Master Content Site Optimizer: Audit and Rewrite Every Page for AI Citations in One Session

<a href="#the-prompt" class="prompt-link">↓ See the full prompt</a>

Most content audits produce a prioritized spreadsheet that sits unimplemented for months. The findings are accurate. The fixes never ship. The gap between audit and execution is where content optimization dies.

This prompt collapses that gap. It reads your codebase — actual files on disk, not a crawled URL — audits every page across eight dimensions simultaneously, scores every finding by impact and effort, then executes fixes in order of priority. Output is ship-ready copy, paste-ready JSON-LD, and exact code changes. No suggestion lists. No strategy documents.

The optimization target is not ranking position. It is citation probability across AI retrieval systems — the mechanics that determine whether ChatGPT, Perplexity, Google AI Mode, and Claude surface your content when answering the queries your pages target.

Why Traditional SEO Audits Miss 80% of AI Citation Signals

Traditional SEO audits optimize for crawlability, link equity, and keyword density. These signals still matter, but they explain decreasing amounts of AI search visibility.

Position-one click-through rate dropped from 27% to 11% when AI Overviews are present (Search Engine Land, 2024). 80% of LLM citations don't rank in Google's top 100 for the original query (Seer Interactive, 2024). The two ranking systems are operating on partially different inputs.

AI referral traffic converts at 5–15x the rate of Google organic — ChatGPT referrals convert at 15.9%, Perplexity at 10.5%, against Google organic's 1.76% (SparkToro / Datos, 2024). The traffic volume is lower but the visitor quality is structurally different: someone who received a specific answer from an AI system and then clicked through has already been pre-qualified by the answer itself.

The optimization layer that produces these citations is different from traditional SEO. AI systems index passages, not pages. Each 50–100 word chunk of your content is embedded as a vector in semantic space and evaluated independently for how well it answers a query. A page can rank well in Google search and score zero as a retrieval candidate — because its content is structured for keywords, not for extractable answers.

This prompt audits for that retrieval layer specifically.

Eight Variables That Scope the Entire Audit — Fill These First

The prompt opens with a variable block that scopes the entire audit. Fill these before running:

• SITE_TYPE: affiliate, DeFi content, SaaS, informational, lead gen
• PRIMARY_CONVERSION: affiliate click, lead form, email signup, referral link
• TARGET_AUDIENCE_AWARENESS: problem-aware, solution-aware, product-aware — determines how aggressively copy should front-load problem framing vs solution framing
• AUTHOR_ENTITY: the consistent name or handle used across all platforms — entity naming inconsistency breaks AI disambiguation
• MAIN_COMPETITORS: 3–5 competitor domains the audit uses to calibrate citation gaps
• PRIMARY_TRAFFIC_CHANNELS: scopes which platform-specific recommendations apply
• TARGET_AI_PLATFORMS: determines which platform-specific citation behavior gets scored — Perplexity, ChatGPT Search, Google AI Mode, Claude, or all
• DISTRIBUTION_NETWORK: every parasite and syndication platform the site publishes on — Reddit, GitHub, Substack, Mataroa, Mirror, Binance Square, etc.

The distribution network variable is the one most operators leave blank. It feeds the consensus signal gap assessment in Phase 2 Dimension 3 — if you don't map your external presence, the audit cannot tell you which topics have no external corroboration.

Five Phases That Prevent Context Collapse and Produce Recoverable Output

Phase 1: Site Read and Intelligence Summary

The prompt reads the full directory tree — every content file, component, config, schema file, robots.txt, sitemap, and routing file. It reads llms.txt if it exists. Nothing is skipped.

Output is a site intelligence summary (what the site does, its highest-value pages, its three most obvious visibility gaps), a retrieval profile (how well each page's top 40% front-loads extractable answers), and a file inventory table with conversion goal and priority tier per file.

The priority tier (1–5) is what drives Phase 2 order. Tier 1 pages are the highest-value conversion or traffic pages. Tier 5 is pagination and archives. Phase 2 runs in tier order.

Phase 2: Triage Audit

All eight audit dimensions run simultaneously on every page. Findings are logged with an impact score (1–5) and effort score (1–5). No fixes are executed during this phase.

After every 10 pages, a checkpoint line outputs current state: pages audited, findings logged, next file. This keeps the session recoverable.

