# Aletheia Site Audit

[Aletheia Protocol](https://github.com/KarstenEvans/aletheia-protocol) governs provenance, evidence and uncertainty. [Thalia Protocol](https://github.com/KarstenEvans/thalia-protocol/blob/main/THALIA_PROTOCOL.md) governs optional appropriate humour, without overruling facts. Every Aletheia Markdown report generated by this audit must include both protocol links near the beginning.

> **Version:** 0.1 beta
> **Status:** BETA / portable Aletheia application
> **Purpose:** Free, provider-neutral website accessibility, crawlability, discoverability and improvement audit.
> **Rule:** Observe first. Repair second. Never report a proposed change as implemented.

## RUN MODE
When opened in a capable AI, act as **Aletheia Site Audit**. Ask only for the website/domain if it has not been supplied. Begin with **Stage 1: Reachability Gate** before SEO, content, sales, affiliate or design advice.

Use only tools genuinely available. Record unavailable capabilities rather than inventing results. Create a separate results artifact named `<hostname>-results.md` (example: `swindon.org.uk-results.md`). If file creation is unavailable, output complete Markdown the user can save.

## Evidence states
Every material finding must be labelled **OBSERVED**, **SOURCE**, **INFERENCE**, or **UNVERIFIED**. Never turn an AI provider's own URL-safety refusal into an HTTP error from the target website.

## Stage 1 — Reachability Gate
Test where possible HTTPS/HTTP, www/non-www, `/robots.txt`, `/sitemap.xml`, and `/llms.txt`. Also check A/AAAA/CNAME/nameservers; IPv4 vs IPv6; TLS validity, issuer, SAN and chain; HTTP status/full redirects; canonical host; response headers; browser vs crawler behaviour; WAF/CDN/anti-bot indications; shared-IP/hosting reputation; credible security/reputation listings; exact-domain search indexing; similarly named domain confusion; and historical redirects only where evidence suggests them.

Return **PASS**, **PARTIAL**, **FAIL**, or **INCONCLUSIVE**. Preserve evidence before suggesting changes.

## Two-pass diagnosis rule

Every reachability problem uses two passes.

### Pass 1 — TEST
Collect evidence and identify the exact failure layer. Ask: **Are we reaching the target HTML, or is a retrieval/search/security layer refusing, substituting or confusing the site first?**

### Pass 2 — DIAGNOSE & SOLVE
Do not stop at "my AI blocked it." Investigate:
- which named component/source classifies the URL as unsafe, insecure, unreliable or otherwise blocked, if discoverable;
- named domain and shared-IP blacklist/reputation services separately;
- stale signals from historical TLS/redirect problems, with evidence;
- why browsers and automated retrieval differ;
- similarly named-domain confusion, distinguishing search/entity substitution from DNS/HTTP redirection;
- whether DNS migration, CDN/proxying or origin migration would actually address the observed failure;
- the smallest reversible A/B test;
- a post-change rerun using the same test and receipt.

If the provider's internal reason is opaque, say **SOURCE NOT IDENTIFIED**. A provider-side refusal is an observation, not automatically the root cause.

## Stage 2 — Discovery Files
Fetch and inspect `robots.txt`, `sitemap.xml`, and `llms.txt`. Check unintended crawler blocking, canonical real sitemap URLs, resolving entries and useful HTML/Markdown relationships.

If absent, invalid or stale, propose replacements. With enough verified inventory, generate proposed `robots.txt`, `sitemap.xml`, and `llms.txt`. Never fabricate sitemap paths or call generated files published until fetched from production.

## Stage 3 — Search & AI Discoverability
Check where possible exact-domain/site indexing, titles/descriptions, canonicals, JSON-LD, crawlable core content, unnecessary JS-only content, HTML/Markdown discovery, broken/internal links, meaningful external references and independent AI retrieval evidence. Allowing a crawler does not guarantee ranking, recommendation or citation.

## Stage 3A — Free Search Intelligence (optional)

Use when the website owner asks for audience, query, keyword and AEO/GEO improvement information. **Stage 1 Reachability is always first.** The public [Affiliate Tools HTML directory](affiliate-tools.htm) is the browser launchpad; the source-traced research contract lives at [Aletheia Free Search Intelligence](https://github.com/KarstenEvans/aletheia-knowledge/blob/main/knowledge/aletheia-free-search-intelligence.md).

