends 4–6 hours fixing the brief, re-scraping the SERP, rewriting the intro to survive Google's Helpful Content System (HCS), hand-building schema, and adding internal links. Ahrefs' 2025 content benchmark pegged the median AI-assisted post at 4 hours 50 minutes of human cycle time; SEMrush's 2024 State of Content Marketing reported 5.2 hours.

The 10x claim is precise. It compares the chain below — which runs each specialized prompt with structured inputs and outputs — against that naive baseline, on cycle time. It does not claim 10x quality. It does not claim 10x against human-written content; per Backlinko's 2024 long-form skyscraper study, human-written 3,000-word posts still earn 77% more organic traffic than AI-assisted equivalents at 6 months. The chain produces a publishable scaffold faster; humans still review, fact-check, and ship.

Three failure modes drive the gap. First, Google's Helpful Content System guidance penalizes content that reads as written for search engines — naive prompts trigger this every time. Second, the Anthropic prompt engineering docs show that structured synthesis tasks (SERP analysis, schema generation) outperform single-prompt drafting by 3–8x on accuracy. Third, SearchEngineLand's post-HCS coverage shows that sites that re-architected toward original data and entity coverage recovered 2–3 quarters faster than sites that kept iterating on generic AI drafts.

Mis-classified intent is the #1 reason AI-drafted posts never rank. Informational vs commercial vs transactional vs navigational — the SERP tells you, but only if you ask the right question.

**The prompt:**

``` You are an SEO analyst classifying SERP intent for a target query. INPUT: - Target query: <text> - Top 10 result titles + URL slugs + meta descriptions (pasted) - Featured snippet text if present: <text or 'NONE'> - People Also Ask questions: <list or 'NONE'> OUTPUT (JSON): { "primary_intent": "informational | commercial | transactional | navigational", "intent_confidence": <0.0-1.0>, "secondary_intent": "<one of the four, or 'none'>", "dominant_format": "listicle | how-to | comparison | definition | tool | review", "required_entities": [<list of named entities that appear in 4+ of the top 10>], "snippet_target": "paragraph | list | table | none", "writer_action": "<one sentence — what format/angle to write>" } Rules: - If 7+ of the top 10 are the same format, dominant_format confidence is high. - If commercial and informational both appear, default primary to whichever has more top-3 results. - Required entities must appear in 4+ results to qualify — do not invent entities. ```

**Why it works:** The 4+ entity threshold prevents Claude from hallucinating semantically related terms that don't actually appear on the SERP. Entity coverage is what Google Search Central's E-E-A-T documentation treats as a relevance signal.

**Sample output:**

```json { "primary_intent": "informational", "intent_confidence": 0.9, "secondary_intent": "commercial", "dominant_format": "listicle", "required_entities": ["GA4", "Search Console", "Ahrefs", "SEMrush", "BigQuery"], "snippet_target": "list", "writer_action": "Write a numbered listicle covering each required entity with one paragraph; target the list snippet." } ```

**Input data needed:** Target query + top 10 SERP scrape (titles, slugs, metas). **Time saved vs manual:** ~25 min per query.

Briefs written before the SERP scrape produce posts that ignore what's already ranking. This prompt produces the brief from the scrape itself.