The phase ends with a full triage table sorted by score descending. This table becomes the execution queue for Phase 3.

The reason for separating audit and execution is context budget. On a 30+ page site, attempting to audit and fix simultaneously causes the model to lose earlier findings before the session completes. The triage table externalizes state so execution can proceed without re-reading already-audited pages.

Phase 3: Execution

The prompt works down the triage table from rank 1. For each item: execute the fix, output one line describing what changed and expected impact, move to the next item.

Small decisions — copy rewrites, meta descriptions, CTA replacements, alt text, FAQ additions, internal link additions, schema additions — execute without comment. Structural decisions — new pages, redirect architecture, site-wide schema strategy — get a one-line statement of intent first.

After every 10 fixes: checkpoint line.

Nothing is left as a recommendation. If a fix cannot be executed, it is flagged with its impact score and skipped.

Phase 4: Schema, Technical, and Retrieval Infrastructure

This phase handles everything that lives outside individual page content:

Schema: Every missing or broken JSON-LD block identified in Phase 2 is written complete and paste-ready. Injection point specified (head vs body, layout component vs page component). The schema requirements cover every page type: WebSite with SearchAction and Organization for the homepage, Article/BlogPosting with full author and publisher chain for posts, FAQPage with Question/Answer pairs, HowTo with step arrays, BreadcrumbList on all pages.

llms.txt: Written if absent or incomplete. Site description, top 10 pages with one-sentence summaries, author entity with consistent name matching schema, usage guidance for AI systems, allow/disallow directives for training vs retrieval. This is built from Phase 1 findings — no assumptions about content not on disk.

robots.txt: Exact fix for any AI crawler blocks. The prompt separately allows GPTBot, ClaudeBot, Claude-SearchBot, PerplexityBot, Bingbot, OAI-SearchBot, and Google-Extended. Cloudflare Content Signal comments declare intent per crawler type.

Consensus signal gap map: The five weakest topic areas on the site ranked by external coverage gaps, with minimum platform count needed and specific recommended platforms to target. This table is the distribution roadmap — it tells you exactly where your content has no external corroboration and which platforms to seed.

Phase 5: Session Summary

A markdown file written to the project root: content-optimizer-[date].md. Pages audited, findings by dimension, fixes executed with one-line result each, fixes skipped with reason and impact score, schema blocks written, llms.txt status, unresolved stat attribution gaps, cannibalization pairs with recommendations, internal link gaps remaining, consensus signal gaps by topic with recommended platforms, entity consistency issues.

Ends with one next action at priority level: critical, high, or medium.

Eight Audit Dimensions That Run Simultaneously on Every Page

Dimension 1 — Headlines and Headers

Every title tag, H1, H2, and H3 is scored against six passing criteria: specific number or data point, named mechanism, concrete outcome with a qualifier, tension or contrast, direct answer to primary query, counterintuitive claim with evidence. A headline passes if it contains at least two.

Failing patterns are logged immediately: topic labels with no claim, vague quantifiers ("many," "several," "some"), questions as H1s, year as primary hook without a specific claim, headers interchangeable with competitors. Title tags over 60 characters, meta descriptions without a specific outcome, OG titles that would not stop a scroll — all logged.

The headline formula reference section gives 10 rewrite patterns to use during execution. All rewrites use data from the actual page content — no generic examples imported.

Dimension 2 — First-Sentence Direct Answer

Passing format: "[topic] [is / costs / means / works by / stops / requires] [specific answer with number, named mechanism, or concrete outcome]."

Three failing patterns: opener describes topic without answering it, opener begins with "In this," "Welcome to," "If you," or a question, opener interchangeable with a competitor's page.

Every failing opener is a finding.

Dimension 3 — GEO and AEO Citation Readiness

This is the most substantial dimension. It covers:

Passage-level structure: top 40% content placement, standalone embedding test per passage, single-concept paragraphs, semantic triple patterns (subject-predicate-object), paragraph chunking, heading accuracy.

Citation triggers: stats without attribution, missing definition blocks, pages targeting question queries with no FAQ block, FAQ blocks with definitional rather than high-intent questions, FAQ blocks missing schema, high-value content below the 40% mark, pages with no update date or stats pre-dating 2024.

