How the outside-in audit works

Scores and coverage

Pass earns 1, partial earns 0.5 and fail earns 0. Confidence does not change health points. Area score = weighted earned points / observed raw check weight × 100.

Observed applicable check weight / all applicable check weight; unavailable stays in the denominator; not applicable is removed.

Each check’s effective coverage weight is its area weight × raw check weight / 12. Removing a not-applicable check never redistributes its weight within the area.

An area needs at least 50% coverage to show a grade; otherwise it says “Not visible”. An area whose checks are all not applicable says “Not applicable”. Overall requires at least 60% weighted coverage, 4 eligible areas and resolved identity. Otherwise you see “Limited snapshot” with observations and no overall number.

Overall is the area-weighted mean of eligible scores, using the original area weights. Coverage and the measured-area count accompany it. Grades use unrounded scores; displaying a rounded score never upgrades the grade. Grades have no plus or minus.

• A: 90 to 100
• B: 75 to below 90
• C: 60 to below 75
• D: 45 to below 60
• F: 0 to below 45

Failed means no scoped useful observation or usable resolved snapshot. Limited means useful observations with insufficient overall eligibility. Useful-partial means overall eligibility with applicable gaps. Complete requires every applicable promised probe to finish with usable evidence. A limited run never counts as complete.

The check catalog

These 48 check definitions come from the executable scoring catalog. Each entry shows its raw weight, confidence ceiling and whether it supplies health points or context. Applicability follows the confirmed business model and evidence.

Choosing priorities

Priority RS = (2 × impact + 6 − effort + confidence points) × scope multiplier. Impact and effort are bounded judgments, each from 1 to 5, not measured revenue effects.

Confidence: high 2, medium 1, low 0. Scope: page ×1, site ×1.2, location ×1.5, brand ×2. Scope never expands beyond supported evidence. RS bands are A ≥24, B ≥16, C ≥10 and D below 10.

Sort by RS, then higher impact, lower effort, higher confidence and check ID. Consolidate a common root cause once. Show up to three fixes, at most one per area, and up to two supported strengths from distinct areas. Never pad a sparse report. Low-confidence AI samples cannot supply verified strengths.

Sources and dates

• Own public-page measurements
• Google PSI/CrUX
• DataForSEO
• Serper
• OpenRouter

Each displayed observation carries its source, instrument, scope and collection date or period. A page review date is not a new measurement date.

Only permitted public evidence enters scores, maps and reports. Restricted sources and their derivatives are excluded; missing data never becomes an invented zero.

A provider estimate remains modeled. Search ETV estimates clicks from tracked keywords; paid traffic cost estimates value from a provider model, not actual spend, invoices, acquisition cost or revenue.

The source list identifies instruments used when available, not a claim that each provider returned data for your business. Source dates on results describe the actual observation; the review date above describes this methodology.

What the evidence can establish

Public-business marketing surfaces only: no private account access, no form submissions, no mystery calls, no patient narratives. We remove incidental sensitive data from evidence before analysis or public output.

Static HTML heuristic: an action link, form field, tag or widget can be detected in fetched source. This does not establish rendered placement, usability, booking availability or staff response.

A detected tag is not proof of event delivery, attribution or correct account configuration. A tag-manager-only page makes downstream checks unavailable (container). Known server-side or consent-delayed measurement remains unavailable where the outside view cannot establish it.

Only a complete covered page sample with no known measurement markers permits the low-confidence finding “we could not detect” analytics. A missing marker behind a container cannot establish broken tracking.

Speed reports label lab tests and field data separately, including page versus origin scope and device/period. Missing CrUX coverage is unavailable. API blocks, timeouts, quota and incomplete datasets are stated separately.

Instant neutral discovery answers are illustrative samples. Show mentions and citations as numerator/denominator per engine with prompt version/date; missing answers are not misses. Branded comparison and reputation panels do not inflate discovery rates. Sample misses alone cannot produce a failing AI check or negative reputation claim.

AI access grades the versioned search/answer crawlers for tested instruments. Policies for training crawlers are observed only, without penalty. Robots group precedence and allow rules matter.

Paid presence and paid traffic cost are context, not health points. With no target ads observed, execution checks are not applicable: “no paid search observed”. Brand defence is graded only when a completed brand query shows competitor advertising pressure; execution is graded only from observed ads.

A matched map needs the target plus at least three relevant competitors with both axes from the same instrument, period and geography. Missing-axis rows are excluded and explained; a comparison table survives when no map is supported.

Map 1 reputation is (v / (v + m)) × R + (m / (v + m)) × C: v is review count, R is rating, m is set median review count with minimum 1, and C is set mean rating. At zero reviews it equals C, flagged “no reviews found”; an unmatched listing is excluded. Review count is the bubble size. AI discovery mentions stay in the comparison table.

Map 2 compares modeled organic ETV with DataForSEO paid traffic cost. Within-set rank percentiles or winsorized scaling describe sampled relative visibility; tied values share ranks, constant values show no spread. A chart does not establish market dominance.

The audit is automated by default. Reports with unresolved quality or sensitive-data concerns can be held for review. No universal human review, delivery time, legal compliance, clinical quality or business outcome is promised.

Deep scope is shown before commitment in a versioned manifest: up to 5 competitors, up to 300 pages, and a 24-prompt panel (12 discovery, 6 comparison, 6 reputation) across 2 engines and 2 repeats. Availability and funded scope determine which optional work is included; final reports identify completed, unavailable, omitted-for-budget and held-for-review stages.

Deep screenshots come from DataForSEO and support a separate experience rubric; screenshots do not test keyboard behavior or completed transactions and do not silently alter scoring/v1. Email and phone are required for the free full report; automated calls and texts are optional. Phone lookup does not prove subscriber assignment.

Reuse and retention

Compatible scope and scoring/pipeline versions can reuse an instant result for 24 hours. A corrected business, location or scope requires a new run. Reuse preserves the original observation dates.

• Raw evidence: 90 days.
• Snapshots and reports: 18 months.
• Leads: 24 months from last activity.
• Consent records: 5 years.

A lead deletion removes personal contact and delivery data and walkthrough requests. It revokes access to that person’s hosted reports, PDFs and evidence routes. We retain only minimum consent proof as a salted hash with its timestamp and text version, plus a hashed suppression record so we do not contact that person again. Public business evidence follows its normal retention windows: raw vendor evidence for 90 days and snapshots and reports for 18 months; a lead deletion does not erase that business evidence. The audit covers public business information and removes incidental sensitive data; it provides no legal or clinical assurance.