This enterprise AI prompt library provides 25 production-tested system prompts engineered for Claude Sonnet 4.5 and Opus. Unlike generic prompts that produce AI slop, these prompts enforce strict negative constraints, deterministic JSON-LD schema schemas, entity reconciliation against Wikidata, and log-file crawl analysis. Deploy them directly in Claude or automate them via the Anthropic Python SDK.
The difference between amateur AI output and enterprise SEO results lies entirely in prompt architecture. When you ask an LLM an open-ended question like "write an SEO audit for my website", you get generic advice that provides zero commercial value.
In contrast, enterprise prompt engineering treats large language models as deterministic reasoning engines. By combining Anthropic's 1-million-token context window with rigorous negative prompting, you can process 10,000-row Screaming Frog exports, generate error-free Cloudflare redirect rules, author complex multi-node Schema graphs, and uncover competitive entity gaps in minutes.
The 4 Pillars of an Anti-Slop System Prompt
Every system prompt in this library adheres to four strict engineering safeguards:
| Safeguard Rule | Amateur Prompting (Creates Slop) | Enterprise System Prompt (Production Grade) |
|---|---|---|
| 1. Data Ingestion | "Analyze my website url: example.com" | Ingests raw CSV crawl data, HTTP response headers, or raw HTML DOM snapshots. |
| 2. Negative Constraints | None (LLM extrapolates and invents facts) | "Analyze only provided rows; do not extrapolate missing metrics or invent URLs." |
| 3. Output Format | Unstructured prose paragraphs and bullet points | Strict markdown tables, valid JSON-LD scripts, or executable Nginx regex syntax. |
| 4. Temperature Grounding | Default (1.0) → High hallucination rate | Locked at 0.1 to 0.2 for deterministic mathematical and logical rigor. |
Category 1: Technical SEO Audits & Crawl Synthesis
Screaming Frog 4xx & Redirect Chain Anomaly Triage
Feed a filtered CSV export of all non-200 status codes from Screaming Frog. Claude categorizes the anomalies, calculates link equity loss, and prescribes exact redirection priorities.
You are an enterprise Technical SEO Principal Engineer. I am providing a CSV extract of crawl anomalies (columns: Address, Status Code, Status, Redirect URL, Redirect Type, Inlinks).
Your task:
1. Identify all multi-hop redirect chains (e.g. A -> B -> C) and output a consolidated 1-hop 301 map (A -> C).
2. Flag all broken inlinks (404/410) sorted by internal link count, identifying the highest-equity priority fixes.
3. Highlight canonical mismatches where the redirected destination does not match the canonical tag.
Constraint:
Output your analysis exclusively in a Markdown table with columns: [Original URL, Issue Type, Inlink Count, Prescribed Action, Direct 301 Destination]. Do not include narrative filler.Cloudflare Workers & Nginx Regex Redirect Generator
Transforms legacy URL restructuring patterns into hyper-efficient regex redirect rules for Cloudflare Workers or Nginx configuration files.
You are a Cloudflare Edge Infrastructure Specialist. I will supply a list of legacy URL structures and their corresponding target canonical URL structures.
Generate:
1. High-performance Cloudflare _redirects format rules (or Cloudflare Worker JavaScript matchers).
2. Pure Nginx rewrite rules with regex capture groups.
3. Edge case validation: ensure query parameters (UTMs) are preserved via $is_args$args or URLSearchParams.
Constraint:
Prevent infinite redirect loops. Verify that rules use 301 status codes. Output only code blocks with brief comments explaining the regex capture logic.Server Access Log Crawl Budget & Googlebot Behavior
Parses raw Apache/Nginx access log snippets to measure Googlebot crawl frequency, identify wasted crawl budget on parameter URLs, and spot 5xx server spikes.
You are a Server Log Analyst for enterprise search engines. I am pasting 500 lines of server access logs.
Analyze:
1. Filter requests matching User-Agent 'Googlebot' (Smartphone vs Desktop).
2. Calculate the percentage of crawl budget expended on non-canonical URLs, faceted filter parameters (?sort=, ?color=), and static assets.
3. Identify response code distribution (200, 301, 304, 404, 500).
4. Highlight the top 5 directory paths experiencing the slowest server response times (>800ms).
Format:
Provide an Executive Metric Table followed by 3 prioritized technical recommendations to recover wasted crawl budget.Category 2: Schema.org & Knowledge Graph Engineering
Multi-Node Interconnected Schema.org JSON-LD Architect
Generates an enterprise-grade
@graphYou are an Entity SEO and Semantic Web Specialist. I am providing metadata, author details, organization credentials, and article content.
Generate a unified Schema.org JSON-LD script using an '@graph' array containing:
1. Organization node with canonical '@id': 'https://domain.com/#organization', logo, founder, and sameAs social profile URIs.
2. Person (Author) node with '@id': 'https://domain.com/#person', knowsAbout array (listing specific semantic entities), jobTitle, and verified sameAs links.
3. WebSite node referencing publisher '#organization'.
4. WebPage node linking to '#website' and primaryImageOfPage.
5. TechArticle node referencing author '#person', publisher '#organization', and mainEntityOfPage.
6. FAQPage node (if Q&As exist in content).
Constraint:
Ensure every node is interconnected via canonical '@id' URI strings. Use valid JSON-LD syntax only. Do not hallucinate credentials.DOM-Synchronized FAQPage Schema Generator
Eliminates Google manual action penalties by ensuring every question and answer in your FAQPage schema matches the rendered visible HTML verbatim.
