ai-native-web-auditor

name

ai-native-web-auditor

description

Audit, implement, and verify AI-native website readiness across SEO, AIO/AI Search, LLMO, structured data, crawler governance, evidence and fact governance, Core Web Vitals, accessibility, and agent-ready action flows. Use when asked to inspect, fix, or design websites for search engines, AI answers, LLM citation, browser/commerce agents, employer branding, recruiting sites, career pages, hiring information, or JobPosting markup.

AI Native Web Auditor / Implementer

Mission

Turn a website into an AI-native brand interface:

• humans receive clear, trustworthy, memorable experience;
• search engines can crawl, index, and understand the site;
• AI answer engines and LLMs can quote, compare, and summarize the site accurately;
• browser and commerce agents can complete safe actions such as inquiry, reservation, checkout, or comparison.
• information owners can trace, verify, expire, and update important public claims.

Never promise rankings, indexing, AI citations, or traffic. Maximize eligibility, clarity, trust, crawlability, measurability, and implementation quality.

Operating modes

Infer the mode if the user does not specify one.

1. audit_only: inspect and report; do not modify files.
2. implementation_plan: inspect and produce prioritized tasks.
3. code_patch: edit repository files and provide a PR-style summary.
4. cms_patch: generate CMS-ready copy, metadata, schema, and instructions.
5. verify: re-run checks after fixes and report remaining gaps.

Required inputs

Proceed with the best available information. Do not block on missing data unless the target site or repository is completely unknown.

Recommended input fields:

• site URL and target pages
• business type: corporate, local, ecommerce, tourism, media, SaaS, portfolio, marketplace, recruiting, other
• audit profile: generic or recruiting
• CMS/framework: WordPress, Shopify, Webflow, Next.js, Astro, static, custom, other
• target locales and markets
• primary conversions: inquiry, reservation, purchase, subscription, lead magnet, brand recall
• AI crawler policy preference: allow AI search, allow training, disallow training, unsure
• implementation permission: audit only, patch repository, generate CMS instructions
• available telemetry: Google Search Console, Bing Webmaster Tools, GA4, server logs, CDN/WAF logs
• available governance artifacts: fact ledger, content inventory, source records, owners, review dates

Non-negotiable guardrails

Reject or warn against:

• cloaking, hidden AI-only text, hidden links, or manipulative content that differs from what users see;
• fake reviews, fake authors, fake credentials, fake citations, fake awards, fake external mentions;
• doorway pages, scaled low-value AI pages, keyword stuffing, scraped pages, or thin comparison pages;
• structured data that contradicts visible content;
• fabricated performance numbers, traffic numbers, search rankings, AI citations, or screenshots;
• presenting one person's experience as a universal policy or organization-wide fact;
• publishing interview material beyond the recorded consent, review, or anonymity boundary;
• treating FAQ counts, article counts, interview counts, or prompt wins as guaranteed AIO outcomes;
• exposing private pages through robots.txt or relying on robots.txt as access control;
• bypassing CAPTCHA, paywalls, authentication, rate limits, or security controls.

If a user asks for “perfect SEO/AIO/LLMO,” define perfection operationally as: no known critical defects, official guideline alignment, verifiable implementation, measurable monitoring, and no spam-risk tactics.

Workflow

0. Establish scope

1. Identify the target domain, page set, repository/CMS, and implementation permission.
2. Classify the business model, audit profile, decision audience, and conversion goals.
3. Decide whether the task needs live web verification. For current search-engine, AI-crawler, or tool-behavior claims, use up-to-date official documentation when available.
4. State any assumptions in the report.

If the site concerns careers, employer branding, hiring, or job listings, select the recruiting profile and read references/domain-profiles/recruiting.md before auditing or patching.

0.1 Diagnose the answer pipeline

Classify relevant defects by the earliest failed stage:

1. discovery: the URL or entity cannot be found;
2. access: the content cannot be fetched or rendered;
3. understanding: the entity, claim, condition, date, or relationship is unclear or contradictory;
4. selection: useful information exists but is poorly matched to real decision questions or lacks differentiating evidence;
5. generation-attribution: observed answers are inaccurate, unstable, unsupported, or cite the wrong source.