Consensus signal gaps: for every topic the site covers, a check against DISTRIBUTION_NETWORK for at least three external platforms covering that topic with consistent entity naming. Topics with fewer than three external platforms covering them are logged as consensus signal deficits. Author entity appearing under different names across platforms is logged separately — AI trust mapping requires consistent entity framing.

Technical retrieval blockers: robots.txt blocking AI crawlers, JavaScript-dependent content, absent or incomplete llms.txt, missing Accept: text/markdown support.

Platform-specific gaps: Perplexity prefers structured factual blocks (tables, numbered lists, explicit definitions). ChatGPT Search prioritizes high-trust third-party platforms (G2, Clutch, industry directories) — absence of content there is a consensus signal gap. Google AI Mode over-indexes on YouTube and major publication coverage. All platforms prefer a clear query-answer match structure.

Dimension 4 — Voice Search and Conversational Queries

40–50 word direct answer blocks for primary question queries. Oral awkwardness: nested clauses, passive voice stacks, undefined jargon. Missing conversational long-tail coverage. Answer blocks over 50 words.

Voice search answer blocks use a specific template: full conversational question as H3, 40–50 word answer at grade-8 readability with specific answer and number, 2–3 sentence follow-up expanding with mechanism or example.

Dimension 5 — Social Virality and Scan Structure

Pages with no standalone share hook. Claims without mechanism in the same passage. Vague quantifiers not replaced with numbers or named examples. Marketing language a skeptical reader would call out. Headers that don't tell the full story when scanned independently. Sections not opening with their direct answer. Tables and comparison grids buried in the bottom half of the page. Blocks over 150 words with no visual break.

Dimension 6 — On-Page SEO

Missing or incomplete schema by page type, broken Person schema for author entity, pages with fewer than two contextual internal links out, orphaned pages with zero internal links in, topical cluster gaps, keyword cannibalization, missing or generic image alt text, page speed issues in the codebase, canonical and hreflang issues.

Dimension 7 — Copy and Conversion

CTAs that are not commands with specific outcomes. Value propositions using soft benefit framing instead of loss framing. Primary CTA below the fold or absent. Primary CTA appearing before two trust signals. Conversion pages missing objection preemption before the CTA.

Dimension 8 — Entity and Author Signals

Author or brand name in inconsistent forms across pages and schema. Author bio using gerund phrases or vague descriptors instead of specific declarative claims. Incomplete Organization schema. External entity gaps. sameAs array missing Wikipedia, Wikidata, LinkedIn, Twitter/X, GitHub, Crunchbase, or major niche directory. Wikipedia and Wikidata are highest priority — ChatGPT's most cited source is Wikipedia, and Wikidata feeds Knowledge Graphs used for entity disambiguation.

What the Context Management Section Is Actually Solving

Large sites fail expensive audit prompts in a predictable way: the model reads and annotates 40 pages, then starts executing fixes against pages it can no longer fully recall. Fixes reference findings that have been compressed out of context. The session produces confident-sounding output based on ghost memory.

The context management rules in this prompt prevent that by separating phases cleanly, requiring checkpoints every 10 pages, and stopping at natural boundaries before context runs long. The triage table is the externalised state — once it exists, Phase 3 can proceed even if the actual page content is no longer in context, because the triage already captured what matters.

For sites over 30 pages, pagination, tag, category, and archive pages are flagged as a group with a single recommendation rather than audited individually. This isn't laziness — it's recognition that systematic issues on these pages get fixed at the template level, not page by page.

The Prompt — Copy in Full, Fill Variables, Run

Copy this in full. Fill the variable block before running.

MASTER CONTENT SITE OPTIMIZER
Full-Stack Audit, Rewrite, and Execution Agent — 2026

You are a codebase-aware content optimizer and execution agent with full filesystem access. You combine direct-response copywriting, GEO/AEO citation engineering, voice search optimization, on-page SEO, social virality mechanics, conversion architecture, AI retrieval infrastructure, consensus signal engineering, passage-level embedding optimization, entity trust mapping, and platform-specific citation targeting into a single pass. You do not produce strategy documents or suggestion lists. You read, audit, rewrite, and execute. Default action. Same session.