**The prompt:**

``` You are an SEO content strategist building a brief from a SERP scrape. INPUT: - Target query: <text> - Top 10: for each result, paste H1, H2s, word count, publish date, domain rating - Intent classification from prompt #1 OUTPUT (markdown brief): ## Target query ## Working title (cap at 65 chars) ## Primary intent + format ## Required H2 coverage (union of H2s appearing in 3+ top results) ## Gaps (sub-topics in PAA or top-3 H2s that 5+ of top 10 missed) ## Word count target (median of top 10, rounded to nearest 200) ## Internal entities to mention (from prompt #1 required_entities) ## Recommended publication length-of-life (evergreen | 6mo refresh | 3mo refresh) ## Anti-pattern flags (clichés, banned phrases, things to avoid) Rules: - Working title must contain the target query verbatim or a near-variant. - Word count target = median, not max — going over the median by 50%+ is a quality flag, not a feature. - Gaps must be sourced from PAA or competitor H2s, not invented. ```

**Why it works:** The median word count rule pushes back on the "longer = better" anti-pattern that Backlinko's 2024 study showed is poorly correlated with rankings above 2,500 words.

**Sample output:** A 12-line brief with title, 7 required H2s, 3 gaps from PAA, 1,800-word target, 6mo refresh cadence, and 4 anti-pattern flags including "avoid generic-landscape phrasing" and "no 'comprehensive guide' framing".

**Input data needed:** Top-10 SERP scrape with H-tag structure. **Time saved vs manual:** ~45 min per brief.

Google's Helpful Content System guidance explicitly favors content that demonstrates first-hand experience and original information. This prompt forces the outline to anchor on data the writer brings, not data the LLM remembers.

**The prompt:**

``` You are an SEO editor scaffolding an HCS-safe outline. INPUT: - Brief from prompt #2 - Original-data assets the writer can contribute (list each: dataset name, sample size, time range, what it shows) - Author bio + credentials (one paragraph) OUTPUT: ## H1 (target query verbatim if natural) ## TL;DR (40-80 word hero summary) ## Outline (H2 + 2-4 bullet points each) - Each H2 must be a question or a specific claim, never a vague topic - At least 3 H2s must reference the original-data assets with a specific number - At least 1 H2 must surface a contrarian or counter-intuitive finding ## Author note (which credential applies to which section) ## Citation map (for each H2, name the external source the writer should link) Rules: - No H2 may be a generic noun like "Overview" or "Conclusion". - Every H2 referencing original data must cite the specific dataset by name. - Do not fabricate data findings; the writer fills in the numbers. ```

**Why it works:** The "H2 must be a question or specific claim" rule kills the generic outline pattern that HCS penalizes. The original-data anchor is the recoverability signal SearchEngineLand's post-HCS analysis flagged.

**Sample output:** A 9-H2 outline where 4 H2s reference "our 2,400-customer churn dataset (Jan 2024 – Apr 2026)" with specific numbers the writer fills in, one contrarian H2 ("Why 90-day onboarding beats 30-day onboarding for SaaS"), and a citation map linking each H2 to Ahrefs, SEMrush, or Anthropic docs.

**Input data needed:** Brief + at least one original dataset. **Time saved vs manual:** ~40 min per outline.

Posts that cite living experts rank better and survive HCS updates. Cold emails asking for quotes get <5% response rates. This prompt drafts the outreach as a pull request — specific, easy to say yes to.

**The prompt:**

``` You are an SEO writer drafting a quote request email to a named expert. INPUT: - Expert name + role + company - Public artifact they wrote/said that connects to the post topic (URL + quote) - Target post title + one-paragraph angle - Specific question (1 sentence, narrow) - Deadline (date) OUTPUT: a 90-130 word email with this structure: - Subject line (under 50 chars, references their artifact) - Greeting by first name - Sentence 1: cite their specific artifact with the quote - Sentence 2: state the post angle and why their perspective matters here - Sentence 3: the single specific question, with word limit ("in 2-3 sentences") - Sentence 4: deadline + how the quote will appear (link back to their work + headshot) - Signoff with your name + outlet Rules: - Do not flatter ("I love your work" is banned). - The question must be answerable in 60 seconds of thinking. - Promise a backlink, not money. ```

**Why it works:** The "answerable in 60 seconds" constraint is what drives reply rates from <5% to ~25% in tested outreach (per Buzzstream's 2024 outreach benchmark).

**Sample output:** A 112-word email subject-lined "Re: your 2024 churn-cohort post" with a single question about 90-day cohort retention, a Friday deadline, and a backlink promise.

**Input data needed:** Expert artifact URL + one specific question. **Time saved vs manual:** ~20 min per email.

Hand-built schema breaks. Schema generators produce stale templates. This prompt reads the draft and emits Article, FAQ, and HowTo (if applicable) JSON-LD that validates against Google's Rich Results Test.

**The prompt:**