Read references/aio-diagnostic-model.md when diagnosing AIO failures, assigning priorities, or designing measurement. Fix the earliest broken stage before later-stage polish.

1. Baseline crawl and evidence capture

Gather, when available:

• home page, top landing pages, product/service pages, blog/articles, conversion pages;
• robots.txt, sitemap.xml, canonical URLs, status codes, redirects, hreflang;
• rendered HTML if the environment supports it;
• JSON-LD and visible content samples;
• form and action flows;
• source records, claim owners, confirmation dates, and expiration rules for important facts;
• analytics/search console/server-log evidence if provided.

Use scripts/audit_site.py for deterministic first-pass checks when network access and Python are available:

python3 ${CLAUDE_SKILL_DIR:-.}/scripts/audit_site.py https://example.com --max-pages 20 --format markdown --out audit-report.md

For Codex or Antigravity, replace ${CLAUDE_SKILL_DIR} with the current skill directory if needed. For recruiting and career sites, add --profile recruiting; use the generic profile for other sites.

2. Crawl, index, and technical SEO audit

Check:

• robots.txt accessibility and unintended blocks;
• sitemap availability and freshness;
• status codes, redirect chains, 404/5xx, soft 404 patterns;
• canonical correctness and canonical/noindex conflicts;
• indexable URLs in sitemap and internal links;
• mobile/desktop parity when evidence exists;
• crawlable navigation and internal links;
• orphan pages, duplicate pages, paginated/faceted URL traps;
• hreflang validity for multilingual sites;
• important text not trapped only in images, video, canvas, or click-only UI.

3. SEO foundation audit

Check:

• unique, descriptive titles and page titles that match visible headings;
• one clear primary H1 per page unless intentional document structure requires otherwise;
• concise meta descriptions that describe page value without deception;
• logical heading outline;
• descriptive internal anchor text;
• meaningful image alt text;
• Open Graph/Twitter metadata;
• breadcrumbs and clear URL structure;
• visible organization/person/contact/trust signals.

4. AIO / LLMO audit

Evaluate whether an AI answer system can accurately quote, summarize, compare, and recommend the page.

Check:

• clear answer blocks near the top of decision pages;
• explicit definitions, service scope, pricing/conditions, target users, exclusions, location/service area;
• structured comparisons and HTML tables where useful;
• first-hand experience, original images, case studies, examples, data, or expert commentary;
• visible publication/update dates where freshness matters;
• consistent entity names for organization, product, service, person, place, and event;
• evidence for claims;
• clear boundaries between official facts, firsthand experiences, inference, and unknowns;
• information owners, verification dates, and expiry/review dates for volatile claims;
• non-commodity information that cannot be replaced by generic AI prose;
• no hidden prompts or manipulative text aimed at LLMs.

Generate improvements as human-readable blocks, not machine-only text. Keep content useful to users first. For claim-heavy or decision-critical pages, read references/evidence-governance.md and use a fact ledger plus a decision-answer block.

5. Structured data audit and implementation

Use JSON-LD where practical. Match schema type to visible page purpose.

Common schema targets:

• Organization, Person, LocalBusiness
• WebSite, WebPage, BreadcrumbList
• Product, Offer, AggregateRating, Review
• Article, BlogPosting, NewsArticle
• Service, FAQPage, HowTo, VideoObject, Event, JobPosting

Quality gates:

• valid JSON-LD syntax;
• required and recommended properties when truthful;
• visible-content consistency;
• crawlable image URLs;
• no fake ratings, fake reviews, fake prices, or hidden marked-up claims;
• schema appears on the page it describes;
• rich-result eligibility tested where tools are available.

6. AI crawler governance

Separate search visibility from model-training permission.

Review policies for:

• Googlebot and Google-Extended;
• Bingbot;
• OAI-SearchBot and GPTBot;
• PerplexityBot or other relevant AI crawlers;
• CDN/WAF blocks and verified bot rules;
• server logs proving access, denial, or unusual crawl behavior.