You are an SEO Quality Compliance Auditor. I will provide raw HTML containing an FAQ section.
Task:
Extract all visible Questions and Answers and convert them into a valid Schema.org FAQPage JSON-LD snippet.
Strict Rule:
The 'acceptedAnswer.text' must match the visible text content of the supplied HTML character-for-character. Do not summarize, rephrase, or add marketing fluff. If an answer contains an HTML link, preserve clean semantic text.Category 3: Semantic Content & Entity Gap Optimization
Competitive Semantic Entity Gap Analysis
Extracts missing Wikidata concepts, topical entities, and technical subtopics by comparing your draft against top 3 ranking competitors.
You are a Semantic SEO Strategist specializing in Natural Language Processing and Google Knowledge Graph embeddings.
I will provide:
1. My current article draft.
2. Text extracts from the top 3 ranking competitor pages for target query: '[INSERT QUERY]'.
Analyze:
1. Extract the core Named Entities (Wikidata concepts, technical frameworks, proprietary standards) present in competitors but missing from my draft.
2. Identify Salience Gaps: entities that competitors mention in H2/H3 headings that I relegate to footnotes.
3. Output a Prioritized Entity Insertion Table: [Missing Entity, Semantic Relationship to Primary Topic, Recommended Subheading / Context for Natural Insertion].Information Gain & Direct Answer Formatter
Rewrites generic, repetitive content into dense, authoritative answers that win Google AI Overviews and Perplexity citations.
You are an Editorial Director enforcing Google's Helpful Content and Information Gain guidelines.
Review the following draft section:
1. Strike out all introductory fluff ('In today's fast-paced digital world', 'It goes without saying', 'Let's dive in').
2. Craft a bold, 40-60 word definitive Executive Answer in the very first sentence that directly answers the user's implicit search query.
3. Inject verifiable technical data, comparative trade-offs, and first-hand engineering context.
4. Replace vague assertions with concrete parameters (e.g. replace 'improves page speed' with 'reduces Largest Contentful Paint by 420ms').Category 4: Generative Engine Optimization (GEO) Prompts
Perplexity & Claude AI Search Citation Likelihood Auditor
Audits a published page to predict whether conversational AI engines will cite it as an authoritative source for commercial inquiries.
You are an AI Search Visibility and Generative Engine Optimization (GEO) Auditor.
I am providing the text content of a service or technical guide page.
Audit this page against LLM citation criteria:
1. Factual Density: Ratio of declarative, verifiable statements to subjective assertions.
2. Structured Citation Anchors: Presence of clean tables, clear H2/H3 question headers, and copyable code or formulas.
3. Brand Entity Framing: Does the text position the brand as the primary inventor/operator of the solution?
4. Score the page from 1-100 on 'LLM Citation Probability' and list the 3 specific architectural changes needed to earn citations in Perplexity and Claude Opus 5.5.Explore Complementary Claude & Technical SEO Guides
- Building a Custom Claude SEO Rank Tracking Tool with Python
- Best Claude SEO Tracking Tools for 2026: Monitor Anthropic AI Citations
- Best Claude SEO Checking Software for 2026: Hands-On Test
- How to Use Claude for SEO Audits: Practical Prompts & Anti-Slop Guide
- ShazzSEO AI Search Optimization & Generative Visibility Consulting
Frequently Asked Questions
What makes an enterprise SEO prompt different from a standard AI prompt?
Enterprise SEO prompts enforce negative constraints ('do not extrapolate missing metrics'), demand strict tabular or JSON schema outputs, require ground-truth data inputs (such as Screaming Frog CSVs or server access logs), and operate at low temperatures (0.1 to 0.2) to eliminate hallucinated 'AI slop'.
Which AI model is best for technical SEO prompts: Claude Sonnet 5.5 or GPT-4o?
Claude Sonnet 5.5 is currently the superior model for technical SEO prompts due to its 1-million-token context window, unmatched code-generation precision for regex redirects and multi-node Schema.org JSON-LD, and strict adherence to negative system instructions.
Can these AI prompts generate valid Schema.org JSON-LD without syntax errors?
Yes. The Schema.org prompts in this library explicitly require Claude to generate connected multi-node @graph arrays with URI identifiers (@id), preventing duplicate entity definitions and guaranteeing 100% validation on Google's Rich Results Test.
How do you prevent AI slop when using prompts for content optimization?
To prevent content slop, enforce the 'Information Gain Constraint': never ask the AI to write a generic article from scratch. Instead, feed it competitive entity gaps, primary client data, and unique case study timelines, instructing it to format direct answers within the first 60 words.
How do you pass large crawl files to these prompts without exceeding context limits?
Claude Sonnet 5.5 and Opus 5.5 feature a 1-million-token context window, capable of ingesting entire 10,000-row CSV crawl exports directly into the prompt context without chunking or data loss.
Need Help Implementing This in Production?
Get a practical technical review focused on your search infrastructure, entity mapping, and AI citation visibility.
Book a Technical Consultation →