Methodology

One question drives everything here: can search engines find the website, and can AI systems understand, trust, and recommend it? This page explains exactly how we score that, and, just as importantly, what we refuse to claim.

Scoring rules

• Undeterminable criteria are skipped, not zeroed. If a check cannot be established from public evidence (e.g. traceroute from Oman, backend geolocation behind a CDN), its weight is excluded and the section score is normalized over what could actually be measured. A site is never punished for the scanner's blind spots.
• Scores have separate models. The established 10-category, 100-point model remains the overall diagnostic view. AI Readability is now calculated independently from the 15 research criteria below, while Oman Localization Readiness keeps its own /100 methodology.
• Oman locality is binary. Evidence either points to Oman or it does not. A UAE or wider-GCC edge earns no Oman-locality points; regional proximity is reflected only in performance-related checks.
• One report per domain per day (Muscat time). Re-scanning the same domain on the same day serves the stored report (on-screen and PDF). Every scan issues a signed, timestamped PDF, and every criterion score is stored individually.

The overall category model (100 points)

Ten categories, weighted by real-world impact, provide the broad overall diagnostic model. SEO, category-level findings and sub-scores still use this model; the dedicated AI Readability and Oman scores are calculated separately.

Rating bands

• 90–100 Excellent
• 75–89 Good
• 60–74 Average
• 40–59 Weak
• 0–39 Poor

Entity, Structured Data, Trust and Agent-layer scores are their categories rescaled to 100. Rating bands are applied to each reported score, rather than blending AI Readability or Oman readiness into an unsupported all-purpose number.

Dedicated AI Readability (15 criteria, normalized /100)

The research rubric assigns 105 raw weight points across 15 criteria. Each applicable criterion is scored as pass, partial or fail. The displayed score is earned applicable weight ÷ total applicable weight × 100, so 105 is the research denominator before N/A exclusions, not a score above 100. Privacy/residency and linkability/authority are explicitly N/A when the scanner lacks measurable public evidence; their weights are then excluded rather than treated as zero.

Canonical Markdown policy

Search access is not training consent

The audit separates bots used for search or user-requested retrieval (for example OAI-SearchBot, ChatGPT-User, Claude-SearchBot and Claude-User) from bots used for model training (for example GPTBot and ClaudeBot). Blocking a training crawler is a consent choice and does not automatically mean the site is unavailable to AI-assisted search. Recommendations identify the affected purpose instead of telling every site to allow every AI bot.

Oman Localization Readiness (separate /100)

Checks whether a website is locally relevant, fast for Oman users, and whether there is evidence of Oman-hosted infrastructure. Locality checks (IP, ASN, DNS, API host) are binary: Oman or not Oman. A .om / .com.om domain is also credited: registration requires an Omani commercial presence, making it registry-verified local proof.

Latency interpretation bands (for a real Oman probe): 30–80ms strong local/GCC signal · 100–180ms possible UAE/GCC · 200–350ms likely farther region · 400ms+ poor regional localization. Latency is always evidence, not proof. The MVP measures TTFB from the scanner’s serverless region and labels it approximate; the architecture accepts future probes in Oman, UAE, Saudi Arabia, India, Europe and the USA.

Confidence labels

Every finding carries one of six labels. They are not decoration. They are the product.

What this scanner will never claim

Scan mechanics

• Fetches the homepage, robots.txt and sitemaps, then crawls up to ~10 prioritized same-domain pages (about, contact, services, pricing, case studies, blog, privacy, terms…).
• Parses static HTML only, with no JavaScript execution. This mirrors how most AI crawlers see the site, and low static-HTML content is itself reported as a finding.
• Requests representative canonical pages as both HTML and Markdown, verifies Content-Type and Vary: Accept, and compares the negotiated bodies.
• Checks optional /llms.txt and /llms-full.txt files, then probes legacy .md patterns only to detect active twins and recommend their 301/308 migration to canonical negotiated URLs.
• Extracts JSON-LD (plus microdata types), asset hosts, API hosts from HTML, form actions and up to 5 same-site JS bundles.
• Resolves DNS, follows CNAME chains, fingerprints CDNs from headers, and geolocates the visible IP via a pluggable provider (free-tier accuracy; treated as evidence, not fact).
• Reports are stored in Supabase Database and private Storage so results are shareable by permalink.