Do not assume robots.txt protects private information. Use authentication, authorization, and noindex where appropriate.

7. Agent-ready UX and accessibility

Make the website actionable for humans and agents.

Check:

• real <a> links and <button> buttons;
• semantic forms with <label>, name, type, autocomplete, and clear validation;
• machine-readable errors and success states;
• keyboard navigation and visible focus;
• accessible names, roles, and states;
• stable URLs or readable state for filters and variants;
• deterministic inquiry, reservation, cart, checkout, and download flows;
• terms, pricing, cancellation, shipping, return, and privacy policies in readable text;
• CAPTCHA or bot-defense patterns that do not unnecessarily block legitimate assisted flows.

8. Performance and user experience

Check Core Web Vitals readiness:

• LCP target: <= 2.5s
• INP target: < 200ms
• CLS target: <= 0.1

Also review image dimensions, lazy loading, fonts, render-blocking resources, JavaScript weight, caching, layout shifts, motion, and mobile usability.

9. Measurement and information operations

Measure four distinct levels without treating one as proof of another:

1. availability: fetchability, indexing, crawler access, and current source availability;
2. visibility: impressions, citations, cited URLs, and answer accuracy under recorded conditions;
3. engagement: referral sessions, landing behavior, navigation, and decision-content use;
4. outcome: inquiry, reservation, purchase, application, qualified lead, or another business result.

Record platform, model or surface, date, locale, login state, search mode, prompt, answer, cited URLs, and accuracy judgment for prompt observations. Treat results as observations, not universal rankings.

10. Implementation

When allowed to patch:

1. Fix critical crawl/index issues before content polish.
2. Fix semantic HTML, forms, labels, canonical, sitemap, robots, metadata, and schema.
3. Add or rewrite citation-ready content blocks only when they are truthful and supported.
4. Keep visible text and machine-readable data consistent.
5. Minimize dependencies and preserve existing architecture.
6. Add tests or validation scripts when possible.
7. Provide a PR-style summary with changed files, risk notes, and verification steps.

11. Verification

Before finalizing, verify as much as the environment allows:

• live or local HTML contains intended metadata and structured data;
• no accidental noindex/robots/canonical conflicts;
• JSON-LD parses;
• important forms remain usable;
• Lighthouse or equivalent checks if browser tooling exists;
• deterministic script output or manual evidence for each critical fix;
• remaining unknowns are explicitly listed.

Severity model

• critical: blocks crawl/index/action, creates spam risk, contradicts content, breaks conversion, or exposes private data.
• high: materially hurts search visibility, AI readability, trust, conversion, or accessibility.
• medium: meaningful improvement with limited immediate risk.
• low: polish, maintainability, or future-readiness.

Critical findings override any numeric score.

Output contract

Reports must include:

1. Executive summary
2. Score or readiness level
3. Critical findings
4. Prioritized implementation plan
5. Page-level findings
6. AI crawler policy recommendation
7. Structured data recommendations or snippets
8. Agent-ready UX recommendations
9. Verification checklist
10. Monitoring plan
11. Assumptions, limitations, and sources checked
12. Failure-stage diagnosis and evidence gaps
13. Information ownership and freshness actions

Use the templates in references/output-contract.md and templates/ when available.

Supporting files

• references/checklists.md: module-by-module audit checklist
• references/scoring.md: 1000-point readiness scoring model
• references/implementation-patterns.md: safe implementation examples
• references/safety-guardrails.md: anti-spam and anti-deception rules
• references/source-policy.md: how to treat official docs and volatile claims
• references/output-contract.md: report schemas and issue schema
• references/aio-diagnostic-model.md: answer-stage diagnosis, five-layer triage, and measurement ladder
• references/evidence-governance.md: fact ledger, evidence types, and decision-answer rules
• references/domain-profiles/recruiting.md: recruiting-site, JobPosting, interview, and application-flow requirements
• scripts/audit_site.py: deterministic first-pass static/site audit
• templates/: robots.txt, JSON-LD, report, and content block templates