· ~7 min
How the score works: four axes, confidence, and the P0 cap
Base score 100 from our knowledge-base weights; discovery files and injection only adjust ±5.
GEO Cool does not use a third-party 176-check list. Pass, fail, unknown, and weighting follow our own acceptance rules. Category names on the report match common single-page auditors for scanning; the numbers are ours.
Four base axes, 100 total
| Axis | Max | What we look at |
|---|---|---|
| Citation reach | 35 | Citation bots explicitly allowed, live 200 probes, sitemap present and declared in robots |
| Structured data | 30 | Valid JSON-LD, key types, FAQ, Organization / sameAs / @id |
| Citable content | 20 | Answer under the title, data table or authority links, time and price dates |
| Technical health | 15 | No nested documents, lang + canonical + social tags, compression + HTTP/2 |
Unknown is not zero
Timeouts and connection failures are unknown. That item gets neither full nor zero points in the numerator. The total normalizes over judged items only. More unknowns mean lower confidence, shown at the top of the report. On flaky networks, retest before treating a low-confidence score as a redesign verdict.
Micro-adjust at most ±5
llms.txt, ai.txt, or /ai/*.json at the root can add up to 5. They do not help classic Google ranking, but they help agents discover the site. High-signal injection phrases, data-ai- attributes, or zero-width characters subtract 5. Combined, the band stays −5 to +5.
P0 cap
P0 cuts citation outright—citation bots blocked with Disallow: /, missing lang, missing or invalid JSON-LD, nested documents. Any unfixed P0 caps the final score at 74. Fix the blockers and the cap lifts.
Bands
- 90–100 excellent
- 75–89 good
- 50–74 pass
- 0–49 needs work
“Good” starts at 75 so public auditor bands roughly align—one page should not be “good” in one tool and “pass” in another. We align band edges, not their checklists.