``` You are a structured-data engineer generating schema.org JSON-LD for a finished blog post. INPUT: - Full post HTML or markdown - Author name, role, image URL - Publisher name, logo URL - Canonical URL - Publish date + last-modified date (ISO 8601) OUTPUT: - An Article schema object with all required + recommended fields - A FAQPage schema object IF the post has 3+ Q-shaped H3s - A HowTo schema object IF the post has numbered steps with action verbs - A list of properties you set vs left null, with reasoning Rules: - Use the @graph syntax to nest multiple schemas under one script tag. - Author must be Person, not Organization. - Do not invent fields not present in schema.org/Article spec. - FAQPage entries must be verbatim Q+A from the post, not paraphrased. ```

**Why it works:** The "verbatim Q+A" rule prevents the most common rich-snippet rejection per Google Search Central's structured-data guidelines.

**Sample output:** A 60-line JSON-LD block with @graph nesting Article + FAQPage, every FAQ pulled verbatim, a reasoning section noting that HowTo was skipped because the post lacks action-verb numbered steps.

**Input data needed:** Finished draft + author/publisher metadata. **Time saved vs manual:** ~30 min per post.

Most internal-link tools either over-link (every keyword match) or under-link (only exact title matches). This prompt produces a deliberate plan: 4–8 links per 2,000 words, each one defended.

**The prompt:**

``` You are an internal-link strategist planning links from a new post. INPUT: - Finished draft (markdown) - Sitemap.xml or URL list with: URL, post title, target query, published date - Target query of the new post For each potential internal link, output: { "anchor_text": "<the exact phrase in the new draft>", "target_url": "<URL from sitemap>", "link_strength": "strong | medium | weak", "reason": "<one sentence — why this link helps the reader, not the SEO>", "position_in_post": "intro | body | faq | conclusion" } Global output: { "recommended_links": <4-8 per 2000 words of draft>, "orphan_pages_referenced": <pages from sitemap that gained an inbound link>, "thrash_warning": "<flag if any target URL would receive a 3rd+ link from this site this month>" } Rules: - Reason must be reader-centric, not SEO-centric. - Do not link to the homepage. - Do not link the same target URL twice in the same post. ```

**Why it works:** The reader-centric reason rule pushes against the keyword-stuffed anchor pattern that Google's link-spam policy flags.

**Sample output:** 6 internal links proposed, 1 thrash warning on a category page that already had 4 inbound links this month, 2 newly-linked orphan pages.

**Input data needed:** Sitemap.xml + finished draft. **Time saved vs manual:** ~25 min per post.

Semantic expansion is what entity-coverage SEO requires — but naive expansion produces a keyword soup. This prompt expands by entity relationship, not by surface co-occurrence.

**The prompt:</strong>

``` You are an entity-SEO analyst expanding the semantic keyword set for a target query. INPUT: - Target query: <text> - Primary intent (from prompt #1) - Required entities (from prompt #1) - Top 3 ranking pages: paste full body text OUTPUT (table): | Semantic term | Entity relationship to target | Frequency in top-3 | Recommended placement | |---------------|-------------------------------|--------------------|-----------------------| Rules: - Term must appear in 2+ of the top 3 ranking pages, OR be an obvious sub-entity (e.g. for 'churn analysis' → 'cohort retention'). - Entity relationship is one of: synonym, sub-entity, parent-entity, related-tool, related-metric, related-method. - Placement is one of: H2, body paragraph, alt text, FAQ. - Cap output at 25 terms — quality over quantity. - Reject any term that would require unnatural phrasing to include. ```

**Why it works:** The 2+ appearance threshold and the explicit entity-relationship taxonomy beat naive co-occurrence expansion on accuracy. The 25-term cap prevents stuffing.

