Free scan 註冊
Menu

評分準則

最後更新：2026年9月11日

絕無黑箱。您評分中的每一分，都來自我們在此公開的檢查項目及其權重。您、您的開發人員或其他代理公司，都可以覆核我們的計算。若某項檢查有誤，請告訴我們，我們會為所有人修正。

兩項評分均源自這套評分準則：21個模組、共104項檢查，按類別加權。本頁由評分程式所讀取的同一批檔案生成，因此不可能與掃描器的實際運作出現偏差。

評分如何計算？

掃描共執行104項檢查，分為21個模組。每項檢查的結果為通過、部分通過或未通過，並附1至5級的嚴重程度。通過計1，部分通過計0.5，未通過計0。

  • 類別得分： category_value = 100 * sum(status_factor * severity) / sum(severity), over scored findings only.
  • 每項評分： score = sum(category_value * weight) / sum(active weight).

有兩種結果不計入計算，既不扣分，也不加分：

needs api
an API we did not connect — never counts as a pass and never penalises; report the count
not applicable
the check does not apply to this site

Active only when their signals are detected. An inactive category leaves the denominator entirely and the remaining weights re-normalise, so a blog is never penalised for having no Product schema.

Search SEO：類別及權重

How well the page is set up to rank in classic search engines. 權重只在適用的類別之間加總；若某個條件類別不適用，權重會重新按比例分配。

類別權重模組深度適用情況
Indexability & Crawl22M1, M2, M3完整一律適用
Core Web Vitals / Performance16M15部分一律適用
On-Page & Meta12M7完整一律適用
Structured Data12M5部分一律適用
Rendering8M4部分一律適用
Internal Linking & Semantics8M10部分一律適用
E-E-A-T7M16部分一律適用
Images / Media5M9部分一律適用
Sitemaps & Discovery5M17完整一律適用
Freshness3M13部分一律適用
Social Cards2M8部分一律適用
E-commerce15M18尚未建立僅在偵測到時適用
Local10M19尚未建立僅在偵測到時適用
International8M20完整僅在偵測到時適用

AI Visibility：類別及權重

How extractable and citable the page is for AI answer engines (GEO/AEO). 權重只在適用的類別之間加總；若某個條件類別不適用，權重會重新按比例分配。

類別權重模組深度適用情況
Answer Extractability20M11部分一律適用
Structured Data16M5部分一律適用
Fact Density / Original Data14M12部分一律適用
AI Crawler Access12M14部分一律適用
Rendering (non-JS)10M4部分一律適用
Entity / Knowledge-Graph10M6部分一律適用
E-E-A-T / Authority9M16部分一律適用
Freshness6M13部分一律適用
Images / Multimodal3M9部分一律適用
llms.txt0M21完整只報告，不計分

為甚麼兩項評分永不合併？

兩項評分永不合併為一個數字。 Two scores are published side by side and never averaged into one. A site can rank well and be uncitable by AI, or the reverse; collapsing that hides the finding we are paid to surface.

甚麼情況下評分不予公布？

每項評分都會列明，掃描實際量度到的評分準則權重佔多少。若低於該權重的70%，我們仍會計算該數字，但不會公布，並會在報告中列出未能評估的類別。只根據一小部分評分準則得出的評分，看似精確，實際描述的卻是一套與本頁不同的評分準則。

甚麼情況會為評分設上限？

嚴重程度5的未通過項目，代表網站完全無法被索引或無法連接；只要問題存在，報告中其他項目都無法產生效果。任何一項評分中出現此類未通過項目，該評分的上限即為40分（F級）。此上限只會拉低評分，絕不會提高評分。

M14: AI1 is this axis's severity-5 trigger. It fires only on the site-wide condition — every retrieval crawler refused, or a robots.txt that does not respond — which is the same evidence T13 reads for the search axis, assessed for its own consequence rather than borrowed from T13's.

