开悟Kaiwu
  1. Home

Method and scorecard

Kaiwu scores can be recomputed. Each part is made of checks, and each check is a pass rate with a stated denominator. Below are the formula, the scorecard and the rules we follow when producing a report.

Scorecard

The checks behind each part:

SEO

100 pts
Technical 20

Search crawlers not blocked; robots and sitemap reachable; correct canonical, html lang and hreflang; http, www and bare domain redirect to the https host in one hop

Page basics 15

Key-page titles contain the main local keyword and fit in 600 px; description present; one H1; image alt text; absolute og:image URL

Content 15

Body language matches the locale folder; pages exist for comparison and review searches; superlatives carry a source

Keyword coverage 15

Tracked queries reach Google page one; brand and model queries reach page one; rankings land on the right locale; key-page opening paragraph contains the main keyword

Internal links 10

Content pages linked from the site, not only the sitemap; internal PageRank of product pages; at least two internal inlinks per page

External links 15

Authority score and referring domains against the median of competitors collected the same way

Speed 10

Field data: mobile LCP and INP, desktop CLS; HTML compression

GEO

100 pts
AI crawler access 15

Major AI crawlers allowed; body text present in the raw HTML without scripts

Citability 20

Description present; consistent body language; sourced superlatives; pages that answer review and comparison questions

Citation rate 20

Share of Google AI Overviews with a source list that cite your site

Mention rate 10

Share of non-branded questions where the AI answer names your brand

Recommendation rate 15

Visibility across major AI platforms per market, or mention rate from repeated live tests

Source coverage 10

Share of all sources cited in AI answers that mention your brand

Answer accuracy 10

Share of brand answers with no factual errors, checked against your published specifications

Combined = SEO × 0.6 + GEO × 0.4 A≥ 85B70–84C55–69D40–54E< 40

Formula

  • Part score = weight × the average pass rate of the checks in that part.
  • Total = sum of tested part scores ÷ tested weight × 100, rounded to an integer.
  • Combined = SEO × 0.6 + GEO × 0.4.
  • Every pass rate is "items that meet the condition ÷ items that should", and the report states the denominator, such as total pages, pages in a language folder or number of queries.

When data is missing

  • A check without data is marked untested and is not estimated. If fewer than two checks in a part, or fewer than half of them, have data, the whole part is untested and its weight is removed from the total.
  • Checks that do not apply are excluded and noted, for example hreflang on a single-language site.
  • This avoids misleading scores. Without field speed data, the speed part would be left with only "HTML is compressed", and scoring it as usual would produce a meaningless full mark.

Scoring rules

  • Each problem loses points in one part only.
  • Description, body language and sourced claims affect both search engines and AI, so they count once in SEO and once in GEO, and the report says so.
  • The GEO total is broken out. AI crawler access is usually full marks and can hide real AI performance, so the report also gives the combined rate for citation, mention, recommendation, source coverage and accuracy.
  • Competitors are scored only when we collected the same data for them. The backlink part uses the median of competitors collected on the same platform, date and basis.

How numbers and conclusions are checked

  • Every number in a report is generated from the raw data by code, and each figure comes from a single source wherever it appears.
  • Conclusions read from tables, such as "ranks first in Germany and Italy", are checked against the data during generation, and generation stops on a mismatch.
  • Each report goes through two independent reviews before delivery: one for language, one for logic and data, each reading the full text.
  • Inferences are stated with qualifiers and tested at re-measurement, never presented as fact.

Re-measurement

  • Re-measurement uses the same questions, the same checks and the same formula.
  • Projected scores in a plan change only the checks the plan covers. All other checks keep their current values.
  • Diagnostic reports carry no targets and no projected scores.

Start with an audit of your own site

Send us your domain and target markets. A person reviews the report before it reaches you, and every number in it states its source and method.

Get a free audit →