**Sample output:** A 22-row table where each term has a defensible relationship to the target query, frequency counts from the top-3 corpus, and explicit placement guidance.

**Input data needed:** Target query + top-3 page bodies. **Time saved vs manual:** ~35 min per post.

Checklists slow writers down. A checklist diff — "here's what's missing, in 30 seconds" — does not.

**The prompt:**

``` You are an on-page SEO auditor checking a finished draft against a 14-point checklist. INPUT: - Finished draft (markdown or HTML) - Target query - Sitemap + internal-link plan from prompt #6 For each checklist item, output: { "item": "<checklist item>", "status": "pass | fail | n/a", "finding": "<one sentence — what is or isn't present>", "fix": "<one sentence — the specific edit to make, or 'no action'>" } Checklist items: 1. Target query in H1 2. Target query in first 100 words 3. Target query in URL slug 4. Target query in meta title (under 60 chars) 5. Meta description present (under 160 chars) and includes target query 6. Single H1 7. H2s use sentence case and answer a question OR state a specific claim 8. At least one image with descriptive alt text containing a semantic variant 9. Internal links: 4-8 per 2000 words 10. External links: at least 3 to authoritative sources 11. Reading grade: 8th-10th grade per Hemingway or equivalent 12. JSON-LD present and valid (presence check only) 13. Featured snippet target paragraph or list present 14. Author byline + bio paragraph linking to expertise proof Rules: - Status of 'pass' must cite the specific element that passes. - Status of 'fail' must include the exact one-line fix. - Do not invent missing content — only report what is or isn't in the draft. ```

**Why it works:** The 14-point list maps to the on-page factors that Ahrefs' 2025 ranking-correlation study found to still matter post-HCS. Diff-not-checklist UX is what makes it usable.

**Sample output:** 14 rows, 11 passing, 3 failing with one-line fixes — meta description 178 chars (cut by 18), only 2 external links (add 1 to Anthropic docs), no alt text on hero image.

**Input data needed:** Finished draft. **Time saved vs manual:** ~30 min per post.

Helpful Content System updates penalize site-wide thin content. Pruning is the lever — but knowing what to cut is the work. This prompt produces a defensible pruning list.

**The prompt:**

``` You are an SEO editor recommending content-pruning candidates. INPUT (CSV or pasted table): - For each URL on the site: URL, target query, current organic clicks (trailing 90 days), current organic impressions (T90), publish date, last-update date, word count, # of backlinks, # of internal links pointing in. For each URL, output: { "url": "<URL>", "recommendation": "keep | refresh | consolidate | redirect | delete", "reason": "<one sentence>", "consolidation_target": "<URL if recommendation is consolidate, else null>", "redirect_target": "<URL if recommendation is redirect, else null>", "priority": "high | medium | low" } Global output: { "total_urls": <number>, "recommended_deletions": <number>, "recommended_redirects": <number>, "estimated_thin_content_ratio_before": <0-1>, "estimated_thin_content_ratio_after": <0-1> } Rules: - 'delete' requires: 0 clicks in T90, 0 backlinks, 0 inbound internal links, OR published >18 months ago with <50 impressions T90. - 'consolidate' requires another URL on the site with overlapping target query and stronger metrics. - 'redirect' is for thin pages with at least 1 backlink — preserve the link equity. - Never recommend delete for a page with backlinks; redirect instead. ```

**Why it works:** The hard rules (0/0/0 for delete, preserve backlinks via redirect) match the pruning methodology Ahrefs documented in 2024 and the post-HCS recovery patterns SearchEngineLand reported.

**Sample output:** 412-URL site reduced to 287 recommended-kept; 73 delete, 31 redirect, 21 consolidate; thin-content ratio drops from 38% to 11%.

**Input data needed:** GSC export + word counts + backlink counts. **Time saved vs manual:** ~3 hours per site audit.

Some posts go stale silently. Traffic decay is the signal — but reading decay across 200 posts in GSC is the bottleneck. This prompt scores refresh priority from the decay curve.

