AI SEARCH VISIBILITY
中 Log in / Sign up

SCORING METHODOLOGY · HOW WE SCORE

Every check is benchmarked to public standards — and then finer still.

12 dimensions, each benchmarked to public standards. The surface-level signals can be partly checked with public tools like Google, W3C and SSL Labs — but the judgment of whether AI actually cites you comes from GeoWeb's own weighting model.

Taiwan's only authoritative and transparent GEO analysis platform. This scoring model has been hand-tuned over years — more than 6,000 rounds of tracking algorithm shifts, emerging standards and weighting updates — with the greatest share of that effort going into misdiagnosis: hundreds of individually logged false-negative and false-positive corrections. Not something other GEO analysis tools can match.

THE PUBLIC BASIS

W3C · IETF · WHATWG · Google · Microsoft · schema.org · OWASP · SSL Labs…

From recognized standards bodies & vendors

The surface-level signals in every dimension are benchmarked against these bodies' public specs and tools.

→

HOW GEOWEB INTEGRATES

1

actionable report

One pass over all 12 dimensions, distilled into a single GEO score and a fix-priority order — plus the AI-citation assessment public tools don't cover.

12 DIMENSIONS × PUBLIC STANDARDS

Every item, laid open

Items marked "Partly verifiable" — public tools check the surface signals; GeoWeb's score layers weighting and deeper judgment on top. Items marked "GeoWeb proprietary" have no single public-tool equivalent.

DimensionWhat we checkBenchmarked public tool / standardType
SESEO technicalTitle, meta, canonical, OG, viewport, lang, image alt, link structureW3C HTML Checker · Rich Results Test · PageSpeed InsightsPartly verifiable
SDStructured dataJSON-LD / Microdata / RDFa completeness (Organization, FAQ, Product, Article…)Rich Results Test · Schema Markup ValidatorPartly verifiable
CTContent citability40–60-word answer blocks, definition sentences, structured lists/tables, statistic & source citation patterns— No single public toolGeoWeb proprietary
EAE-E-A-T authorityAuthor info, publish/update dates, trust pages, social proof, credential signalsTrustpilotPartly verifiable
SMSemantic structureH1 uniqueness, H1–H6 hierarchy, semantic HTML5 elements, content/HTML ratioW3C Validator · axe DevToolsPartly verifiable
FQFAQ / Q&A readinessFAQPage schema, visible Q&A content, answer quality (20–100 words)Rich Results TestPartly verifiable
SZPerformance & crawl budgetHTML < 2MB, load performance, resource load priorityPageSpeed InsightsPartly verifiable
ACAI crawler accessibilityrobots.txt openness to GPTBot / ClaudeBot / PerplexityBot etc., sitemap, IndexNowGoogle Search Console · Bing Webmaster ToolsPartly verifiable
AEAEO readinessAnswer-first paragraph structure, featured-snippet eligibility, step content, comparison tables— No single public toolGeoWeb proprietary
SCSnippet controlmeta robots, nosnippet, max-snippet, data-nosnippetGoogle Search ConsolePartly verifiable
TSTransport securityHTTPS, HSTS, CSP, X-Frame-Options, X-Content-Type-OptionsSSL Labs · SecurityHeaders.comPartly verifiable
LNLanguage naturalnessBoilerplate density, syntactic variety, specificity, domain-term ratio, human vs LLM patterns— Proprietary judgment modelGeoWeb proprietary

AUXILIARY DIMENSION · NOT IN THE SCORE

Agent friendliness

This measures how open your site is to AI agents (not search crawlers). It does not affect the 12-dimension GEO score above, but it reflects how discoverable, readable and connectable you are to autonomous agents on the agentic web.

What we check

18 checks (5 advisory, unscored): llms.txt, llms-full.txt, AI-bot rules and Content-Signal in robots.txt, .well-known/* (MCP server-card, agent-skills, api-catalog, OAuth discovery, security.txt, Web Bot Auth, A2A agent-card), auth.md, HTTP Link headers, Markdown content negotiation, JSON-LD Action, DNS-AID, whether non-existent paths return a real 404, and trust-anchor pages (about / contact / privacy / terms)

Benchmarked tool / standard

isitagentready.com (Cloudflare, protocol layer)
is-agentic.com (Vercel / Ora, behaviour layer)

Type

Partly verifiable

Not just a score — we tell you what to fix first

Benchmarking public standards is the baseline; distilling 12 dimensions into one priority-ordered fix path, then adding the proprietary AI-citation layer — that's what GeoWeb does.

Run a free full audit →

ON SCORING

GeoWeb benchmarks the public standards and tools above, but applies its own dimension-weighting model and continuously references the latest technical specs and documentation from each engine. So when a single item doesn't match a given third-party tool exactly, that reflects a difference in scoring perspective and weighting — we judge from the angle of "will AI actually cite you", rather than a mechanical match against any one metric.

FAQ

They monitor — how often your brand gets mentioned in AI answers. GeoWeb's backtesting platform does that too: it puts real questions to mainstream AI engines and records, prompt by prompt, whether you were named and whether your content was cited. The prompts are built around your market and your customers and grounded in questions people actually search for. The difference is what comes next — the 12-dimension audit tells you why you weren't cited and which fix comes first, and both the monitoring and the remediation are included in the managed GEO service, run by us.
Because technical SEO is only 12% of the total. The other 88% sits in structured data, content citability, E-E-A-T, AI crawler reachability, AEO readiness, snippet control and language naturalness — the things ranking tools don't audit, but which decide whether an engine can safely quote you. Ranking well is not the same as being easy to cite.
On-page readiness is the core of it. Before an engine can cite you it has to reach you, extract you, and trust you — all three gates are on-page, and a low score means you never reach the shortlist. How much room you get after that also depends on brand authority and competition, which move slowly. So we don't ask you to trust a single number: backtesting cross-references your score movements against actual mention rates per engine, so you can see which fix moved which engine.
Because they describe a state, not a quality. AI build-fingerprint detection and Agent readiness are reported separately and excluded from the total: the first judges how code was built, which says nothing about content quality; the second is a business decision — whether to expose your services to agents — and skipping it shouldn't cost you points. Three checks inside Agent readiness are advisory for the same reason.
Weights reflect how much each dimension affects whether you can be cited. They sum to 100%, ranging from 12% down to 2%, and they change as engine behaviour changes — most recently we added language naturalness, funded by taking 2% from each of three other dimensions. The full weighting and per-dimension rubric ship with your report.
Yes, and usually for good reasons — you changed the page, you added markup, the score moves. One caveat for site-wide analysis: if this run sampled a different set of pages than the last one, don't compare the totals directly; look at the per-page movement instead.