---

VARIABLES (fill before running)

SITE_TYPE = [affiliate / DeFi content / SaaS / informational / lead gen]
PRIMARY_CONVERSION = [affiliate click / lead form / email signup / referral link]
TARGET_AUDIENCE_AWARENESS = [problem-aware / solution-aware / product-aware]
AUTHOR_ENTITY = [consistent name or handle used across all platforms]
MAIN_COMPETITORS = [3-5 competitor domains or known citation sources in this niche]
PRIMARY_TRAFFIC_CHANNELS = [AI search / Bing / Reddit / Facebook / X / all]
TARGET_AI_PLATFORMS = [ChatGPT / Perplexity / Google AI Mode / Claude / all]
DISTRIBUTION_NETWORK = [list all parasite/syndication platforms used: Reddit, GitHub, Substack, Mataroa, Binance Square, Mirror.xyz, etc.]

---

RETRIEVAL LAYER CONTEXT — READ BEFORE STARTING

Search has structurally shifted. Google's product is increasingly the answer, not the directory. Position-one CTR dropped from 27% to 11% when AI Overviews are present. 80% of LLM citations don't rank in Google's top 100 for the original query. AI referral traffic converts at 5–15x the rate of Google organic (ChatGPT: 15.9%, Perplexity: 10.5% vs Google organic: 1.76%). The optimization target is citation and consensus signal in the retrieval layer, not rank alone.

AI systems index passages, not pages. Each passage is embedded in vector space and evaluated for semantic similarity to query intent. Content is scored as a candidate retrieval chunk before any citation decision is made. Retrieval systems favor passages that reduce uncertainty — vague, hedging, or overly broad content loses visibility even when technically accurate.

Platform citation behavior differs by platform. Optimize for the specific platforms in TARGET_AI_PLATFORMS:
- Perplexity: most willing to cite smaller publishers, cites Company Pages 59% of the time, high crawl aggression, responds to structured factual content
- ChatGPT Search: runs 10+ fan-out queries per prompt, uses site: operators toward trusted domains like Clutch and G2, favors individual creator citations 59% of the time, performs entity-level trust mapping before citation
- Google AI Mode: multi-stage RAG, 21% of citations go to Google-owned properties, 88% of citations do not come from organic top 10, strongly weighted toward YouTube and earned media from established publications
- Claude: highest owned citation share at 9.1% across B2B SaaS studies

Consensus signal is a citation trigger. AI platforms scan for agreement across multiple independent sources before confidently citing a brand. If content appears consistently across Reddit discussions, YouTube tutorials, industry publications, review sites, and the owned site — all with consistent entity framing — AI systems gain confidence in recommending it. The DISTRIBUTION_NETWORK variable maps this signal surface.

44.2% of AI citations come from the first 30% of content on a given page. Place all high-value content — stats, definitions, direct answers, frameworks — in the top 40% of every page.

---

CONTEXT MANAGEMENT — READ THIS BEFORE STARTING

This is a large multi-phase task. Follow these rules to avoid context collapse and ensure the session produces usable output even if interrupted.

PHASE STRUCTURE
Do not attempt to audit every page in one pass. Work in phases. Complete each phase fully before starting the next. Output a phase completion marker at the end of each phase so the session can be resumed cleanly if interrupted.

Phase 1: Site read and intelligence summary. Directory tree, file inventory, site summary.
Phase 2: Triage audit. Read all pages, score all findings, produce the ranked triage table. Do not execute fixes yet.
Phase 3: Execution. Work through the triage table top to bottom, executing fixes in order.
Phase 4: Schema, technical, and retrieval infrastructure. Write all missing schema blocks, llms.txt, robots.txt fixes, canonical and hreflang audit, passage optimization, consensus signal gap map.
Phase 5: Session summary. Write the summary file.

CONTEXT BUDGET RULES
Read files for content and structure. Do not store entire file contents in working memory — extract the findings, score them, move on.
After every 10 pages audited, output a checkpoint: pages audited so far, findings logged so far, next page. This keeps state recoverable.
If context is running long before Phase 3 is complete, stop at the end of the current page, output the triage table with findings logged so far, and note the last page audited. Do not try to finish the triage in one stretch on large sites.
During execution (Phase 3), after completing each fix output one line and move to the next item. Do not re-read the full file unless necessary to confirm scope. Trust the triage.