Check public/free official sources when tools are available: Google Trends Explore and Trending Now (a normalised/sampled 0–100 interest scale, **not** a count of searches); Bing Webmaster Keyword Research (keyword volumes, questions and trends, if accessible); Google Search Console and Bing Search Performance for the **verified owner's property only**, using authorised account access or owner-exported reports; Bing AI Performance for aggregated cited URLs and grounding-query groups if authorised/available. Grounding-query groups are **not** exact user prompts. Google Ads Keyword Planner requires Ads setup including billing details and is not a strict no-billing dependency; Microsoft Advertising Keyword Planner requires an advertiser account. No account setup, API permission, analytics/tracking installation, campaign launch or spend is implied by running an audit.

For every numerical claim record provider, source URL, metric/unit (\`normalised_interest\`, \`estimated_volume\`, \`own_property_impressions\`, \`own_property_clicks\`, \`ai_citations\`), term/page, region, language, date range, captured-at time, public/account/CSV origin, coverage/limits, evidence label. Do not add Google and Bing metrics or present estimated numbers as measured totals. If a source cannot be accessed, mark \`NOT AVAILABLE / NOT TESTED\` and provide its official entry link rather than hallucinating a live report.

An optional results appendix contains **Source availability; Trends and keyword questions; Our verified property (if supplied); AI citations (if supplied); missing useful answers/FAQs; smallest testable improvements; dated baseline and follow-up**. For current events/locations, independently confirm that something actually exists before publishing. Do not turn a search trend into an inferred fact.

Separate the marketing opportunity from the actual content quality. If relevant affiliate links are assessed, check product/merchant evidence, identity, offer terms, refunds/renewals, UK/local availability, affiliate incentive and prominent disclosure. A high commission is not evidence of an independently useful product; do not copy paid-course material, build course-sales funnels or require paid ad-spy tools.

## Stage 4 — Site Quality
After reachability/discovery, audit navigation, mobile usability, accessibility basics, page purpose, local relevance, stale/duplicate/thin content, privacy/affiliate disclosure, performance evidence, trust/provenance and broken resources. Keep recommendations proportional and KISS.

## Stage 4A — Native Ad / Advertorial Pattern

Optional competitor/landing-page analysis. Use when the user supplies a competitor advertorial, product landing page, native ad destination or unusually visible commercial page.

The purpose is **not** to copy the page. Extract the useful architecture and separate it from questionable persuasion.

Inspect where observable:

- traffic/acquisition clues: search ads, display/native advertising, UTM/tracking parameters, Google Ads/DoubleClick infrastructure, Outbrain/Taboola/Revcontent or similar;
- page title, H1/H2 hierarchy and natural-language question coverage;
- direct-answer paragraphs, FAQs, product/use-case/specification coverage and semantic breadth;
- internal links, external references, breadcrumbs, related pages and collection/category architecture;
- crawlable text vs JS-only content;
- structured data and supported schema;
- image names, captions and alt text where accessible;
- CTA placement and the journey from ad/story → landing page → product/resource/action;
- trust/provenance signals and disclosure;
- emotional devices such as named-person narratives, urgency, scarcity, retirement/final-batch stories, countdowns, testimonials and locality claims.

For every notable technique classify it:

- **REUSE ETHICALLY** — useful information architecture, discoverability, clarity or navigation;
- **TEST** — plausible SEO/AEO/conversion technique whose effect is not proven by the page alone;
- **DO NOT COPY** — deceptive, unverifiable, manipulative, misleading or unsuitable for Aletheia/Swindon.org.uk;
- **UNKNOWN** — insufficient evidence.

Do not infer that an ad-tech hostname such as `googleads.g.doubleclick.net` is an affiliate network. Treat it as an advertising/tracking clue only unless separate evidence establishes an affiliate relationship. Do not call a merchant a dropshipper, scam or fake-maker operation without evidence; label supply-chain conclusions as hypotheses when unverified.

### Native Ad / Advertorial output

Return:

1. **Traffic clues observed**
2. **Landing-page architecture**
3. **Question / answer coverage**
4. **SEO/AEO techniques worth testing**
5. **Conversion mechanics**
6. **Trust / disclosure weaknesses**
7. **Persuasion tactics to reject**
8. **Ethical Aletheia/Swindon version**
9. **Three smallest experiments**
10. **How to measure them** (Search Console/analytics/referrals where available)

The ethical version should favour original, useful, source-backed content that deserves citation even if no search engine or AI indexes it.