字母等級代表甚麼？

公布的評分按其所屬區間的下限，對應一個字母等級，但有一項調整：未解決的檢查結果達8項或以上，等級下調一級；未解決的檢查結果達12項或以上，等級下調2級。評分本身不變。

等級最低分數含義
A90Your site is easy for Google and AI assistants to read; what remains is refinement.
B80Your site is easy to read, with a few gaps worth closing before your competitors do.
C70The basics work, but gaps are holding you back; a few well-chosen fixes move the score the most here.
D60The basics work only in part; several problems stop Google or AI assistants reading key pages.
F0Google and AI assistants struggle to read your site; answers about your line of business come from other businesses' pages.

各嚴重程度代表甚麼？

嚴重程度含義
5The site cannot be indexed or reached at all. Nothing else on the report matters while it holds, and it caps the axis at 40.
4A primary signal is broken or absent. Directly costs indexing, ranking or citability.
3A real defect with measurable impact, but the page still works without it.
2Hygiene. Worth fixing, cheap to fix, rarely decisive on its own.
1Minor. Reported for completeness; a site can be excellent without it.

各模組不涵蓋甚麼？

在21個模組中，有14個尚未達到評分準則所述的完整深度。每個模組未有檢查的範圍，均列於此處及每份採用該模組的報告中，而不是留待讀者從數字自行推斷。

模組名稱深度未涵蓋範圍
M4Rendering / JS dependency部分Raw HTML only. We see what a non-JS crawler receives, which is the question that matters for AI citability, but with no rendered-DOM diff we cannot separate 'missing for everyone' from 'missing only for AI', nor classify SSR/SSG/ISR.
M5Structured data部分Type presence, markup/body parity (S6, 2026-09-13) and, on the offer page, Offer presence and Offer parity (OF2, OF3, 2026-10-02). No validation of the required properties per schema.org type, so an Organization block with no name still passes.
M6Entity / knowledge graph部分The site's own declaration only: whether a primary entity exists, what identity links it publishes, and whether it names itself consistently. Nothing is reconciled against Wikidata or a company register, so we can see that a site claims an identity, never that the claim resolves to the right one. No NAP consistency across third-party directories.
M8Social cards部分Open Graph only. No twitter:card family.
M9Images / media部分Alt coverage only. No alt quality, no modern-format check, no VideoObject.
M10Internal linking & semantics部分Orphan detection only. No anchor-text grading, pillar/cluster mapping or click depth.
M11Answer extractability部分Structure only: question headings and the length of what answers them, extractable blocks, segmentation, and whether a section opens on an unresolved reference. Whether an answer is correct, complete or better than a competitor's is not judged, and self-containment is tested on the opening sentence rather than the whole passage. Since 2026-10-02 (OF1) the offer page is read for where its prices sit; whether the offer itself is clear is not judged.
M12Fact density / original data部分Counts figures, outbound sources and promotional claims. It cannot tell original data from restated data — a proprietary survey and a quoted industry statistic look identical to a counter — so the original-data half of this category is named, not measured. The claim list is English-only, which under-reads a Chinese-language page. PP7 (thin content) is reported under this heading as context but is not scored: word count is not fact density.
M13Freshness部分Presence of a visible timestamp, sitemap lastmod, and a claimed freshness date judged against the crawl date (O10, 2026-09-13). No staleness judgement against topic volatility: a rates page a year old is stale, a evergreen guide a year old is not, and telling them apart needs a per-vertical model we do not yet have. A visible updated date is compared with dateModified (O11, 2026-10-02, reported, not scored); whether a date is older than the content's last change is not observable from outside and is not claimed.
M14AI crawler access部分robots.txt policy only — what the site declares, not what its edge actually serves. A WAF or bot-management rule can refuse a crawler that robots.txt admits, and that is invisible from a crawl; confirming it needs a request sent under each crawler's user-agent, or Bing Webmaster's AI performance report. The user-agent lists are verified first-party in docs/ai-visibility-research-2026-09-09.md.
M15Core Web Vitals / performance部分Crawl-only runs measure proxies: TTFB, compression, caching, image weight, script loading. Real field data needs --pagespeed, which raises this module to full (PS1-PS5).
M16E-E-A-T / authority部分Presence of the trust signals only: a byline or author node on content pages, an about and a contact page, a published privacy policy, a contact route in the markup. We can see that a page names an author; we cannot see that the author is qualified, and no credential, affiliation or citation of one is verified.
M18E-commerce尚未建立No Product/Offer checks. The category stays inactive and re-normalises out, which is correct for every seed site and for most HK/SG SME prospects.
M19Local尚未建立No LocalBusiness, NAP or opening-hours check. This one matters commercially — most HK/SG SME prospects are local businesses — and it re-normalises out today rather than scoring them on it.