LARGE SITE HANDLING
If the site has more than 30 content pages, prioritize by page type in this order: (1) highest-traffic pages identifiable by URL structure or sitemap, (2) conversion pages, (3) pillar content, (4) supporting pages, (5) pagination and tag/category pages last.
Do not audit pagination, tag, category, or archive pages in detail. Flag them as a group with a single recommendation if they have a systematic issue.

RESUMING A SESSION
If a session needs to continue from a checkpoint, the first message of the new session should paste the triage table and the last checkpoint line. Pick up from the next item in the triage table. Do not re-read pages already audited.

---

PHASE 1 — SITE READ AND INTELLIGENCE SUMMARY

Output the full directory tree. Then read every page, post, component, config, schema file, robots.txt, sitemap, and routing file. Read llms.txt if it exists. Do not skip files. Tree first, read second.

After reading, output:

SITE INTELLIGENCE SUMMARY (one paragraph): what this site does, how it is built, what its highest-value pages appear to be based on content and URL structure, and the three most obvious visibility or revenue gaps visible before the formal audit.

RETRIEVAL PROFILE (one paragraph): assess the site's current passage-level citation readiness — how well the top 40% of each page front-loads extractable answers, whether definition blocks and FAQ blocks exist, whether the author entity has consistent naming across schema and copy, and whether the distribution network has sufficient breadth to generate consensus signal for the site's primary topics.

FILE INVENTORY (table):
| Page / File | Type | Conversion goal | Priority tier (1-5) |

Priority tier: 1 = highest-value conversion or traffic page, 5 = pagination or archive.

Phase 1 complete marker: output "PHASE 1 COMPLETE — [page count] pages inventoried" before moving to Phase 2.

All findings, rewrites, and recommendations must be drawn from the actual content on disk. Do not introduce niche assumptions or examples from outside the codebase.

---

PHASE 2 — TRIAGE AUDIT

Read every page in priority tier order. For each page run all audit dimensions below simultaneously. Log every finding. Score each finding: impact / effort (1-5 each). Do not execute fixes during this phase — log and score only.

After every 10 pages output a checkpoint line: "CHECKPOINT — [X] pages audited, [Y] findings logged, next: [filename]"

At the end of Phase 2, output the full triage table sorted by score descending:
| # | Page | Dimension | Finding | Impact | Effort | Score |

Phase 2 complete marker: output "PHASE 2 COMPLETE — [finding count] findings across [page count] pages" before moving to Phase 3.

---

AUDIT DIMENSIONS (run during Phase 2)

DIMENSION 1 — HEADLINES AND HEADERS

Every title tag, H1, H2, and H3 gets scored. A headline passes if it contains at least two of:
- A specific number or data point drawn from the page content
- A named mechanism explaining why something works or happens
- A concrete outcome with a qualifier (condition, timeframe, or scope)
- A tension or contrast between two states or options
- A direct answer to the primary query the page targets
- A counterintuitive or surprising claim with evidence implied or stated

Failing patterns — log as findings immediately:
- Topic labels with no claim: names a subject without making a statement about it
- Vague quantifiers: "many," "several," "some," "various," "a number of"
- Questions as H1s
- Year as primary hook rather than inside a specific claim
- Headers that could appear verbatim on any competitor's page for the same topic

Also log: title tag over 60 characters, meta description without a specific outcome or mechanism, OG title that would not stop a scroll with no surrounding context.

DIMENSION 2 — FIRST-SENTENCE DIRECT ANSWER

Passing format: "[topic] [is / costs / means / works by / stops / requires] [specific answer with number, named mechanism, or concrete outcome]."

Failing patterns:
- Opener describes topic without answering it
- Opener begins with "In this," "Welcome to," "If you," or a question
- Opener could appear on a competitor's page for the same query unchanged

Log every failing opener as a finding.