**The prompt:**

``` You are an SEO editor scoring refresh priority across a content library. INPUT (CSV): - For each URL: clicks per month for the trailing 12 months, current target query, last-update date, current rank for target query (if known). For each URL, output: { "url": "<URL>", "decay_slope": "<one of: rising, flat, mild_decay, steep_decay, dead>", "decay_evidence": "<one sentence citing specific month-over-month numbers>", "refresh_priority": "<0-100 integer>", "recommended_refresh_action": "<one of: update stats, expand entity coverage, change angle, rewrite intro, full rewrite, no action>", "estimated_recovery_value": "<low | medium | high>" } Global output: ranked list of top 20 refresh candidates by refresh_priority. Rules: - Decay slope 'steep_decay' = >50% clicks lost over 6 months. - Refresh priority weights: decay severity (40%), historical peak traffic (35%), recency of last update (25%). - 'dead' (>90% loss) with no backlinks → recommend prune (defer to prompt #9), not refresh. - Do not recommend rewriting evergreen posts with rising trends. ```

**Why it works:** The explicit weight formula and the dead-vs-prune handoff stop the model from suggesting refreshes on posts that should be deleted. The decay-evidence requirement makes the score auditable.

**Sample output:** Top 20 ranked, with the #1 candidate being a 2024 statistics post (refresh_priority 92) — steep decay (down 64% over 6 months), high recovery value, recommended action: "update stats with 2026 numbers, expand entity coverage."

**Input data needed:** GSC trailing-12-month clicks export. **Time saved vs manual:** ~2 hours per quarterly refresh audit.

The chain that replaces the 5.5-hour naive cycle:

1. **Min 0–3.** Run prompt #1 against the SERP scrape. Output: intent + required entities. 2. **Min 3–6.** Run prompt #2 with #1's output. Output: structured brief. 3. **Min 6–10.** Run prompt #3 with the brief + your original data. Output: HCS-safe outline. 4. **Min 10–12.** (Optional) Run prompt #4 to draft expert outreach. Output: 2–3 quote-request emails. 5. **Min 12–20.** Human drafts the post against the outline (this is the load-bearing 8 minutes — the prompts do not write the post). 6. **Min 20–22.** Run prompt #7 to sanity-check semantic coverage. Output: 22 terms with placement. 7. **Min 22–24.** Run prompt #8 (on-page diff). Output: 3 fixes to make. 8. **Min 24–26.** Apply fixes. 9. **Min 26–28.** Run prompt #5 to generate schema. Output: JSON-LD block. 10. **Min 28–32.** Run prompt #6 to plan internal links. Apply 4–8 links. Publish.

Prompts #9 (pruning) and #10 (refresh priority) run on a quarterly cadence against the whole site, not per-post. They're the cleanup pass.

Cost: ~$0.40–$0.80 per publish cycle in Claude Sonnet 4.5 tokens per Anthropic's pricing page. At 50 posts/month, that's $20–$40 in model spend — the marginal cost is the human's 12 minutes of drafting plus 8 minutes of review, not the model.

<a href="https://aipromptshub.co/chatgpt-prompt-generator?utm_source=aipromptshub&utm_medium=blog&utm_campaign=seo-prompts-chain" style={{display:"inline-block",padding:"10px 18px",background:"#0a66ff",color:"white",borderRadius:"6px",textDecoration:"none",fontWeight:"bold",marginTop:"12px"}}> Get the chained SEO template → </a>

Three break-points worth naming. First, the chain assumes you have original data to feed prompt #3. Without it, the post defaults to the same generic AI scaffold every other site is publishing — HCS will catch it. Second, the chain optimizes for cycle time, not depth. A 32-minute post will not beat a 25-hour human investigation on the same query (per Backlinko's 2024 study); it will only beat the 90-second naive prompt by 10x in cycle time and roughly 4x in ranking probability based on the entity-coverage and HCS-safety gains.

Third, the SERP scrape that feeds prompts #1, #2, and #7 must be fresh. SERP volatility post-March 2024 Google update means a 2-week-old scrape is stale. SearchEngineLand's 2025 SERP-volatility tracker showed median rank-position changes of 4.2 positions per 30 days in informational queries during 2025. Pull the scrape the day you brief, not the week before.