會執行哪些檢查？

全部104項計分檢查，並列出每項所屬的模組、計入哪項評分，以及其嚴重程度。

編號模組計入評分嚴重程度檢查項目
T1M3Search SEO4HTTPS redirect
T2M2Search SEO4non-canonical host folds to canonical
T3M2Search SEO3alt host redirects, not merely canonical
T4M2Search SEO3/index.html redirects to /
T5M17Search SEO3sitemap.xml reachable
T6M17Search SEO2sitemap URL count
T7M17Search SEO2sitemap lastmod
T8M1Search SEO3robots.txt present and references a sitemap
T10M3Search SEO2custom 404
T11M2Search SEO4homepage canonical
T12M7Search SEO3meta viewport
T13M1Search SEO5robots.txt crawl access
C1M2Search SEO4homepage canonical exact
C2M2Search SEO3/index.html 301s
C3M2Search SEO3alt-host canonical when it serves 200
C4M2Search SEO2no internal index.html links
C5M2Search SEO2no links to the non-canonical host
S1M5兩者4JSON-LD on homepage
S2M5兩者3homepage schema types
S3M5兩者3JSON-LD on article
S4M5兩者2article schema type
S5M2Search SEO5no page is noindex
S6M5兩者4markup/body parity
O1M7Search SEO4title tag
O2M7Search SEO2meta description
O3M7Search SEO3single H1
O4M7Search SEO2html lang
O5M8Search SEO2Open Graph tags
O6M8Search SEO2OG image reachable
O7M13兩者2visible timestamp
O8M7Search SEO2article title lengths
O9M9兩者3image alt coverage
O10M13兩者3visible freshness
P2M15Search SEO2LCP fetchpriority
P3M15Search SEO2LCP preload
P4M15Search SEO3image dimensions (CLS)
P5M15Search SEO2script defer/async
P6M15Search SEO3large images
I1M17Search SEO1sitemap URL list
I2M3Search SEO4sitemap URLs return 200
I3M10Search SEO3orphan detection
X1M7Search SEO3duplicate titles
X2M7Search SEO2duplicate meta descriptions
X3M2Search SEO4self-referential canonical
X4M3Search SEO4mixed content
HD1M3Search SEO2HSTS header
HD2M3Search SEO1X-Content-Type-Options
HD3M3Search SEO1frame protection
HD4M3Search SEO1Referrer-Policy
HD5M15Search SEO3compression
HD6M15Search SEO2Cache-Control
HD7M15Search SEO3TTFB
HD8M3Search SEO3redirect chain
HL1M20Search SEO3hreflang tags
HL2M20Search SEO3hreflang values valid
HL3M20Search SEO2self-referencing hreflang
HL4M20Search SEO2hreflang x-default
HL5M20Search SEO3hreflang reciprocity
HL6M20Search SEO3hreflang targets 200
L1M21AI Visibility2llms.txt present
L2M21AI Visibility1llms.txt starts with H1
L3M21AI Visibility2llms.txt links resolve
L4M21AI Visibility1llms.txt declared in head
R1M4兩者4content without JS
R2M4兩者3headings without JS
R3M4兩者4links without JS
R4M4兩者4client-side app shell
PS1M15Search SEO4mobile performance
PS2M15Search SEO4LCP (mobile)
PS3M15Search SEO3CLS (mobile)
PS4M15Search SEO3TBT (mobile)
PS5M15Search SEO2desktop performance
G4M2Search SEO5homepage indexed
B2M3Search SEO3Bing page alerts
B3M3Search SEO2Bing crawl stats
AI1M14AI Visibility5AI answer engines can reach the site
AI2M14AI Visibility4citation crawlers individually allowed
AI3M14AI Visibility3user-triggered fetchers allowed
AB1M11AI Visibility4question headings answered directly
AB2M11AI Visibility3extractable structures present
AB3M11AI Visibility3content is segmented into passages
AB4M11AI Visibility2sections open self-contained
FD1M12AI Visibility4statistic density
FD2M12AI Visibility3outbound citations to third-party sources
FD3M12AI Visibility2claims supported by figures
EN1M6AI Visibility4primary entity declared
EN2M6AI Visibility4identity links (sameAs)
EN3M6AI Visibility2the entity is named consistently
EE1M16兩者3content carries an author
EE2M16兩者3the business is identifiable
EE3M16兩者2published policies
EE4M16兩者3a real contact route
PP1M7Search SEO3inner-page H1
PP2M7Search SEO3inner-page titles
PP3M7Search SEO2inner-page meta
PP4M2Search SEO4inner-page canonical
PP5M5兩者3inner-page schema
PP6M9兩者3inner-page image alt
PP8M2Search SEO4noindex pages
PP9M2Search SEO3duplicate content
OF1M11AI Visibility3prices in an extractable structure
OF2M5AI Visibility3offer markup present
OF3M5AI Visibility3offer markup matches the page
L5M21AI Visibility2llms.txt agrees with the pages