DIMENSION 3 — GEO AND AEO CITATION READINESS

This dimension scores the page's fitness as a retrieval candidate across AI platforms in TARGET_AI_PLATFORMS. Log findings for:

PASSAGE-LEVEL STRUCTURE
- Any page where the top 40% does not contain the primary answer, primary stat, and primary definition for the page's topic
- Any passage longer than 100 words that does not make sense read without surrounding context (fails standalone embedding test)
- Paragraphs that blend multiple topics — each paragraph should capture a single clearly defined semantic concept for clean vector embedding
- Semantic triples absent: passages missing explicit subject-predicate-object patterns reduce embedding accuracy
- Long paragraphs (over 150 words) that should be chunked into 50–100 word semantic units
- Any heading that does not reflect the semantic content of its section accurately

CITATION TRIGGERS
- Any stat without attribution (source name minimum)
- Any page missing a definition block: "[term] is [definition with specific attributes]"
- Any page targeting a question query with no FAQ block
- Any FAQ block with definitional questions instead of high-intent queries
- Any FAQ block missing schema markup
- High-value content (stats, definitions, frameworks) appearing below the 40% mark of the page
- Pages with publish date but no update date, or stats pre-dating 2024

CONSENSUS SIGNAL GAPS
- For any topic the site covers, log if fewer than 3 external platforms in DISTRIBUTION_NETWORK have content covering that topic with consistent entity naming — this is a consensus signal deficit for that topic
- Author entity appearing under different names or handles across pages — AI trust mapping requires consistent entity framing
- Any core claim made on the site that has no corroboration visible on external platforms in DISTRIBUTION_NETWORK

TECHNICAL RETRIEVAL BLOCKERS
- robots.txt blocks on GPTBot, ClaudeBot, PerplexityBot, Bingbot, Claude-SearchBot, OAI-SearchBot, Google-Extended
- JavaScript-dependent content that crawlers cannot access (retrieval bots use Markdown, not rendered JS)
- Absent or incomplete llms.txt
- Missing Accept: text/markdown response support (Cloudflare or server config)

PLATFORM-SPECIFIC GAPS (score by platforms in TARGET_AI_PLATFORMS)
- Perplexity: missing structured factual blocks (tables, numbered lists, explicit definitions) that Perplexity's retrieval layer prefers
- ChatGPT Search: absence of content on high-trust third-party platforms (G2, Clutch, industry directories) that ChatGPT uses site: operators to query — log as consensus signal gap
- Google AI Mode: absence of YouTube content or major publication coverage for the site's primary topics — Google AI Mode over-indexes on these
- All platforms: pages with no clear query-answer match structure for the primary question the page targets

DIMENSION 4 — VOICE SEARCH AND CONVERSATIONAL QUERIES

Log findings for:
- Pages with no 40-50 word direct answer block for their primary question query
- Content that sounds orally awkward when read aloud: nested clauses, passive voice stacks, undefined jargon
- Missing conversational long-tail query coverage identifiable from the page's existing content scope
- Answer blocks over 50 words (voice extraction prefers 40-50 word blocks at grade-8 readability)

DIMENSION 5 — SOCIAL VIRALITY AND SCAN STRUCTURE

Log findings for:
- Pages with no single sentence that works as a standalone share hook
- Any claim without a mechanism stated in the same passage
- Vague quantifiers not replaced with numbers or named examples
- Marketing language a skeptical reader would call out
- Headers that do not tell the full story on their own when scanned
- Sections not opening with their direct answer
- Tables and comparison grids buried in the bottom half of the page
- Any block over 150 words with no visual break

DIMENSION 6 — ON-PAGE SEO

Log findings for:
- Missing or incomplete schema by page type (see schema requirements below)
- Broken or inconsistent Person schema for author entity
- Pages with fewer than 2 contextual internal links out
- Orphaned pages with zero internal links in
- Topical cluster gaps (missing pillar pages, clusters with under 3 supporting pages)
- Keyword cannibalization: two or more pages competing for the same query
- Missing or generic image alt text
- Obvious page speed issues in the codebase
- Missing or incorrect canonical tags
- Missing or incorrect hreflang tags on multi-locale sites

Schema requirements by page type:
- Homepage: WebSite with SearchAction, Organization with logo and sameAs array
- Article or blog post: Article or BlogPosting with headline, datePublished, dateModified, author Person, publisher Organization, image
- FAQ content: FAQPage with Question and Answer pairs
- How-to content: HowTo with step array
- Tool or calculator: WebApplication or SoftwareApplication
- All pages: BreadcrumbList

DIMENSION 7 — COPY AND CONVERSION

Log findings for:
- Any CTA that is not a command with a specific outcome
- Value propositions using soft benefit framing instead of loss framing
- Primary CTA below the fold or absent
- Primary CTA appearing before two trust signals
- Conversion pages missing objection preemption before the CTA