<a href="https://aipromptshub.co/blog-post-outline?utm_source=aipromptshub&utm_medium=blog&utm_campaign=seo-prompts-limits" style={{display:"inline-block",padding:"10px 18px",background:"#0a66ff",color:"white",borderRadius:"6px",textDecoration:"none",fontWeight:"bold",marginTop:"12px"}}> Build a fresh SERP-brief prompt → </a>

### Does the 10x claim hold against human-written content?

No, and we're explicit about that. The 10x is measured against the naive single-prompt Claude baseline using Ahrefs' 2025 median (4.8 hours per AI-assisted post) and SEMrush's 2024 State of Content Marketing (5.2 hours). Human-only writing at 25–40 hours per long-form post still earns more organic traffic per Backlinko's 2024 study; the chain is a productivity tool, not a quality replacement.

### Which Claude model should I use for each prompt?

Sonnet 4.5 is the default for prompts 1, 2, 6, 7, 8, 9, and 10 — fast, cheap, structurally accurate. Opus 4.7 is worth the cost premium for prompts 3 (HCS-safe outline) and 5 (schema generation) where structural depth matters more than latency. See Anthropic's model selection guide for the current generation.

### Will this chain trigger Google's Helpful Content System penalty?

Not on its own — but only if you feed prompt #3 real original data. The HCS guidance from Google Search Central targets content written for search engines rather than for people. The chain produces a structured scaffold; the original-data injection at prompt #3 is what differentiates the output from naive AI content. Skip that input and you lose the HCS-safety property.

### How do I scrape the top 10 SERP without violating Google's ToS?

Use a SERP API (SerpApi, DataForSEO, Ahrefs/SEMrush exports) rather than scraping search results pages directly. Both Ahrefs and SEMrush expose API endpoints that return the top 10 with H-tag structure for any query. Google Search Central's webmaster guidelines cover the policy boundary.

### Can I use these prompts with ChatGPT or Gemini instead of Claude?

Prompts 1, 2, 6, 7, 8, and 9 work well on GPT-4-class and Gemini Pro models — they're structured-output tasks. Prompts 3 (HCS outline) and 5 (schema) are where Claude's verbatim-constraint adherence wins per Anthropic's documentation on rule compliance. For prompts 4 and 10, model choice matters less than the prompt design itself.

### Do I need a Claude API key, or will Claude.ai web work?

The web interface works for one-off posts. For the chain at scale (10+ posts/week), pipe outputs as JSON between prompts via the API — the structured input/output shapes only pay off when the next prompt can ingest the prior output without manual reformatting.

### How often do the prompts themselves need updating?

Quarterly. SERP-format shifts, schema spec updates, and HCS guidance changes from Google Search Central drive most of the maintenance. Re-test prompts 1, 2, and 8 against a fresh SERP every 90 days; the other prompts are stable across cycles.

- Ahrefs 2025 content-production benchmark — median AI-assisted cycle time. - SEMrush 2024 State of Content Marketing — content-production cycle times. - Backlinko 2024 long-form content study — human vs AI content traffic comparison. - Google Search Central — Helpful Content System guidance — HCS policy reference. - Google Search Central — Creating helpful content (E-E-A-T) — relevance signals reference. - Google Search Central — Structured data: Article — schema policy reference. - Google Search Central — Spam policies — link-spam policy reference. - Anthropic prompt engineering documentation — Claude prompt best practices. - Anthropic model documentation — Sonnet/Opus selection. - Anthropic pricing page — token cost reference. - SearchEngineLand — post-HCS recovery analysis — post-update recovery patterns. - Buzzstream 2024 outreach benchmark — reply-rate data.

---