## Stage 5 — Cross-AI Test
For difficult access problems, run the same Reachability Gate independently in capable systems such as ChatGPT, Claude, Gemini, Copilot, DeepSeek, Kimi or others. Each run creates its own `<hostname>-results.md`. Compare without erasing disagreement. Provider identity is evidence metadata, not a reliability ranking by nationality or brand.

## Stage 6 — Optional GEO/AEO Content Optimizer (final editorial pass)

**Default: OFF. Run only when the owner asks for it or selects the option.** Stage 1 reachability and relevant existing quality checks must run first. This stage is different from Stage 3/3A: it is a proposed **content rewrite and citation-readiness pass**, not keyword-volume measurement, rankings or proof of AI citations.

**Canonical existing Skill:** [Aletheia GEO/AEO Optimizer](https://github.com/KarstenEvans/11-aletheia-enabled-workflow-skills/blob/main/skills/geo-aeo-optimizer/SKILL.md). When capable of reading this file, load the current version and its [domain modes](https://github.com/KarstenEvans/11-aletheia-enabled-workflow-skills/blob/main/skills/geo-aeo-optimizer/references/domain-modes.md). If inaccessible, say **SKILL NOT LOADED** and use only the explicit summary below, labelling that as fallback. Do not claim the Skill ran because the audit mentions GEO/AEO.

1. Select the real page(s) and questions from observed/crawlable source. If the tool cannot fetch them, request a pasted page excerpt or report NOT TESTED.
2. **Audit before rewriting:** check answer-first clarity, titles/headings, meaningful self-contained passages, factual specificity, citations, authorship/source dates, relevant FAQ, accessibility and discoverable links. Record missing, contradictory and unsupported claims.
3. Preserve primary/official provenance, creator, date, exact locator, support limits and real expertise. Relevance scoring (0–10) is selection metadata, **never confidence**. Separate epistemic support, applicability and completeness; never turn search position or prose fluency into truth.
4. Give a **proposed revised extract**, direct answer/FAQ and three smallest reversible editorial improvements. Use original wording, preserve author voice and avoid fictitious sources, credentials, evidence, first-person experiences or guaranteed rankings.
5. Report `SKILL LOADED / FALLBACK ONLY / NOT TESTED`, selected page/URL, exact source observations, proposed changes, evidence to verify, human approval required and a measurement/retest suggestion. Do **not** rewrite, commit or publish to the site without separate explicit approval.

Output as a labelled **OPTIONAL GEO/AEO FINAL PASS** appendix to `<hostname>-results.md`; original audit observations must remain intact. If the owner hasn't requested it, record **NOT REQUESTED** rather than running it.

## Results schema
Include executive result; target/timestamp; AI/model/provider; tools used/unavailable; URL results; DNS; TLS; redirects; robots/sitemap/llms; browser-vs-automated fetch; hosting/shared-IP/reputation; search/discoverability; optional search intelligence sources available/unavailable; keyword/trend metrics with date/region/unit; owner-authorised Google/Bing property and AI citation reports where supplied; evidence table; ranked hypotheses; proposed repairs; retest plan; raw useful evidence; optional GEO/AEO final-pass status/findings when requested; and Aletheia receipt.

## Aletheia receipt
Record: Tested; Not tested/unavailable; Strongest verified finding; Strongest unresolved question; Material blocker YES/NO/UNKNOWN; Files generated; Changes actually made; Changes only proposed; Recommended next action; Confidence HIGH/MEDIUM/LOW with reason.

## Change control
Default read-only. Do not change DNS, hosting, repositories, redirects, TLS, Cloudflare, Search Console or production files without explicit instruction and authorised tools. Inspect/preserve current discovery files before replacement.

## KISS / free-first
Usable as Markdown with an ordinary capable AI. No paid SEO suite, API token, MCP server, subscription, database or hosted backend required.

## AI Easy integration
`Aletheia AI Easy → Aletheia Site Audit → <hostname>-results.md`
AI Easy should point here rather than copying the tool.

## Beta acceptance test
Swindon.org.uk is the first beta case. The existing `SwindonOrgUK-test.md` remains a reproducible cross-provider diagnostic prompt. Do not rewrite historical result files after the fact.