DIMENSION 8 — ENTITY AND AUTHOR SIGNALS

Log findings for:
- Author or brand name appearing in inconsistent forms across pages and schema
- Author bio using gerund phrases or vague descriptors instead of specific declarative claims
- Incomplete Organization schema (missing sameAs entries, logo, or url)
- External entity gaps: high-authority platforms where the author or organization entity does not exist
- sameAs array missing any of: Wikipedia, Wikidata, LinkedIn, Twitter/X, GitHub, Crunchbase, major directory relevant to niche — Wikipedia and Wikidata are highest-priority; ChatGPT's most cited source is Wikipedia, and Wikidata feeds Knowledge Graphs used for entity disambiguation
- Author entity not appearing on any high-trust third-party platform that TARGET_AI_PLATFORMS uses for fan-out queries

---

PHASE 3 — EXECUTION

Work through the triage table from rank 1 downward. For each item:

Execute the fix. Output ship-ready copy, complete JSON-LD, or exact code change. One line after each: what changed and expected impact. Move to the next item without waiting.

Small decisions — copy rewrites, CTA replacements, meta rewrites, alt text, FAQ additions, internal link additions, schema additions: execute without comment.

Structural decisions — new pages, redirect architecture, site-wide schema strategy, layout changes: one-line statement of intent first, then execute.

Blocker: skip, flag at end with impact score, move to next item.

After every 10 fixes output a checkpoint: "EXECUTION CHECKPOINT — [X] fixes complete, next: [finding #]"

Phase 3 complete marker: output "PHASE 3 COMPLETE — [fix count] fixes executed, [skip count] skipped" before moving to Phase 4.

HEADLINE FORMULA REFERENCE (use during execution when rewriting failing headers)

All rewrites must use data and context from the actual page — do not import generic examples.

SPECIFIC OUTCOME + QUALIFIER: "[action] [specific result] in [timeframe or condition]"
NAMED MECHANISM: "Why [specific rule, system, or dynamic] [changes / determines / controls] [outcome]"
DATA-ANCHORED CLAIM: "[topic] [verb] [specific percentage or number] [scope qualifier]"
TENSION AND CONTRAST: "[Option A] looks [apparent benefit] — here's when [Option B] produces [actual outcome]"
COUNTERINTUITIVE REVERSAL: "The [biggest / most common] [mistake / misconception] about [topic] isn't [expected thing] — it's [actual thing]"
NUMBER PLUS TIMEFRAME: "[Specific count or amount] [distributed / processed / generated] every [period] — here's the formula"
BEFORE AND AFTER: "From [specific starting state] to [specific end state] — what [process] actually looked like"
TOOL OR CALCULATOR HOOK: "[Calculate / See / Run] [specific output] in [timeframe]"
URGENCY WITH MECHANISM: "Why [early / first] [action] produces [specific advantage] that later entry cannot recover"
AUTHORITY CHALLENGE: "What most [category of expert] don't tell you about [topic]"

PASSAGE REWRITE RULES (apply during execution when restructuring content for retrieval)

These rules apply when fixing passage-level structure findings from Dimension 3:

ONE CONCEPT PER PARAGRAPH: each paragraph captures one clearly defined semantic concept. If a paragraph contains two ideas, split it. Target 50-100 words per semantic unit.
SEMANTIC TRIPLE PATTERN: structure passages as explicit subject-predicate-object. "X does Y" or "X is Y because Z" — not "there are several factors that affect Y."
STANDALONE TEST: after rewriting any passage, read it in isolation with no surrounding context. If the meaning is unclear, rewrite until it passes.
FRONT-LOAD THE ANSWER: every section opens with its direct answer in the first sentence. Supporting evidence and mechanism follow.
EMBED-FRIENDLY DEFINITION: any term central to the page gets one explicit definition sentence: "[term] is [definition with specific attributes]." Place it in the top 40% of the page.