哪些檢查只報告、不計分？

另有21項檢查列於報告附錄，不計任何分數。 Reported in the appendix, contributing zero points. Two standing reasons: the check encodes a convention of ours rather than a property of the client's site, or it returns data rather than a verdict. A check may fault a site for being inconsistent with its own declared intent, never for disagreeing with ours. A third, temporary reason: a new check is reported for one cycle of client data before it scores (O11, since 2 October 2026).

編號不計分原因
T9ads.txt — only meaningful if the site sells ad inventory. Absence is not a defect.
P1GA4 tag — analytics choice, not an SEO property. Plenty of good sites use something else or nothing.
A1AdSense snippet — monetisation, not visibility.
A2AdSense pub-ID match — monetisation, not visibility.
I4cross-site partner links — our own inter-property linking convention, driven by partnerDomains config. Meaningless for a client.
IN1IndexNow key file — our deployment practice, and only observable when we hold the filesystem.
H1.htaccess present — needs filesystem access we do not have on a prospect, and duplicates the redirect and security-header tests, which we can see from outside.
H2.htaccess HTTPS rule — same; the outcome is already measured by the HTTPS-redirect test.
H3.htaccess www rule — same; the outcome is already measured by the www-redirect test.
H4.htaccess index.html rule — same; the outcome is already measured by the index.html redirect test.
H5.htaccess security headers — same; the outcome is already measured by the HSTS-header test.
G1GSC property access — a diagnostic of our own connection, not of the site.
G2GSC search performance — data for the narrative, not a verdict.
G3GSC top queries — data for the narrative, not a verdict.
B1Bing access — a diagnostic of our own connection.
B4Bing performance — data, not a verdict.
B5Bing top queries — data, not a verdict.
B6Bing top pages — data, not a verdict.
AI4training-crawler policy — which model-training crawlers robots.txt refuses. Refusing them is a legitimate business choice that costs no citations, per docs/ai-visibility-research-2026-09-09.md, so this returns data for the report to state rather than a verdict to score.
PP7thin content — reported under M12 as context. Word count is not fact density; scoring it there would overstate what we measure.
O11Visible updated date against dateModified. Reported, not scored, for its first cycle, and not for either reason above: the check is new and scores once a month of client data shows how often a correct site trips it.

評分準則以甚麼為依據？

以搜尋引擎及標準機構公開發布的指引為依據，而非對排名因素的臆測。評分準則版本0.3.0，檢查對應表版本0.2.0。