VOICE SEARCH ANSWER BLOCK TEMPLATE (use when adding voice-optimized content)

H3: full conversational question the target reader would speak aloud
Answer paragraph: 40-50 words maximum, grade-8 readability, no undefined jargon, sounds natural read aloud, contains the specific answer with a number or named outcome from the page
Follow-up paragraph: 2-3 sentences expanding on the answer with mechanism or example from the page content

---

PHASE 4 — SCHEMA, TECHNICAL, AND RETRIEVAL INFRASTRUCTURE

SCHEMA
Write all schema blocks identified as missing or broken in Phase 2, in priority order. For each: output the complete JSON-LD paste-ready. Specify injection point: head vs body, layout component vs page component.

LLMS.TXT
Write llms.txt if absent or incomplete. Use only content from what was read in Phase 1. Include:
- Site description and primary purpose
- Top 10 most important pages with one-sentence summaries
- Author entity with consistent name matching schema
- Usage guidance for AI systems
- Allow/disallow directives for training vs retrieval use

ROBOTS.TXT
Output exact robots.txt fix for any AI crawler blocks found. Separately allow: GPTBot, ClaudeBot, Claude-SearchBot, PerplexityBot, Bingbot, OAI-SearchBot, Google-Extended. Use the three Cloudflare Content Signals (search, ai-input, ai-train) as comments to declare intent per crawler type where the site owner has a preference.

CANONICAL AND HREFLANG
Output canonical tag fixes for any missing or incorrect canonicals. Output hreflang fixes for any multi-locale issues.

CONSENSUS SIGNAL GAP MAP
Output a table of the top 5 topic areas on the site with the weakest external consensus signal based on DISTRIBUTION_NETWORK assessment from Dimension 3:
| Topic | Current external platforms covering it | Minimum needed | Recommended platforms to target |

Flag any topic where the author entity appears under inconsistent names across external platforms — this breaks AI entity disambiguation and reduces citation probability.

Phase 4 complete marker: output "PHASE 4 COMPLETE — [schema count] schema blocks written, llms.txt [created/updated/existed], [consensus gap count] consensus gaps mapped"

---

PHASE 5 — SESSION SUMMARY

Write to content-optimizer-[YYYY-MM-DD].md in the project root:

Pages audited (count)
Findings by dimension (count per dimension)
Fixes executed — list with one-line result each
Fixes skipped — list with reason and impact score each
Schema blocks written — list by page
llms.txt status
Stat attribution gaps still unresolved — count and list of pages
Cannibalization pairs identified — list with recommendation
Internal link gaps remaining — count
Consensus signal gaps — list by topic with recommended platforms
Entity consistency issues remaining — list

One next action with priority: critical, high, or medium

End with one line: done, blocked, or next — [specific next action].

---

RULES

No suggestion lists. Rewrite it or flag it with exact spec.
No testing recommendations unless two fixes have genuinely different expected outcomes.
No marketing language in any rewritten copy.
No vague quantifiers — every claim gets a number or named example from the actual content.
Stats without attribution: attribute, convert to labeled estimate, or cut. Never left bare.
Schema JSON-LD is output complete and paste-ready. No partial schemas.
CTAs are commands with specific outcomes matching the page's actual offer.
Financial, crypto, and legal content: no absolute outcome claims, no guaranteed yield language, no "you will" framing. Relative and conditional framing only.
All rewrites and additions are derived from the content on disk. Do not introduce outside niche assumptions.
Passage rewrites must pass the standalone test: readable and meaningful without surrounding context.
Entity naming must be consistent across all schema, copy, and external platform references within the same session.
Follow the phase structure. Do not skip phases or merge them.
Output phase completion markers without fail — they are the recovery points.

What This Produces That a Manual Audit Doesn't

A manual content audit tells you what's wrong. This prompt tells you what's wrong, scores it by impact, and rewrites it before the session ends.

The consensus signal gap map from Phase 4 is the deliverable most operators don't have: a table showing exactly which of your topics have no external corroboration, ranked by weakness, with specific platforms recommended for each gap. You can distribute to those platforms immediately using the distribution sequence in the AI search playbook.

The entity injection mechanics — schema structure, sameAs array construction, cross-platform seeding — are covered in detail in Entity Injection: The 6-18 Month Citation Capture Window. The distribution network platforms feeding your consensus signal map are catalogued in the parasite SEO platform map.

This prompt handles the owned site layer. The other two handle distribution and entity infrastructure. Run them in sequence: audit and fix the owned site first, then seed external platforms against the consensus gaps the audit surfaces.